Spaces:
Running
Running
Sync from felladrin/MiniSearch@79128cb
Browse files- .env.example +4 -1
- .github/CODE_OF_CONDUCT.md +3 -0
- .github/CONTRIBUTING.md +1 -0
- .github/SECURITY.md +8 -0
- .github/SUPPORT.md +1 -0
- README.md +1 -1
- agents.md +10 -1
- changelog.md +10 -6
- client/components/App/App.test.tsx +1 -0
- client/components/App/App.tsx +4 -1
- client/modules/config.test.ts +1 -0
- client/modules/config.ts +2 -0
- client/modules/pubSub.ts +2 -1
- client/modules/settings.test.ts +47 -3
- client/modules/settings.ts +34 -2
- client/modules/textGenerationWithWllama.test.ts +35 -0
- client/modules/textGenerationWithWllama.ts +34 -4
- client/modules/wllama.test.ts +54 -1
- client/modules/wllama.ts +55 -40
- docs/ai-integration.md +30 -2
- docs/api.md +1 -0
- docs/coding-conventions.md +1 -0
- docs/configuration.md +25 -11
- docs/core-technologies.md +1 -0
- docs/development-commands.md +36 -0
- docs/glossary.md +7 -0
- docs/overview.md +10 -0
- docs/project-structure.md +1 -0
- docs/pull-requests.md +6 -0
- docs/quick-start.md +6 -0
- docs/reranking.md +3 -0
- docs/search-history.md +5 -0
- docs/security.md +2 -0
- docs/ui-components.md +25 -0
- docs/wllama-model-evaluation.md +129 -0
- e2e/main-thread-budget.spec.ts +1 -0
- eval/README.md +2 -0
- knip.json +2 -1
- package-lock.json +184 -136
- package.json +2 -1
.env.example
CHANGED
|
@@ -4,9 +4,12 @@ ACCESS_KEYS=""
|
|
| 4 |
# The timeout in hours for access key validation. Set to 0 to require validation on every page load.
|
| 5 |
ACCESS_KEY_TIMEOUT_HOURS="24"
|
| 6 |
|
| 7 |
-
# The default model ID for Wllama.
|
| 8 |
WLLAMA_DEFAULT_MODEL_ID="littlelamb-290m"
|
| 9 |
|
|
|
|
|
|
|
|
|
|
| 10 |
# The base URL for the internal OpenAI compatible API. Example: `INTERNAL_OPENAI_COMPATIBLE_API_BASE_URL="https://api.openai.com/v1"`. Leave blank to disable internal OpenAI compatible API.
|
| 11 |
INTERNAL_OPENAI_COMPATIBLE_API_BASE_URL=""
|
| 12 |
|
|
|
|
| 4 |
# The timeout in hours for access key validation. Set to 0 to require validation on every page load.
|
| 5 |
ACCESS_KEY_TIMEOUT_HOURS="24"
|
| 6 |
|
| 7 |
+
# The default model ID for Wllama in browsers without WebGPU, where the model runs on the CPU.
|
| 8 |
WLLAMA_DEFAULT_MODEL_ID="littlelamb-290m"
|
| 9 |
|
| 10 |
+
# The default model ID for Wllama in browsers with WebGPU.
|
| 11 |
+
WLLAMA_DEFAULT_GPU_MODEL_ID="gemma-3-4b"
|
| 12 |
+
|
| 13 |
# The base URL for the internal OpenAI compatible API. Example: `INTERNAL_OPENAI_COMPATIBLE_API_BASE_URL="https://api.openai.com/v1"`. Leave blank to disable internal OpenAI compatible API.
|
| 14 |
INTERNAL_OPENAI_COMPATIBLE_API_BASE_URL=""
|
| 15 |
|
.github/CODE_OF_CONDUCT.md
CHANGED
|
@@ -9,6 +9,7 @@ We pledge to act and interact in ways that contribute to an open, welcoming, div
|
|
| 9 |
## Our Standards
|
| 10 |
|
| 11 |
Examples of behavior that contributes to a positive environment for our community include:
|
|
|
|
| 12 |
- Demonstrating empathy and kindness toward other people
|
| 13 |
- Being respectful of differing opinions, viewpoints, and experiences
|
| 14 |
- Giving and gracefully accepting constructive feedback
|
|
@@ -16,6 +17,7 @@ Examples of behavior that contributes to a positive environment for our communit
|
|
| 16 |
- Focusing on what is best not just for us as individuals, but for the overall community
|
| 17 |
|
| 18 |
Examples of unacceptable behavior include:
|
|
|
|
| 19 |
- The use of sexualized language or imagery, and sexual attention or advances of any kind
|
| 20 |
- Trolling, insulting or derogatory comments, and personal or political attacks
|
| 21 |
- Public or private harassment
|
|
@@ -35,6 +37,7 @@ This Code of Conduct applies within all community spaces, and also applies when
|
|
| 35 |
## Enforcement
|
| 36 |
|
| 37 |
Instances of abusive, harassing, or otherwise unacceptable behavior may be reported to the community leaders responsible for enforcement at:
|
|
|
|
| 38 |
- GitHub: [@felladrin](https://github.com/felladrin)
|
| 39 |
- Email: (private contact can be provided upon request)
|
| 40 |
|
|
|
|
| 9 |
## Our Standards
|
| 10 |
|
| 11 |
Examples of behavior that contributes to a positive environment for our community include:
|
| 12 |
+
|
| 13 |
- Demonstrating empathy and kindness toward other people
|
| 14 |
- Being respectful of differing opinions, viewpoints, and experiences
|
| 15 |
- Giving and gracefully accepting constructive feedback
|
|
|
|
| 17 |
- Focusing on what is best not just for us as individuals, but for the overall community
|
| 18 |
|
| 19 |
Examples of unacceptable behavior include:
|
| 20 |
+
|
| 21 |
- The use of sexualized language or imagery, and sexual attention or advances of any kind
|
| 22 |
- Trolling, insulting or derogatory comments, and personal or political attacks
|
| 23 |
- Public or private harassment
|
|
|
|
| 37 |
## Enforcement
|
| 38 |
|
| 39 |
Instances of abusive, harassing, or otherwise unacceptable behavior may be reported to the community leaders responsible for enforcement at:
|
| 40 |
+
|
| 41 |
- GitHub: [@felladrin](https://github.com/felladrin)
|
| 42 |
- Email: (private contact can be provided upon request)
|
| 43 |
|
.github/CONTRIBUTING.md
CHANGED
|
@@ -22,6 +22,7 @@ docker compose exec development-server npm run lint
|
|
| 22 |
```
|
| 23 |
|
| 24 |
This runs:
|
|
|
|
| 25 |
- Biome (formatting/linting)
|
| 26 |
- TypeScript (type checking)
|
| 27 |
- knip (dead code detection)
|
|
|
|
| 22 |
```
|
| 23 |
|
| 24 |
This runs:
|
| 25 |
+
|
| 26 |
- Biome (formatting/linting)
|
| 27 |
- TypeScript (type checking)
|
| 28 |
- knip (dead code detection)
|
.github/SECURITY.md
CHANGED
|
@@ -20,6 +20,7 @@ Report security vulnerabilities through GitHub's [Private Vulnerability Reportin
|
|
| 20 |
### What to Include
|
| 21 |
|
| 22 |
When reporting a vulnerability, please include:
|
|
|
|
| 23 |
- A clear description of the vulnerability
|
| 24 |
- Steps to reproduce the issue
|
| 25 |
- Potential impact of the vulnerability
|
|
@@ -60,23 +61,27 @@ When reporting a vulnerability, please include:
|
|
| 60 |
MiniSearch is designed as a privacy-focused search application with the following security assumptions:
|
| 61 |
|
| 62 |
**Trust Boundaries:**
|
|
|
|
| 63 |
- **Browser Environment**: The application runs entirely in the user's browser
|
| 64 |
- **Server Component**: Optional backend for search and AI processing
|
| 65 |
- **AI Models**: Local or remote AI processing with configurable endpoints
|
| 66 |
|
| 67 |
**Data Flow:**
|
|
|
|
| 68 |
- User queries are sent to SearXNG instances (configurable)
|
| 69 |
- AI processing can be local (Wllama) or remote (API endpoints)
|
| 70 |
- Search history is stored locally in the browser
|
| 71 |
- No tracking or analytics by default
|
| 72 |
|
| 73 |
**Security Controls:**
|
|
|
|
| 74 |
- Optional access key protection for deployment
|
| 75 |
- Configurable AI endpoints for privacy
|
| 76 |
- Local-first data storage
|
| 77 |
- No third-party tracking or analytics
|
| 78 |
|
| 79 |
**Potential Risks:**
|
|
|
|
| 80 |
- Malicious SearXNG instances could log queries
|
| 81 |
- Remote AI endpoints could access user queries
|
| 82 |
- Browser extensions could interfere with the application
|
|
@@ -118,6 +123,7 @@ A few of MiniSearch's dependencies ship native code or WebAssembly rather than a
|
|
| 118 |
### Approving a new install script
|
| 119 |
|
| 120 |
Under `strict-allow-scripts`, any package that ships an install script without a recorded decision fails the install. To unblock it, a maintainer must:
|
|
|
|
| 121 |
1. Read the script first (`npm view <pkg> scripts`, or the unpacked tarball) and decide whether it is acceptable.
|
| 122 |
2. Record the decision: `npm approve-scripts <pkg>` writes a version-pinned `"<name>@<version>": true` entry to `package.json#allowScripts` (`allow-scripts-pin` is on); `npm deny-scripts <pkg>` writes `"<name>": false`, denying every version of the package.
|
| 123 |
3. Update the decision list above in the same PR, with the reasoning.
|
|
@@ -156,6 +162,7 @@ Because native/WASM deps are exact pins, a Renovate bump arrives as a one-line v
|
|
| 156 |
## Security Updates
|
| 157 |
|
| 158 |
Security updates will be:
|
|
|
|
| 159 |
- Released as new versions
|
| 160 |
- Announced in release notes
|
| 161 |
- Coordinated with dependency updates when applicable
|
|
@@ -163,6 +170,7 @@ Security updates will be:
|
|
| 163 |
## Security Team
|
| 164 |
|
| 165 |
The MiniSearch security team is currently the project maintainer:
|
|
|
|
| 166 |
- [@felladrin](https://github.com/felladrin) - Project Maintainer
|
| 167 |
|
| 168 |
## Acknowledgments
|
|
|
|
| 20 |
### What to Include
|
| 21 |
|
| 22 |
When reporting a vulnerability, please include:
|
| 23 |
+
|
| 24 |
- A clear description of the vulnerability
|
| 25 |
- Steps to reproduce the issue
|
| 26 |
- Potential impact of the vulnerability
|
|
|
|
| 61 |
MiniSearch is designed as a privacy-focused search application with the following security assumptions:
|
| 62 |
|
| 63 |
**Trust Boundaries:**
|
| 64 |
+
|
| 65 |
- **Browser Environment**: The application runs entirely in the user's browser
|
| 66 |
- **Server Component**: Optional backend for search and AI processing
|
| 67 |
- **AI Models**: Local or remote AI processing with configurable endpoints
|
| 68 |
|
| 69 |
**Data Flow:**
|
| 70 |
+
|
| 71 |
- User queries are sent to SearXNG instances (configurable)
|
| 72 |
- AI processing can be local (Wllama) or remote (API endpoints)
|
| 73 |
- Search history is stored locally in the browser
|
| 74 |
- No tracking or analytics by default
|
| 75 |
|
| 76 |
**Security Controls:**
|
| 77 |
+
|
| 78 |
- Optional access key protection for deployment
|
| 79 |
- Configurable AI endpoints for privacy
|
| 80 |
- Local-first data storage
|
| 81 |
- No third-party tracking or analytics
|
| 82 |
|
| 83 |
**Potential Risks:**
|
| 84 |
+
|
| 85 |
- Malicious SearXNG instances could log queries
|
| 86 |
- Remote AI endpoints could access user queries
|
| 87 |
- Browser extensions could interfere with the application
|
|
|
|
| 123 |
### Approving a new install script
|
| 124 |
|
| 125 |
Under `strict-allow-scripts`, any package that ships an install script without a recorded decision fails the install. To unblock it, a maintainer must:
|
| 126 |
+
|
| 127 |
1. Read the script first (`npm view <pkg> scripts`, or the unpacked tarball) and decide whether it is acceptable.
|
| 128 |
2. Record the decision: `npm approve-scripts <pkg>` writes a version-pinned `"<name>@<version>": true` entry to `package.json#allowScripts` (`allow-scripts-pin` is on); `npm deny-scripts <pkg>` writes `"<name>": false`, denying every version of the package.
|
| 129 |
3. Update the decision list above in the same PR, with the reasoning.
|
|
|
|
| 162 |
## Security Updates
|
| 163 |
|
| 164 |
Security updates will be:
|
| 165 |
+
|
| 166 |
- Released as new versions
|
| 167 |
- Announced in release notes
|
| 168 |
- Coordinated with dependency updates when applicable
|
|
|
|
| 170 |
## Security Team
|
| 171 |
|
| 172 |
The MiniSearch security team is currently the project maintainer:
|
| 173 |
+
|
| 174 |
- [@felladrin](https://github.com/felladrin) - Project Maintainer
|
| 175 |
|
| 176 |
## Acknowledgments
|
.github/SUPPORT.md
CHANGED
|
@@ -7,6 +7,7 @@ If you need help with MiniSearch, these are the best support channels.
|
|
| 7 |
### GitHub Discussions
|
| 8 |
|
| 9 |
For questions, feature requests, and general discussion, use [GitHub Discussions](https://github.com/felladrin/MiniSearch/discussions). This is the best place to:
|
|
|
|
| 10 |
- Ask how to configure MiniSearch for your use case
|
| 11 |
- Request new features
|
| 12 |
- Share ideas and feedback
|
|
|
|
| 7 |
### GitHub Discussions
|
| 8 |
|
| 9 |
For questions, feature requests, and general discussion, use [GitHub Discussions](https://github.com/felladrin/MiniSearch/discussions). This is the best place to:
|
| 10 |
+
|
| 11 |
- Ask how to configure MiniSearch for your use case
|
| 12 |
- Request new features
|
| 13 |
- Share ideas and feedback
|
README.md
CHANGED
|
@@ -37,7 +37,7 @@ The AI can run entirely inside your browser tab, on GPU or CPU, so a working set
|
|
| 37 |
## Features
|
| 38 |
|
| 39 |
- **Private by design.** No tracking, no telemetry, no accounts. Search history, cached results, and chats are stored in your browser and never leave your machine.
|
| 40 |
-
- **AI in your browser.** Pick from several curated models
|
| 41 |
- **Any backend you like.** Connect an OpenAI-compatible API (Ollama, LM Studio, vLLM, llama.cpp server, or a hosted provider), use the crowdsourced AI Horde, or let the server proxy your own API without exposing its key.
|
| 42 |
- **A real search pipeline.** Text and image results aggregated by SearXNG, reranked locally by a cross-encoder model, cached, and rate-limited; all inside the container.
|
| 43 |
- **Answers you can verify.** Responses cite the sources they draw from, support follow-up questions with conversation memory, reveal the model's reasoning on demand, and can be read aloud.
|
|
|
|
| 37 |
## Features
|
| 38 |
|
| 39 |
- **Private by design.** No tracking, no telemetry, no accounts. Search history, cached results, and chats are stored in your browser and never leave your machine.
|
| 40 |
+
- **AI in your browser.** Pick from several curated models that run on WebGPU where available and on CPU elsewhere. Models are downloaded once and cached by the browser.
|
| 41 |
- **Any backend you like.** Connect an OpenAI-compatible API (Ollama, LM Studio, vLLM, llama.cpp server, or a hosted provider), use the crowdsourced AI Horde, or let the server proxy your own API without exposing its key.
|
| 42 |
- **A real search pipeline.** Text and image results aggregated by SearXNG, reranked locally by a cross-encoder model, cached, and rate-limited; all inside the container.
|
| 43 |
- **Answers you can verify.** Responses cite the sources they draw from, support follow-up questions with conversation memory, reveal the model's reasoning on demand, and can be read aloud.
|
agents.md
CHANGED
|
@@ -5,11 +5,13 @@ This is your navigation hub. Start here, follow the links, and return when you n
|
|
| 5 |
## Before You Start
|
| 6 |
|
| 7 |
**New to this codebase?** Read in this order:
|
|
|
|
| 8 |
1. `docs/quick-start.md` - Get it running
|
| 9 |
2. `docs/overview.md` - Understand the system
|
| 10 |
3. `docs/project-structure.md` - Navigate the code
|
| 11 |
|
| 12 |
**Making changes?** Check:
|
|
|
|
| 13 |
- `docs/coding-conventions.md` - Code style
|
| 14 |
- `docs/development-commands.md` - Available commands
|
| 15 |
- `docs/pull-requests.md` - How to submit
|
|
@@ -32,6 +34,7 @@ This is your navigation hub. Start here, follow the links, and return when you n
|
|
| 32 |
|
| 33 |
- **`docs/api.md`** - HTTP endpoint reference: parameters, responses, and error codes
|
| 34 |
- **`docs/ai-integration.md`** - AI inference types (Wllama, OpenAI, AI Horde, Internal)
|
|
|
|
| 35 |
- **`docs/ui-components.md`** - Component architecture and PubSub patterns
|
| 36 |
- **`docs/search-history.md`** - History database schema and management
|
| 37 |
- **`docs/conversation-memory.md`** - Token budgeting and rolling summaries
|
|
@@ -188,7 +191,7 @@ Need to:
|
|
| 188 |
|
| 189 |
### Add a new AI model
|
| 190 |
|
| 191 |
-
1. Check the wasm build can run it: the architecture must exist in `node_modules/@wllama/wllama/esm/wasm/wllama.wasm` (grep the binary for the arch name), and the quant must stay under the ceilings in `docs/ai-integration.md` (shard every file under 2000 MiB; total weights under ~3.7 GB)
|
| 192 |
2. Shard the GGUF (`llama-gguf-split --split --size`, or an equivalent) and publish to a `Felladrin/gguf-sharded-<QUANT>-<Name>` repo
|
| 193 |
3. Add to `client/modules/wllama.ts`; thinking-by-default templates need nothing extra (their `reasoning_content` channel is streamed into the thinking UI) but do need `reasoningBudgetTokens` so they cannot think past the context without answering
|
| 194 |
4. Run `node scripts/wllama-catalog-check.cjs` and `npx vitest run client/modules/wllama.test.ts`
|
|
@@ -222,6 +225,7 @@ Need to:
|
|
| 222 |
## Quality Gates
|
| 223 |
|
| 224 |
Before any change:
|
|
|
|
| 225 |
```bash
|
| 226 |
npm run lint
|
| 227 |
```
|
|
@@ -229,6 +233,7 @@ npm run lint
|
|
| 229 |
Inside the dev container, the same command is `docker compose exec development-server npm run lint`.
|
| 230 |
|
| 231 |
This runs:
|
|
|
|
| 232 |
- Biome (formatting/linting)
|
| 233 |
- TypeScript (type checking)
|
| 234 |
- knip (dead code detection)
|
|
@@ -251,21 +256,25 @@ So `npx tsc --noEmit` reports success while type errors sit in `server/` or `eva
|
|
| 251 |
## Agent-First Principles
|
| 252 |
|
| 253 |
**Repository as System of Record:**
|
|
|
|
| 254 |
- All knowledge lives in versioned docs/ structure
|
| 255 |
- This file is your entry point - start here
|
| 256 |
- Follow links, don't assume - verify in code
|
| 257 |
|
| 258 |
**Context Efficiency:**
|
|
|
|
| 259 |
- Use this map to navigate quickly
|
| 260 |
- Return to this file when context drifts
|
| 261 |
- Follow the decision tree for common tasks
|
| 262 |
|
| 263 |
**Architecture & Boundaries:**
|
|
|
|
| 264 |
- Respect PubSub boundaries - don't cross concerns
|
| 265 |
- Client vs server - keep them separate
|
| 266 |
- Feature-based organization - one folder per feature
|
| 267 |
|
| 268 |
**Documentation Maintenance:**
|
|
|
|
| 269 |
- Update these docs when you learn something new
|
| 270 |
- Add cross-references when linking concepts
|
| 271 |
- Keep examples current with actual code
|
|
|
|
| 5 |
## Before You Start
|
| 6 |
|
| 7 |
**New to this codebase?** Read in this order:
|
| 8 |
+
|
| 9 |
1. `docs/quick-start.md` - Get it running
|
| 10 |
2. `docs/overview.md` - Understand the system
|
| 11 |
3. `docs/project-structure.md` - Navigate the code
|
| 12 |
|
| 13 |
**Making changes?** Check:
|
| 14 |
+
|
| 15 |
- `docs/coding-conventions.md` - Code style
|
| 16 |
- `docs/development-commands.md` - Available commands
|
| 17 |
- `docs/pull-requests.md` - How to submit
|
|
|
|
| 34 |
|
| 35 |
- **`docs/api.md`** - HTTP endpoint reference: parameters, responses, and error codes
|
| 36 |
- **`docs/ai-integration.md`** - AI inference types (Wllama, OpenAI, AI Horde, Internal)
|
| 37 |
+
- **`docs/wllama-model-evaluation.md`** - How in-browser models are scored on MiniSearch's task
|
| 38 |
- **`docs/ui-components.md`** - Component architecture and PubSub patterns
|
| 39 |
- **`docs/search-history.md`** - History database schema and management
|
| 40 |
- **`docs/conversation-memory.md`** - Token budgeting and rolling summaries
|
|
|
|
| 191 |
|
| 192 |
### Add a new AI model
|
| 193 |
|
| 194 |
+
1. Check the wasm build can run it: the architecture must exist in `node_modules/@wllama/wllama/esm/wasm/wllama.wasm` (grep the binary for the arch name), and the quant must stay under the ceilings in `docs/ai-integration.md` (shard every file under 2000 MiB; total weights under ~3.7 GB). Then run it for real with `npm run bench:wllama -- --model owner/repo/file.gguf` (see `docs/development-commands.md#wllama-benchmark`), on both devices: speed and answer quality can rank differently on WebGPU and on CPU
|
| 195 |
2. Shard the GGUF (`llama-gguf-split --split --size`, or an equivalent) and publish to a `Felladrin/gguf-sharded-<QUANT>-<Name>` repo
|
| 196 |
3. Add to `client/modules/wllama.ts`; thinking-by-default templates need nothing extra (their `reasoning_content` channel is streamed into the thinking UI) but do need `reasoningBudgetTokens` so they cannot think past the context without answering
|
| 197 |
4. Run `node scripts/wllama-catalog-check.cjs` and `npx vitest run client/modules/wllama.test.ts`
|
|
|
|
| 225 |
## Quality Gates
|
| 226 |
|
| 227 |
Before any change:
|
| 228 |
+
|
| 229 |
```bash
|
| 230 |
npm run lint
|
| 231 |
```
|
|
|
|
| 233 |
Inside the dev container, the same command is `docker compose exec development-server npm run lint`.
|
| 234 |
|
| 235 |
This runs:
|
| 236 |
+
|
| 237 |
- Biome (formatting/linting)
|
| 238 |
- TypeScript (type checking)
|
| 239 |
- knip (dead code detection)
|
|
|
|
| 256 |
## Agent-First Principles
|
| 257 |
|
| 258 |
**Repository as System of Record:**
|
| 259 |
+
|
| 260 |
- All knowledge lives in versioned docs/ structure
|
| 261 |
- This file is your entry point - start here
|
| 262 |
- Follow links, don't assume - verify in code
|
| 263 |
|
| 264 |
**Context Efficiency:**
|
| 265 |
+
|
| 266 |
- Use this map to navigate quickly
|
| 267 |
- Return to this file when context drifts
|
| 268 |
- Follow the decision tree for common tasks
|
| 269 |
|
| 270 |
**Architecture & Boundaries:**
|
| 271 |
+
|
| 272 |
- Respect PubSub boundaries - don't cross concerns
|
| 273 |
- Client vs server - keep them separate
|
| 274 |
- Feature-based organization - one folder per feature
|
| 275 |
|
| 276 |
**Documentation Maintenance:**
|
| 277 |
+
|
| 278 |
- Update these docs when you learn something new
|
| 279 |
- Add cross-references when linking concepts
|
| 280 |
- Keep examples current with actual code
|
changelog.md
CHANGED
|
@@ -2,6 +2,16 @@
|
|
| 2 |
|
| 3 |
Notable changes, grouped by date, most recent first.
|
| 4 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
|
| 6 |
## 2026-11-05
|
| 7 |
|
|
@@ -10,34 +20,28 @@ Notable changes, grouped by date, most recent first.
|
|
| 10 |
- A result from the site the query names gets a bonus on its semantic score before the score filter: +0.5 when a query word equals the domain's label ("disney" and disney.com, subdomains included), +0.75 when the label is the query's own words joined ("disney plus" and disneyplus.com, "coca cola" and coca-cola.com). The bonus is added rather than multiplied because the reranker returns a raw logit that is usually negative, where a factor above one would demote the named site instead of promoting it. Substrings do not count (disneyfan.com gets nothing), filler words never match on their own (plus.com, news.com), and a query that names a year will not promote a page that states a different year, so a brand's recap of last year's event cannot displace fresher reporting; a page that states no year at all keeps the bonus, because relative-time queries are anchored to a year before they reach the search. A label that spells only the start of the query leaves words unused and gets nothing (`tesla investor relations` and teslainvestor.com), a label that starts with a filler word is skipped (`the disney store` and thedisney.com), a split label counts from four characters, and a trailing dot on the hostname changes nothing. The site's name comes from the public suffix list, private section included (the `tldts` package), so bbc.co.uk answers "bbc" and a tenant of a shared host such as someone.github.io answers its own name, not "github". The pinned top result is never boosted, so the score spread the client reads relevance tags from keeps its shape. The reranker still never sees the URL, and a result the search did not return cannot be added here: the boost reorders and re-selects what the pipeline already holds, which is enough to move a page into the six the model reads. `server/authorityRanking.ts` holds the rule and `docs/reranking.md` records it.
|
| 11 |
- A text search that asks for an official page can now reach the official site even when the engine's results never included it, behind a new setting, `SEARCH_OFFICIAL_LANE_ENABLED`, which is off by default because it sends the engines a second request. When it is on, the query carries navigation intent (`ๅฎ็ฝ`, `ๆ็`, `่ฅไธๆถ้ด`, `official`, `hours`, `showtimes` and the like, in Chinese and English only), and one of the returned results names a site, the server sends a second, `site:`-scoped query to SearXNG and merges up to three of its results, deduplicated by URL and kept to that registrable domain, behind the pinned first result: the ranking re-sorts only what follows the top result, so a page appended past that window would never be re-sorted at all. The domain has to come from the results themselves, by the same rule the score bonus uses, because guessing `<token>.com` from the query would turn a search box into a scanner; a query that names a brand whose official site is absent from every result gets no second query. The second query is one request with no retry, and its failures count toward neither the SearXNG circuit nor the engine health in `/status`, so it cannot slow down or degrade the searches users are waiting on. SearXNG still sees the extra request, so an engine that suspends the instance after a CAPTCHA does so for every search. It runs only when SearXNG served the search (never after the fallback), never uses the last five seconds of the search deadline, and is canceled at eight seconds; a failure only logs and is counted, so it can add results but never turn a served search into a 502. `/status` reports `searches.officialLane` (searches, `resultsAdded`, `failures`), and `server/officialLane.ts` holds the rule.
|
| 12 |
|
| 13 |
-
|
| 14 |
## 2026-11-03
|
| 15 |
|
| 16 |
- The five thinking models added on 2026-11-02 (Spark X2.6 1.7B and 4B, Granite 4.2 3B, Qwen 3.5 4B, Gemma 4 E2B) no longer ship with thinking switched off. Their chain of thought arrives on wllama's `reasoning_content` channel and is now wrapped in the configured reasoning markers, so the existing thinking UI shows it while the model reasons and the answer streams in after. The model warmup counts a reasoning token as a working model, so thinking models no longer log a failed warmup.
|
| 17 |
- The same five models now cap their chain of thought with `reasoningBudgetTokens` (wllama's `reasoning_budget_tokens`, 256 tokens): when the model has thought that long without answering, the engine closes the reasoning channel and forces the final answer. Without the cap, Spark X2.6 and Granite 4.2 could spend the whole 4096-token context on reasoning and deliver no answer at all.
|
| 18 |
|
| 19 |
-
|
| 20 |
## 2026-11-02
|
| 21 |
|
| 22 |
- Five new browser models joined the Wllama catalog: Spark X2.6 1.7B, Granite 4.2 3B, Qwen 3.5 4B, Gemma 4 E2B and Spark X2.6 4B, each tested end to end against the shipped wasm build with the search-answer prompt. The Gemma 4 and Qwen 3.5 architectures were previously absent from the catalog, and the new entries run on the wasm engine's supported architectures (`gemma4`, `qwen35`, `spark2_5`, `granite`). Models whose chat template thinks by default ship with `chatTemplateKwargs: { enable_thinking: false }`, a new per-model field passed to wllama as `chat_template_kwargs`, because MiniSearch renders only the content channel and a thinking template would otherwise spend the whole token budget on reasoning and answer with nothing. `docs/ai-integration.md` now records the two measured ceilings a candidate must fit (shard files under 2000 MiB, since Chromium's OPFS silently truncates beyond it; total weights under ~3.7 GB, since the wasm linear memory caps at 4 GiB), and `npm run wllama-catalog-check` HEADs every shard of every catalog model on HuggingFace to verify the claimed sizes.
|
| 23 |
|
| 24 |
-
|
| 25 |
## 2026-10-10
|
| 26 |
|
| 27 |
- The text-search fallback now cascades over Parallel and You.com instead of asking a single provider. On each search the configured providers are tried in a random order, one after the other, until one returns results, under the same 15-second budget, which is split across the providers still untried and not paused so a hanging one cannot starve the rest. A provider that answers HTTP 429 is paused until its `Retry-After` and skipped meanwhile, as You.com's acceptable use policy asks. A single search can now reach more than one provider; each sees the query text and this server's IP address, never the user's. `SEARCH_FALLBACK_PROVIDERS` picks the providers (all of them when empty), and `SEARCH_FALLBACK_PARALLEL_API_KEY` and `SEARCH_FALLBACK_YOUCOM_API_KEY` hold the optional keys. `SEARCH_FALLBACK_PROVIDER` and `SEARCH_FALLBACK_API_KEY` are removed: if either is still set while the fallback is enabled, the server stops at start with an error naming the replacement, and so does a provider key with a control character inside it. `/status` gained per-provider counters under `searches.fallbackProviders`.
|
| 28 |
|
| 29 |
-
|
| 30 |
## 2026-10-08
|
| 31 |
|
| 32 |
- `/status` now reports `authorization.distinctRejectedTokens`, the number of distinct dead tokens the server has cached and keeps refusing. A steady stream of invalid-token rejections with a value of 1 there is one client replaying one stale token, not many old search links, which the existing counters could not tell apart.
|
| 33 |
- Added an optional second source for text searches, so a SearXNG outage no longer has to end in a 502. It stays off unless `SEARCH_FALLBACK_ENABLED` is `true` or `1`, and flipping it takes a restart, since a process keeps the environment it was started with. When it is on, the second source is consulted only after SearXNG has failed (an HTTP error, a network error, every engine unresponsive, or an open circuit), under one 15-second budget for the whole exchange and under a 25-second deadline measured from the start of the search, and only for text searches: image searches still answer 502 as before. The query goes to `search.parallel.ai`, which sees the query text and this server's IP address, never the user's. A retry that comes back with nothing usable now answers HTTP 200 with no results, so the UI shows "No results found" rather than "Text search unavailable"; a retry that fails itself, or that the search deadline has left under two seconds for, falls through to the old 502. `SEARCH_FALLBACK_API_KEY` is optional and is sent as a Bearer token, which raises the provider's free-tier rate limits. `/status` gained `searchesServedByFallback` and `searchesFailedOnFallback`, plain counts that never carry a query, so you can see whether the second source is earning its place.
|
| 34 |
|
| 35 |
-
|
| 36 |
## 2026-10-07
|
| 37 |
|
| 38 |
- Security: cleared the flagged copy of `postcss-selector-parser` that lives inside the npm the image ships. npm pins its own dependency tree exactly, so neither `npm install -g npm@latest` nor a dependency bump reaches a bundled copy, and the npm in the image still carried 7.1.4. This was the one of the image's three code scanning findings that no dependency bump could fix. The image's npm-bundle repair now lists that package among its targets, so the build replaces the bundled copy with the patched release and fails if any copy of that major line survives.
|
| 39 |
|
| 40 |
-
|
| 41 |
## 2026-09-23
|
| 42 |
|
| 43 |
- Moved the Dictate control inside the search field, as an icon button at its right edge, and gave the follow-up question box the same control, so dictating a follow-up no longer means typing it. It shows a microphone when idle, a red stop button while it listens, and the model download progress in its tooltip and its accessible name. Pressing it never takes the focus off the field, so the caret stays where it was and a phone keeps its on-screen keyboard open. In the chat box it is disabled rather than hidden while an answer is generating, and a session still running when generation starts is stopped instead of writing into a field that has gone read-only.
|
|
|
|
| 2 |
|
| 3 |
Notable changes, grouped by date, most recent first.
|
| 4 |
|
| 5 |
+
## 2026-11-08
|
| 6 |
+
|
| 7 |
+
- A model download that was interrupted, for example by reloading the page halfway, no longer stops every model from loading. wllama 3.8.1 checks every model in its cache before loading any, and throws on the first one that misses a file; before each load, MiniSearch now deletes the files of models whose download did not finish, so only that model downloads again.
|
| 8 |
+
- The default system prompt's citation example is now a placeholder, `[example.com](https://example.com/page)`, instead of a real YouTube video link that small models copied as if it were a source. On the benchmark's eleven core questions, over 110 answers from five small models, links to URLs that are not among the results fell from 12 to 7, the YouTube link disappeared, and the share of answers citing a result stayed level (48% before, 50% after). A saved prompt that is exactly the previous default moves to the new one; an edited prompt stays as it is.
|
| 9 |
+
- Four models leave the Wllama catalog because they answer worse than LittleLamb 290M, the default for browsers without WebGPU and the catalog's quality floor: StableLM 2 Zephyr 1.6B, SmolLM 2 135M, Granite 4.0 350M and Pints 1.5B (16 to 25 points out of 84 on the benchmark, against 38 for LittleLamb). Granite 4.0 1B goes too: on WebGPU it writes only `@` characters until it is stopped, though it answers well on the CPU. `docs/wllama-model-evaluation.md` now holds the scores of every catalog model and the rule that decided.
|
| 10 |
+
- A browser whose saved settings point at a model that left the catalog switches to the default model on its next answer, and records the switch in the logs. Before, it failed to answer at all.
|
| 11 |
+
- Browsers with WebGPU now start on their own default model, Gemma 3 4B, set with the new `WLLAMA_DEFAULT_GPU_MODEL_ID`; browsers without it keep LittleLamb 290M from `WLLAMA_DEFAULT_MODEL_ID`, which runs well on the CPU. Both only seed a profile that never saved settings. The GPU variable does not fall back to the CPU one, because `.env.example` sets that one and most deployments would otherwise hand GPUs the CPU model; a deployment that pinned `WLLAMA_DEFAULT_MODEL_ID` and wants the same model on GPUs should set both. Gemma 3 4B scored highest of twelve GPU candidates on the benchmark (`docs/wllama-model-evaluation.md`), with every point in two runs of fifteen questions in eleven languages, and on WebGPU it started answering in about 8 seconds.
|
| 12 |
+
- Qwen 3.5 0.8B (533 MB) joins the Wllama catalog. On the benchmark's eleven core questions it scored 70 of 84 (one run), above larger models such as Qwen 3 1.7B (66) and Granite 4.2 3B (59), and on WebGPU it writes about 32 tokens per second. Like Qwen 3.5 4B it ignores `reasoning_effort`, so it thinks for the full 1024 tokens before most answers (about 33 seconds on an Apple M1 Max).
|
| 13 |
+
- The README and the docs no longer give a number of models or a size range, so adding or removing a model does not leave them out of date.
|
| 14 |
+
- Thinking models get a 1024-token reasoning budget instead of 256, because a chain of thought cut off that early degraded the answer. Every model is also asked for `reasoning_effort: "low"`, which Granite 4.2 and Spark X2.6 honor by thinking less; the other templates ignore it. Five models that think by default had no budget at all and now get the same 1024 tokens: LittleLamb 290M, Qwen 3 0.6B, Qwen 3 1.7B, Polaris 4B and DeepSeek R1 Distill Qwen 1.5B. LittleLamb, the default for browsers without WebGPU, thought for up to 3,663 tokens on a single answer and sometimes never answered; with the budget it answered all 33 questions of three benchmark runs.
|
| 15 |
|
| 16 |
## 2026-11-05
|
| 17 |
|
|
|
|
| 20 |
- A result from the site the query names gets a bonus on its semantic score before the score filter: +0.5 when a query word equals the domain's label ("disney" and disney.com, subdomains included), +0.75 when the label is the query's own words joined ("disney plus" and disneyplus.com, "coca cola" and coca-cola.com). The bonus is added rather than multiplied because the reranker returns a raw logit that is usually negative, where a factor above one would demote the named site instead of promoting it. Substrings do not count (disneyfan.com gets nothing), filler words never match on their own (plus.com, news.com), and a query that names a year will not promote a page that states a different year, so a brand's recap of last year's event cannot displace fresher reporting; a page that states no year at all keeps the bonus, because relative-time queries are anchored to a year before they reach the search. A label that spells only the start of the query leaves words unused and gets nothing (`tesla investor relations` and teslainvestor.com), a label that starts with a filler word is skipped (`the disney store` and thedisney.com), a split label counts from four characters, and a trailing dot on the hostname changes nothing. The site's name comes from the public suffix list, private section included (the `tldts` package), so bbc.co.uk answers "bbc" and a tenant of a shared host such as someone.github.io answers its own name, not "github". The pinned top result is never boosted, so the score spread the client reads relevance tags from keeps its shape. The reranker still never sees the URL, and a result the search did not return cannot be added here: the boost reorders and re-selects what the pipeline already holds, which is enough to move a page into the six the model reads. `server/authorityRanking.ts` holds the rule and `docs/reranking.md` records it.
|
| 21 |
- A text search that asks for an official page can now reach the official site even when the engine's results never included it, behind a new setting, `SEARCH_OFFICIAL_LANE_ENABLED`, which is off by default because it sends the engines a second request. When it is on, the query carries navigation intent (`ๅฎ็ฝ`, `ๆ็`, `่ฅไธๆถ้ด`, `official`, `hours`, `showtimes` and the like, in Chinese and English only), and one of the returned results names a site, the server sends a second, `site:`-scoped query to SearXNG and merges up to three of its results, deduplicated by URL and kept to that registrable domain, behind the pinned first result: the ranking re-sorts only what follows the top result, so a page appended past that window would never be re-sorted at all. The domain has to come from the results themselves, by the same rule the score bonus uses, because guessing `<token>.com` from the query would turn a search box into a scanner; a query that names a brand whose official site is absent from every result gets no second query. The second query is one request with no retry, and its failures count toward neither the SearXNG circuit nor the engine health in `/status`, so it cannot slow down or degrade the searches users are waiting on. SearXNG still sees the extra request, so an engine that suspends the instance after a CAPTCHA does so for every search. It runs only when SearXNG served the search (never after the fallback), never uses the last five seconds of the search deadline, and is canceled at eight seconds; a failure only logs and is counted, so it can add results but never turn a served search into a 502. `/status` reports `searches.officialLane` (searches, `resultsAdded`, `failures`), and `server/officialLane.ts` holds the rule.
|
| 22 |
|
|
|
|
| 23 |
## 2026-11-03
|
| 24 |
|
| 25 |
- The five thinking models added on 2026-11-02 (Spark X2.6 1.7B and 4B, Granite 4.2 3B, Qwen 3.5 4B, Gemma 4 E2B) no longer ship with thinking switched off. Their chain of thought arrives on wllama's `reasoning_content` channel and is now wrapped in the configured reasoning markers, so the existing thinking UI shows it while the model reasons and the answer streams in after. The model warmup counts a reasoning token as a working model, so thinking models no longer log a failed warmup.
|
| 26 |
- The same five models now cap their chain of thought with `reasoningBudgetTokens` (wllama's `reasoning_budget_tokens`, 256 tokens): when the model has thought that long without answering, the engine closes the reasoning channel and forces the final answer. Without the cap, Spark X2.6 and Granite 4.2 could spend the whole 4096-token context on reasoning and deliver no answer at all.
|
| 27 |
|
|
|
|
| 28 |
## 2026-11-02
|
| 29 |
|
| 30 |
- Five new browser models joined the Wllama catalog: Spark X2.6 1.7B, Granite 4.2 3B, Qwen 3.5 4B, Gemma 4 E2B and Spark X2.6 4B, each tested end to end against the shipped wasm build with the search-answer prompt. The Gemma 4 and Qwen 3.5 architectures were previously absent from the catalog, and the new entries run on the wasm engine's supported architectures (`gemma4`, `qwen35`, `spark2_5`, `granite`). Models whose chat template thinks by default ship with `chatTemplateKwargs: { enable_thinking: false }`, a new per-model field passed to wllama as `chat_template_kwargs`, because MiniSearch renders only the content channel and a thinking template would otherwise spend the whole token budget on reasoning and answer with nothing. `docs/ai-integration.md` now records the two measured ceilings a candidate must fit (shard files under 2000 MiB, since Chromium's OPFS silently truncates beyond it; total weights under ~3.7 GB, since the wasm linear memory caps at 4 GiB), and `npm run wllama-catalog-check` HEADs every shard of every catalog model on HuggingFace to verify the claimed sizes.
|
| 31 |
|
|
|
|
| 32 |
## 2026-10-10
|
| 33 |
|
| 34 |
- The text-search fallback now cascades over Parallel and You.com instead of asking a single provider. On each search the configured providers are tried in a random order, one after the other, until one returns results, under the same 15-second budget, which is split across the providers still untried and not paused so a hanging one cannot starve the rest. A provider that answers HTTP 429 is paused until its `Retry-After` and skipped meanwhile, as You.com's acceptable use policy asks. A single search can now reach more than one provider; each sees the query text and this server's IP address, never the user's. `SEARCH_FALLBACK_PROVIDERS` picks the providers (all of them when empty), and `SEARCH_FALLBACK_PARALLEL_API_KEY` and `SEARCH_FALLBACK_YOUCOM_API_KEY` hold the optional keys. `SEARCH_FALLBACK_PROVIDER` and `SEARCH_FALLBACK_API_KEY` are removed: if either is still set while the fallback is enabled, the server stops at start with an error naming the replacement, and so does a provider key with a control character inside it. `/status` gained per-provider counters under `searches.fallbackProviders`.
|
| 35 |
|
|
|
|
| 36 |
## 2026-10-08
|
| 37 |
|
| 38 |
- `/status` now reports `authorization.distinctRejectedTokens`, the number of distinct dead tokens the server has cached and keeps refusing. A steady stream of invalid-token rejections with a value of 1 there is one client replaying one stale token, not many old search links, which the existing counters could not tell apart.
|
| 39 |
- Added an optional second source for text searches, so a SearXNG outage no longer has to end in a 502. It stays off unless `SEARCH_FALLBACK_ENABLED` is `true` or `1`, and flipping it takes a restart, since a process keeps the environment it was started with. When it is on, the second source is consulted only after SearXNG has failed (an HTTP error, a network error, every engine unresponsive, or an open circuit), under one 15-second budget for the whole exchange and under a 25-second deadline measured from the start of the search, and only for text searches: image searches still answer 502 as before. The query goes to `search.parallel.ai`, which sees the query text and this server's IP address, never the user's. A retry that comes back with nothing usable now answers HTTP 200 with no results, so the UI shows "No results found" rather than "Text search unavailable"; a retry that fails itself, or that the search deadline has left under two seconds for, falls through to the old 502. `SEARCH_FALLBACK_API_KEY` is optional and is sent as a Bearer token, which raises the provider's free-tier rate limits. `/status` gained `searchesServedByFallback` and `searchesFailedOnFallback`, plain counts that never carry a query, so you can see whether the second source is earning its place.
|
| 40 |
|
|
|
|
| 41 |
## 2026-10-07
|
| 42 |
|
| 43 |
- Security: cleared the flagged copy of `postcss-selector-parser` that lives inside the npm the image ships. npm pins its own dependency tree exactly, so neither `npm install -g npm@latest` nor a dependency bump reaches a bundled copy, and the npm in the image still carried 7.1.4. This was the one of the image's three code scanning findings that no dependency bump could fix. The image's npm-bundle repair now lists that package among its targets, so the build replaces the bundled copy with the patched release and fails if any copy of that major line survives.
|
| 44 |
|
|
|
|
| 45 |
## 2026-09-23
|
| 46 |
|
| 47 |
- Moved the Dictate control inside the search field, as an icon button at its right edge, and gave the follow-up question box the same control, so dictating a follow-up no longer means typing it. It shows a microphone when idle, a red stop button while it listens, and the model download progress in its tooltip and its accessible name. Pressing it never takes the focus off the field, so the caret stays where it was and a phone keeps its on-screen keyboard open. In the chat box it is disabled rather than hidden while an answer is generating, and a session still running when generation starts is stopped instead of writing into a field that has gone read-only.
|
client/components/App/App.test.tsx
CHANGED
|
@@ -25,6 +25,7 @@ const serverConfig: ServerConfig = {
|
|
| 25 |
accessKeysEnabled: false,
|
| 26 |
accessKeyTimeoutHours: 24,
|
| 27 |
wllamaDefaultModelId: "littlelamb-290m",
|
|
|
|
| 28 |
internalApiEnabled: false,
|
| 29 |
internalApiName: "Internal API",
|
| 30 |
defaultInferenceType: "browser",
|
|
|
|
| 25 |
accessKeysEnabled: false,
|
| 26 |
accessKeyTimeoutHours: 24,
|
| 27 |
wllamaDefaultModelId: "littlelamb-290m",
|
| 28 |
+
wllamaDefaultGpuModelId: "gemma-3-4b",
|
| 29 |
internalApiEnabled: false,
|
| 30 |
internalApiName: "Internal API",
|
| 31 |
defaultInferenceType: "browser",
|
client/components/App/App.tsx
CHANGED
|
@@ -17,7 +17,9 @@ import {
|
|
| 17 |
applyServerConfig,
|
| 18 |
defaultSettings,
|
| 19 |
hasStoredUserSettings,
|
|
|
|
| 20 |
} from "@/modules/settings";
|
|
|
|
| 21 |
import "@mantine/notifications/styles.css";
|
| 22 |
import { verifyStoredAccessKey } from "@/modules/accessKey";
|
| 23 |
import {
|
|
@@ -115,9 +117,10 @@ function useInitializeSettings(config: ConfigState) {
|
|
| 115 |
|
| 116 |
setSettings(
|
| 117 |
applyServerConfig(
|
| 118 |
-
{ ...defaultSettings, ...settings },
|
| 119 |
config === "unavailable" ? FALLBACK_CONFIG : config,
|
| 120 |
hasStoredUserSettings,
|
|
|
|
| 121 |
),
|
| 122 |
);
|
| 123 |
setSettingsInitialized(true);
|
|
|
|
| 17 |
applyServerConfig,
|
| 18 |
defaultSettings,
|
| 19 |
hasStoredUserSettings,
|
| 20 |
+
upgradeUntouchedSystemPrompt,
|
| 21 |
} from "@/modules/settings";
|
| 22 |
+
import { isWebGPUAvailable } from "@/modules/webGpu";
|
| 23 |
import "@mantine/notifications/styles.css";
|
| 24 |
import { verifyStoredAccessKey } from "@/modules/accessKey";
|
| 25 |
import {
|
|
|
|
| 117 |
|
| 118 |
setSettings(
|
| 119 |
applyServerConfig(
|
| 120 |
+
upgradeUntouchedSystemPrompt({ ...defaultSettings, ...settings }),
|
| 121 |
config === "unavailable" ? FALLBACK_CONFIG : config,
|
| 122 |
hasStoredUserSettings,
|
| 123 |
+
isWebGPUAvailable,
|
| 124 |
),
|
| 125 |
);
|
| 126 |
setSettingsInitialized(true);
|
client/modules/config.test.ts
CHANGED
|
@@ -8,6 +8,7 @@ const serverConfig: ServerConfig = {
|
|
| 8 |
accessKeysEnabled: true,
|
| 9 |
accessKeyTimeoutHours: 24,
|
| 10 |
wllamaDefaultModelId: "some-model",
|
|
|
|
| 11 |
internalApiEnabled: true,
|
| 12 |
internalApiName: "Custom LLM",
|
| 13 |
defaultInferenceType: "internal",
|
|
|
|
| 8 |
accessKeysEnabled: true,
|
| 9 |
accessKeyTimeoutHours: 24,
|
| 10 |
wllamaDefaultModelId: "some-model",
|
| 11 |
+
wllamaDefaultGpuModelId: "some-gpu-model",
|
| 12 |
internalApiEnabled: true,
|
| 13 |
internalApiName: "Custom LLM",
|
| 14 |
defaultInferenceType: "internal",
|
client/modules/config.ts
CHANGED
|
@@ -9,6 +9,7 @@
|
|
| 9 |
import {
|
| 10 |
DEFAULT_INFERENCE_TYPE,
|
| 11 |
DEFAULT_INTERNAL_API_NAME,
|
|
|
|
| 12 |
DEFAULT_WLLAMA_MODEL_ID,
|
| 13 |
type ServerConfig,
|
| 14 |
} from "@shared/serverConfig";
|
|
@@ -28,6 +29,7 @@ export const FALLBACK_CONFIG: ServerConfig = {
|
|
| 28 |
accessKeysEnabled: false,
|
| 29 |
accessKeyTimeoutHours: 0,
|
| 30 |
wllamaDefaultModelId: DEFAULT_WLLAMA_MODEL_ID,
|
|
|
|
| 31 |
internalApiEnabled: false,
|
| 32 |
internalApiName: DEFAULT_INTERNAL_API_NAME,
|
| 33 |
defaultInferenceType: DEFAULT_INFERENCE_TYPE,
|
|
|
|
| 9 |
import {
|
| 10 |
DEFAULT_INFERENCE_TYPE,
|
| 11 |
DEFAULT_INTERNAL_API_NAME,
|
| 12 |
+
DEFAULT_WLLAMA_GPU_MODEL_ID,
|
| 13 |
DEFAULT_WLLAMA_MODEL_ID,
|
| 14 |
type ServerConfig,
|
| 15 |
} from "@shared/serverConfig";
|
|
|
|
| 29 |
accessKeysEnabled: false,
|
| 30 |
accessKeyTimeoutHours: 0,
|
| 31 |
wllamaDefaultModelId: DEFAULT_WLLAMA_MODEL_ID,
|
| 32 |
+
wllamaDefaultGpuModelId: DEFAULT_WLLAMA_GPU_MODEL_ID,
|
| 33 |
internalApiEnabled: false,
|
| 34 |
internalApiName: DEFAULT_INTERNAL_API_NAME,
|
| 35 |
defaultInferenceType: DEFAULT_INFERENCE_TYPE,
|
client/modules/pubSub.ts
CHANGED
|
@@ -117,7 +117,8 @@ export const settingsPubSub = createLocalStoragePubSub(
|
|
| 117 |
defaultSettings,
|
| 118 |
);
|
| 119 |
|
| 120 |
-
export const [, listenToSettingsChanges, getSettings] =
|
|
|
|
| 121 |
|
| 122 |
export const modelSizeInMegabytesPubSub = createPubSub(0);
|
| 123 |
|
|
|
|
| 117 |
defaultSettings,
|
| 118 |
);
|
| 119 |
|
| 120 |
+
export const [updateSettings, listenToSettingsChanges, getSettings] =
|
| 121 |
+
settingsPubSub;
|
| 122 |
|
| 123 |
export const modelSizeInMegabytesPubSub = createPubSub(0);
|
| 124 |
|
client/modules/settings.test.ts
CHANGED
|
@@ -1,3 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import { describe, expect, it, vi } from "vitest";
|
| 2 |
import type { ServerConfig } from "./config";
|
| 3 |
import {
|
|
@@ -6,12 +10,14 @@ import {
|
|
| 6 |
getDefaultCpuThreads,
|
| 7 |
getInferenceTypes,
|
| 8 |
prefersLocalDictationModel,
|
|
|
|
| 9 |
} from "./settings";
|
| 10 |
|
| 11 |
const mockConfig: ServerConfig = {
|
| 12 |
accessKeysEnabled: false,
|
| 13 |
accessKeyTimeoutHours: 0,
|
| 14 |
wllamaDefaultModelId: "littlelamb-290m",
|
|
|
|
| 15 |
internalApiEnabled: false,
|
| 16 |
internalApiName: "Internal API",
|
| 17 |
defaultInferenceType: "browser",
|
|
@@ -102,11 +108,25 @@ describe("Settings Module", () => {
|
|
| 102 |
wllamaDefaultModelId: "custom-model",
|
| 103 |
defaultInferenceType: "internal",
|
| 104 |
};
|
| 105 |
-
const applied = applyServerConfig(defaultSettings, config, false);
|
| 106 |
expect(applied.wllamaModelId).toBe("custom-model");
|
| 107 |
expect(applied.inferenceType).toBe("internal");
|
| 108 |
});
|
| 109 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 110 |
it("should preserve stored settings that differ from the server defaults", () => {
|
| 111 |
const config = {
|
| 112 |
...mockConfig,
|
|
@@ -118,7 +138,7 @@ describe("Settings Module", () => {
|
|
| 118 |
wllamaModelId: "user-picked-model",
|
| 119 |
inferenceType: "horde",
|
| 120 |
};
|
| 121 |
-
const applied = applyServerConfig(userSettings, config, true);
|
| 122 |
expect(applied.wllamaModelId).toBe("user-picked-model");
|
| 123 |
expect(applied.inferenceType).toBe("horde");
|
| 124 |
});
|
|
@@ -129,7 +149,7 @@ describe("Settings Module", () => {
|
|
| 129 |
wllamaDefaultModelId: "admin-preferred-model",
|
| 130 |
defaultInferenceType: "internal",
|
| 131 |
};
|
| 132 |
-
const applied = applyServerConfig(defaultSettings, config, true);
|
| 133 |
expect(applied.wllamaModelId).toBe(defaultSettings.wllamaModelId);
|
| 134 |
expect(applied.inferenceType).toBe(defaultSettings.inferenceType);
|
| 135 |
});
|
|
@@ -203,3 +223,27 @@ describe("dictation model default", () => {
|
|
| 203 |
expect(settings.enableLocalDictationModel).toBe(true);
|
| 204 |
});
|
| 205 |
});
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import {
|
| 2 |
+
CITATION_EXAMPLE,
|
| 3 |
+
DEFAULT_SYSTEM_PROMPT,
|
| 4 |
+
} from "@shared/defaultSystemPrompt";
|
| 5 |
import { describe, expect, it, vi } from "vitest";
|
| 6 |
import type { ServerConfig } from "./config";
|
| 7 |
import {
|
|
|
|
| 10 |
getDefaultCpuThreads,
|
| 11 |
getInferenceTypes,
|
| 12 |
prefersLocalDictationModel,
|
| 13 |
+
upgradeUntouchedSystemPrompt,
|
| 14 |
} from "./settings";
|
| 15 |
|
| 16 |
const mockConfig: ServerConfig = {
|
| 17 |
accessKeysEnabled: false,
|
| 18 |
accessKeyTimeoutHours: 0,
|
| 19 |
wllamaDefaultModelId: "littlelamb-290m",
|
| 20 |
+
wllamaDefaultGpuModelId: "gemma-3-4b",
|
| 21 |
internalApiEnabled: false,
|
| 22 |
internalApiName: "Internal API",
|
| 23 |
defaultInferenceType: "browser",
|
|
|
|
| 108 |
wllamaDefaultModelId: "custom-model",
|
| 109 |
defaultInferenceType: "internal",
|
| 110 |
};
|
| 111 |
+
const applied = applyServerConfig(defaultSettings, config, false, false);
|
| 112 |
expect(applied.wllamaModelId).toBe("custom-model");
|
| 113 |
expect(applied.inferenceType).toBe("internal");
|
| 114 |
});
|
| 115 |
|
| 116 |
+
it("should seed the GPU model default when the browser has WebGPU", () => {
|
| 117 |
+
const config = {
|
| 118 |
+
...mockConfig,
|
| 119 |
+
wllamaDefaultModelId: "cpu-model",
|
| 120 |
+
wllamaDefaultGpuModelId: "gpu-model",
|
| 121 |
+
};
|
| 122 |
+
expect(
|
| 123 |
+
applyServerConfig(defaultSettings, config, false, true).wllamaModelId,
|
| 124 |
+
).toBe("gpu-model");
|
| 125 |
+
expect(
|
| 126 |
+
applyServerConfig(defaultSettings, config, false, false).wllamaModelId,
|
| 127 |
+
).toBe("cpu-model");
|
| 128 |
+
});
|
| 129 |
+
|
| 130 |
it("should preserve stored settings that differ from the server defaults", () => {
|
| 131 |
const config = {
|
| 132 |
...mockConfig,
|
|
|
|
| 138 |
wllamaModelId: "user-picked-model",
|
| 139 |
inferenceType: "horde",
|
| 140 |
};
|
| 141 |
+
const applied = applyServerConfig(userSettings, config, true, true);
|
| 142 |
expect(applied.wllamaModelId).toBe("user-picked-model");
|
| 143 |
expect(applied.inferenceType).toBe("horde");
|
| 144 |
});
|
|
|
|
| 149 |
wllamaDefaultModelId: "admin-preferred-model",
|
| 150 |
defaultInferenceType: "internal",
|
| 151 |
};
|
| 152 |
+
const applied = applyServerConfig(defaultSettings, config, true, true);
|
| 153 |
expect(applied.wllamaModelId).toBe(defaultSettings.wllamaModelId);
|
| 154 |
expect(applied.inferenceType).toBe(defaultSettings.inferenceType);
|
| 155 |
});
|
|
|
|
| 223 |
expect(settings.enableLocalDictationModel).toBe(true);
|
| 224 |
});
|
| 225 |
});
|
| 226 |
+
|
| 227 |
+
describe("upgradeUntouchedSystemPrompt", () => {
|
| 228 |
+
const previousDefault = DEFAULT_SYSTEM_PROMPT.replace(
|
| 229 |
+
CITATION_EXAMPLE,
|
| 230 |
+
"[youtube.com](https://www.youtube.com/watch?v=dQw4w9WgXcQ)",
|
| 231 |
+
);
|
| 232 |
+
|
| 233 |
+
it("replaces a saved prompt that is exactly the previous default", () => {
|
| 234 |
+
const upgraded = upgradeUntouchedSystemPrompt({
|
| 235 |
+
...defaultSettings,
|
| 236 |
+
systemPrompt: previousDefault,
|
| 237 |
+
});
|
| 238 |
+
expect(upgraded.systemPrompt).toBe(DEFAULT_SYSTEM_PROMPT);
|
| 239 |
+
});
|
| 240 |
+
|
| 241 |
+
it("keeps a prompt the user edited", () => {
|
| 242 |
+
const edited = `${previousDefault}\nAlways answer in haiku.`;
|
| 243 |
+
const kept = upgradeUntouchedSystemPrompt({
|
| 244 |
+
...defaultSettings,
|
| 245 |
+
systemPrompt: edited,
|
| 246 |
+
});
|
| 247 |
+
expect(kept.systemPrompt).toBe(edited);
|
| 248 |
+
});
|
| 249 |
+
});
|
client/modules/settings.ts
CHANGED
|
@@ -1,6 +1,10 @@
|
|
| 1 |
-
import {
|
|
|
|
|
|
|
|
|
|
| 2 |
import {
|
| 3 |
DEFAULT_INFERENCE_TYPE,
|
|
|
|
| 4 |
DEFAULT_WLLAMA_MODEL_ID,
|
| 5 |
type ServerConfig,
|
| 6 |
} from "@shared/serverConfig";
|
|
@@ -136,23 +140,51 @@ export function getInferenceTypes(config: ServerConfig) {
|
|
| 136 |
];
|
| 137 |
}
|
| 138 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 139 |
/**
|
| 140 |
* Applies the server-provided defaults to a settings object.
|
| 141 |
*
|
| 142 |
* These are defaults, not overrides: they seed a profile that has never been
|
| 143 |
* saved and are ignored once the user has settings of their own, since a stored
|
| 144 |
* value is a choice even when it matches what the code ships with.
|
|
|
|
|
|
|
|
|
|
| 145 |
*/
|
| 146 |
export function applyServerConfig(
|
| 147 |
settings: typeof defaultSettings,
|
| 148 |
config: ServerConfig,
|
| 149 |
userSettingsWereStored: boolean,
|
|
|
|
| 150 |
): typeof defaultSettings {
|
| 151 |
if (userSettingsWereStored) return settings;
|
| 152 |
|
| 153 |
return {
|
| 154 |
...settings,
|
| 155 |
-
wllamaModelId:
|
|
|
|
|
|
|
| 156 |
inferenceType: config.defaultInferenceType || DEFAULT_INFERENCE_TYPE,
|
| 157 |
};
|
| 158 |
}
|
|
|
|
| 1 |
+
import {
|
| 2 |
+
CITATION_EXAMPLE,
|
| 3 |
+
DEFAULT_SYSTEM_PROMPT,
|
| 4 |
+
} from "@shared/defaultSystemPrompt";
|
| 5 |
import {
|
| 6 |
DEFAULT_INFERENCE_TYPE,
|
| 7 |
+
DEFAULT_WLLAMA_GPU_MODEL_ID,
|
| 8 |
DEFAULT_WLLAMA_MODEL_ID,
|
| 9 |
type ServerConfig,
|
| 10 |
} from "@shared/serverConfig";
|
|
|
|
| 140 |
];
|
| 141 |
}
|
| 142 |
|
| 143 |
+
/** The citation example the default prompt carried until 2026-11-08. */
|
| 144 |
+
const PREVIOUS_CITATION_EXAMPLE =
|
| 145 |
+
"[youtube.com](https://www.youtube.com/watch?v=dQw4w9WgXcQ)";
|
| 146 |
+
|
| 147 |
+
/**
|
| 148 |
+
* Moves a saved prompt that is exactly the previous default onto the current
|
| 149 |
+
* one. Settings store the whole prompt, so without this a profile that saved
|
| 150 |
+
* its settings would keep the old citation example forever. A prompt the user
|
| 151 |
+
* edited is their choice and stays.
|
| 152 |
+
*/
|
| 153 |
+
export function upgradeUntouchedSystemPrompt(
|
| 154 |
+
settings: typeof defaultSettings,
|
| 155 |
+
): typeof defaultSettings {
|
| 156 |
+
const previousDefault = DEFAULT_SYSTEM_PROMPT.replace(
|
| 157 |
+
CITATION_EXAMPLE,
|
| 158 |
+
PREVIOUS_CITATION_EXAMPLE,
|
| 159 |
+
);
|
| 160 |
+
return settings.systemPrompt === previousDefault
|
| 161 |
+
? { ...settings, systemPrompt: DEFAULT_SYSTEM_PROMPT }
|
| 162 |
+
: settings;
|
| 163 |
+
}
|
| 164 |
+
|
| 165 |
/**
|
| 166 |
* Applies the server-provided defaults to a settings object.
|
| 167 |
*
|
| 168 |
* These are defaults, not overrides: they seed a profile that has never been
|
| 169 |
* saved and are ignored once the user has settings of their own, since a stored
|
| 170 |
* value is a choice even when it matches what the code ships with.
|
| 171 |
+
*
|
| 172 |
+
* The Wllama model default depends on `hasWebGpu`, because a model that
|
| 173 |
+
* answers quickly on a GPU can take minutes per answer on the CPU.
|
| 174 |
*/
|
| 175 |
export function applyServerConfig(
|
| 176 |
settings: typeof defaultSettings,
|
| 177 |
config: ServerConfig,
|
| 178 |
userSettingsWereStored: boolean,
|
| 179 |
+
hasWebGpu: boolean,
|
| 180 |
): typeof defaultSettings {
|
| 181 |
if (userSettingsWereStored) return settings;
|
| 182 |
|
| 183 |
return {
|
| 184 |
...settings,
|
| 185 |
+
wllamaModelId: hasWebGpu
|
| 186 |
+
? config.wllamaDefaultGpuModelId || DEFAULT_WLLAMA_GPU_MODEL_ID
|
| 187 |
+
: config.wllamaDefaultModelId || DEFAULT_WLLAMA_MODEL_ID,
|
| 188 |
inferenceType: config.defaultInferenceType || DEFAULT_INFERENCE_TYPE,
|
| 189 |
};
|
| 190 |
}
|
client/modules/textGenerationWithWllama.test.ts
CHANGED
|
@@ -1,7 +1,10 @@
|
|
| 1 |
import { beforeEach, describe, expect, it, vi } from "vitest";
|
|
|
|
| 2 |
import {
|
|
|
|
| 3 |
updateModelLoadingProgress,
|
| 4 |
updateModelSizeInMegabytes,
|
|
|
|
| 5 |
} from "./pubSub";
|
| 6 |
import {
|
| 7 |
composeStreamedText,
|
|
@@ -24,9 +27,18 @@ vi.mock("./pubSub", () => ({
|
|
| 24 |
updateModelLoadingProgress: vi.fn(),
|
| 25 |
updateModelSizeInMegabytes: vi.fn(),
|
| 26 |
updateResponse: vi.fn(),
|
|
|
|
| 27 |
updateTextGenerationState: vi.fn(),
|
| 28 |
}));
|
| 29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
vi.mock("./systemPrompt", () => ({
|
| 31 |
getSystemPrompt: vi.fn().mockReturnValue("system"),
|
| 32 |
}));
|
|
@@ -102,6 +114,29 @@ describe("generateTextWithWllama", () => {
|
|
| 102 |
]);
|
| 103 |
});
|
| 104 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
it("keeps the last known size when the total is not available yet", async () => {
|
| 106 |
mockDownload([{ loaded: 0, total: 0 }]);
|
| 107 |
|
|
|
|
| 1 |
import { beforeEach, describe, expect, it, vi } from "vitest";
|
| 2 |
+
import { addLogEntry } from "./logEntries";
|
| 3 |
import {
|
| 4 |
+
getSettings,
|
| 5 |
updateModelLoadingProgress,
|
| 6 |
updateModelSizeInMegabytes,
|
| 7 |
+
updateSettings,
|
| 8 |
} from "./pubSub";
|
| 9 |
import {
|
| 10 |
composeStreamedText,
|
|
|
|
| 27 |
updateModelLoadingProgress: vi.fn(),
|
| 28 |
updateModelSizeInMegabytes: vi.fn(),
|
| 29 |
updateResponse: vi.fn(),
|
| 30 |
+
updateSettings: vi.fn(),
|
| 31 |
updateTextGenerationState: vi.fn(),
|
| 32 |
}));
|
| 33 |
|
| 34 |
+
vi.mock("./config", () => ({
|
| 35 |
+
getConfig: vi.fn().mockResolvedValue({
|
| 36 |
+
wllamaDefaultModelId: "test-model",
|
| 37 |
+
wllamaDefaultGpuModelId: "test-gpu-model",
|
| 38 |
+
}),
|
| 39 |
+
FALLBACK_CONFIG: {},
|
| 40 |
+
}));
|
| 41 |
+
|
| 42 |
vi.mock("./systemPrompt", () => ({
|
| 43 |
getSystemPrompt: vi.fn().mockReturnValue("system"),
|
| 44 |
}));
|
|
|
|
| 114 |
]);
|
| 115 |
});
|
| 116 |
|
| 117 |
+
it("switches to the default model when the saved one left the catalog", async () => {
|
| 118 |
+
const savedSettings = getSettings();
|
| 119 |
+
vi.mocked(getSettings).mockReturnValue({
|
| 120 |
+
...savedSettings,
|
| 121 |
+
wllamaModelId: "removed-model",
|
| 122 |
+
});
|
| 123 |
+
mockDownload([]);
|
| 124 |
+
|
| 125 |
+
try {
|
| 126 |
+
await generateTextWithWllama();
|
| 127 |
+
} finally {
|
| 128 |
+
vi.mocked(getSettings).mockReturnValue(savedSettings);
|
| 129 |
+
}
|
| 130 |
+
|
| 131 |
+
expect(vi.mocked(initializeWllama).mock.calls[0]?.[0]).toBe("repo");
|
| 132 |
+
expect(updateSettings).toHaveBeenCalledWith(
|
| 133 |
+
expect.objectContaining({ wllamaModelId: "test-model" }),
|
| 134 |
+
);
|
| 135 |
+
expect(addLogEntry).toHaveBeenCalledWith(
|
| 136 |
+
expect.stringContaining("removed-model"),
|
| 137 |
+
);
|
| 138 |
+
});
|
| 139 |
+
|
| 140 |
it("keeps the last known size when the total is not available yet", async () => {
|
| 141 |
mockDownload([{ loaded: 0, total: 0 }]);
|
| 142 |
|
client/modules/textGenerationWithWllama.ts
CHANGED
|
@@ -1,4 +1,5 @@
|
|
| 1 |
import type { Wllama } from "@wllama/wllama";
|
|
|
|
| 2 |
import { addLogEntry } from "./logEntries";
|
| 3 |
import {
|
| 4 |
getSettings,
|
|
@@ -6,6 +7,7 @@ import {
|
|
| 6 |
updateModelLoadingProgress,
|
| 7 |
updateModelSizeInMegabytes,
|
| 8 |
updateResponse,
|
|
|
|
| 9 |
updateTextGenerationState,
|
| 10 |
} from "./pubSub";
|
| 11 |
import {
|
|
@@ -33,8 +35,7 @@ export async function generateTextWithWllama(): Promise<void> {
|
|
| 33 |
try {
|
| 34 |
const response = await generateWithWllama({
|
| 35 |
messages: async () => {
|
| 36 |
-
const
|
| 37 |
-
const modelConfig = wllamaModels[settings.wllamaModelId];
|
| 38 |
return getDefaultChatMessages(
|
| 39 |
await getFormattedSearchResults(
|
| 40 |
modelConfig.shouldIncludeUrlsOnPrompt,
|
|
@@ -180,9 +181,38 @@ async function generateWithWllama({
|
|
| 180 |
}
|
| 181 |
}
|
| 182 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 183 |
async function initializeWllamaInstance(progressCallback?: ProgressCallback) {
|
| 184 |
-
const { initializeWllama
|
| 185 |
-
const model =
|
| 186 |
|
| 187 |
updateModelSizeInMegabytes(model.fileSizeInMegabytes);
|
| 188 |
|
|
|
|
| 1 |
import type { Wllama } from "@wllama/wllama";
|
| 2 |
+
import { FALLBACK_CONFIG, getConfig } from "./config";
|
| 3 |
import { addLogEntry } from "./logEntries";
|
| 4 |
import {
|
| 5 |
getSettings,
|
|
|
|
| 7 |
updateModelLoadingProgress,
|
| 8 |
updateModelSizeInMegabytes,
|
| 9 |
updateResponse,
|
| 10 |
+
updateSettings,
|
| 11 |
updateTextGenerationState,
|
| 12 |
} from "./pubSub";
|
| 13 |
import {
|
|
|
|
| 35 |
try {
|
| 36 |
const response = await generateWithWllama({
|
| 37 |
messages: async () => {
|
| 38 |
+
const modelConfig = await getSelectedWllamaModel();
|
|
|
|
| 39 |
return getDefaultChatMessages(
|
| 40 |
await getFormattedSearchResults(
|
| 41 |
modelConfig.shouldIncludeUrlsOnPrompt,
|
|
|
|
| 181 |
}
|
| 182 |
}
|
| 183 |
|
| 184 |
+
/**
|
| 185 |
+
* The model the settings point at, or the default when that model has left
|
| 186 |
+
* the catalog. Without this, removing a model from the catalog would break
|
| 187 |
+
* every answer for whoever had picked it.
|
| 188 |
+
*/
|
| 189 |
+
async function getSelectedWllamaModel(): Promise<WllamaModel> {
|
| 190 |
+
const { wllamaModelId } = getSettings();
|
| 191 |
+
const model = wllamaModels[wllamaModelId];
|
| 192 |
+
if (model) return model;
|
| 193 |
+
|
| 194 |
+
// Safe without the server: the fallback is only a model to switch to.
|
| 195 |
+
const config = await getConfig().catch(() => FALLBACK_CONFIG);
|
| 196 |
+
const defaultId = isWebGPUAvailable
|
| 197 |
+
? config.wllamaDefaultGpuModelId
|
| 198 |
+
: config.wllamaDefaultModelId;
|
| 199 |
+
const defaultModel = wllamaModels[defaultId];
|
| 200 |
+
if (!defaultModel) {
|
| 201 |
+
throw new Error(
|
| 202 |
+
`Neither the selected model "${wllamaModelId}" nor the default "${defaultId}" is in the catalog`,
|
| 203 |
+
);
|
| 204 |
+
}
|
| 205 |
+
|
| 206 |
+
addLogEntry(
|
| 207 |
+
`Model "${wllamaModelId}" is no longer available; switched to "${defaultId}"`,
|
| 208 |
+
);
|
| 209 |
+
updateSettings({ ...getSettings(), wllamaModelId: defaultId });
|
| 210 |
+
return defaultModel;
|
| 211 |
+
}
|
| 212 |
+
|
| 213 |
async function initializeWllamaInstance(progressCallback?: ProgressCallback) {
|
| 214 |
+
const { initializeWllama } = await import("./wllama");
|
| 215 |
+
const model = await getSelectedWllamaModel();
|
| 216 |
|
| 217 |
updateModelSizeInMegabytes(model.fileSizeInMegabytes);
|
| 218 |
|
client/modules/wllama.test.ts
CHANGED
|
@@ -1,5 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import { describe, expect, it, vi } from "vitest";
|
| 2 |
-
import { wllamaModels } from "./wllama";
|
| 3 |
|
| 4 |
vi.mock("./logEntries", () => ({ addLogEntry: vi.fn() }));
|
| 5 |
vi.mock("./textGenerationUtilities", () => ({ defaultContextSize: 4096 }));
|
|
@@ -46,6 +51,17 @@ describe("wllamaModels catalog", () => {
|
|
| 46 |
}
|
| 47 |
});
|
| 48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
it("keeps reasoning budgets small enough to leave room for the answer", () => {
|
| 50 |
// The budget forces the final answer once the chain of thought runs
|
| 51 |
// this long. It must stay far below the 4096-token context, or the
|
|
@@ -58,9 +74,46 @@ describe("wllamaModels catalog", () => {
|
|
| 58 |
}
|
| 59 |
});
|
| 60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
it("uses the shared default context size for every model", () => {
|
| 62 |
for (const [id, model] of models) {
|
| 63 |
expect(model.contextSize, id).toBe(4096);
|
| 64 |
}
|
| 65 |
});
|
| 66 |
});
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import {
|
| 2 |
+
DEFAULT_WLLAMA_GPU_MODEL_ID,
|
| 3 |
+
DEFAULT_WLLAMA_MODEL_ID,
|
| 4 |
+
} from "@shared/serverConfig";
|
| 5 |
+
import type { CacheEntry, CacheManager } from "@wllama/wllama";
|
| 6 |
import { describe, expect, it, vi } from "vitest";
|
| 7 |
+
import { deleteIncompleteCachedModels, wllamaModels } from "./wllama";
|
| 8 |
|
| 9 |
vi.mock("./logEntries", () => ({ addLogEntry: vi.fn() }));
|
| 10 |
vi.mock("./textGenerationUtilities", () => ({ defaultContextSize: 4096 }));
|
|
|
|
| 51 |
}
|
| 52 |
});
|
| 53 |
|
| 54 |
+
it("asks every thinking model for low reasoning effort", () => {
|
| 55 |
+
// A chain of thought cut off by the budget degrades the answer, so the
|
| 56 |
+
// budget is only a backstop; templates that read `reasoning_effort`
|
| 57 |
+
// must be steered to think less in the first place.
|
| 58 |
+
for (const [id, model] of models) {
|
| 59 |
+
if (model.reasoningBudgetTokens !== undefined) {
|
| 60 |
+
expect(model.chatTemplateKwargs?.reasoning_effort, id).toBe("low");
|
| 61 |
+
}
|
| 62 |
+
}
|
| 63 |
+
});
|
| 64 |
+
|
| 65 |
it("keeps reasoning budgets small enough to leave room for the answer", () => {
|
| 66 |
// The budget forces the final answer once the chain of thought runs
|
| 67 |
// this long. It must stay far below the 4096-token context, or the
|
|
|
|
| 74 |
}
|
| 75 |
});
|
| 76 |
|
| 77 |
+
it("includes both default models", () => {
|
| 78 |
+
expect(wllamaModels).toHaveProperty([DEFAULT_WLLAMA_MODEL_ID]);
|
| 79 |
+
expect(wllamaModels).toHaveProperty([DEFAULT_WLLAMA_GPU_MODEL_ID]);
|
| 80 |
+
});
|
| 81 |
+
|
| 82 |
it("uses the shared default context size for every model", () => {
|
| 83 |
for (const [id, model] of models) {
|
| 84 |
expect(model.contextSize, id).toBe(4096);
|
| 85 |
}
|
| 86 |
});
|
| 87 |
});
|
| 88 |
+
|
| 89 |
+
describe("deleteIncompleteCachedModels", () => {
|
| 90 |
+
const shard = (repo: string, index: number, total: number) =>
|
| 91 |
+
`https://huggingface.co/${repo}/resolve/main/model-0000${index}-of-0000${total}.gguf`;
|
| 92 |
+
const entry = (url: string): CacheEntry => ({
|
| 93 |
+
name: url,
|
| 94 |
+
size: 1,
|
| 95 |
+
metadata: { originalURL: url, etag: "", originalSize: 1 },
|
| 96 |
+
});
|
| 97 |
+
|
| 98 |
+
it("deletes only the shards of a model whose download did not finish", async () => {
|
| 99 |
+
let files = [
|
| 100 |
+
entry(shard("complete", 1, 2)),
|
| 101 |
+
entry(shard("complete", 2, 2)),
|
| 102 |
+
entry(shard("interrupted", 1, 3)),
|
| 103 |
+
entry(shard("interrupted", 3, 3)),
|
| 104 |
+
];
|
| 105 |
+
const cacheManager = {
|
| 106 |
+
list: async () => files,
|
| 107 |
+
deleteMany: async (predicate: (file: CacheEntry) => boolean) => {
|
| 108 |
+
files = files.filter((file) => !predicate(file));
|
| 109 |
+
},
|
| 110 |
+
} as unknown as CacheManager;
|
| 111 |
+
|
| 112 |
+
await deleteIncompleteCachedModels(cacheManager);
|
| 113 |
+
|
| 114 |
+
expect(files.map((file) => file.metadata.originalURL)).toEqual([
|
| 115 |
+
shard("complete", 1, 2),
|
| 116 |
+
shard("complete", 2, 2),
|
| 117 |
+
]);
|
| 118 |
+
});
|
| 119 |
+
});
|
client/modules/wllama.ts
CHANGED
|
@@ -1,5 +1,7 @@
|
|
| 1 |
import {
|
|
|
|
| 2 |
type LoadModelParams,
|
|
|
|
| 3 |
type SamplingParams,
|
| 4 |
Wllama,
|
| 5 |
type WllamaConfig,
|
|
@@ -33,6 +35,35 @@ async function createWllamaInstance(config?: WllamaConfig): Promise<Wllama> {
|
|
| 33 |
}
|
| 34 |
}
|
| 35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
/**
|
| 37 |
* Initializes Wllama with a specific model from HuggingFace.
|
| 38 |
* Performs a single-threaded warmup run before loading with the full config.
|
|
@@ -47,6 +78,8 @@ export async function initializeWllama(
|
|
| 47 |
try {
|
| 48 |
const wllama = await createWllamaInstance(config?.wllama);
|
| 49 |
|
|
|
|
|
|
|
| 50 |
await wllama.loadModelFromHF(
|
| 51 |
{ repo: hfRepoId, file: hfFilePath },
|
| 52 |
{
|
|
@@ -161,17 +194,14 @@ const createDefaultModelConfig = (): Omit<
|
|
| 161 |
flash_attn: true,
|
| 162 |
contextSize: defaultContextSize,
|
| 163 |
shouldIncludeUrlsOnPrompt: true,
|
|
|
|
|
|
|
|
|
|
|
|
|
| 164 |
getSampling: () => ({}),
|
| 165 |
});
|
| 166 |
|
| 167 |
export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
| 168 |
-
"smollm2-135m": {
|
| 169 |
-
...createDefaultModelConfig(),
|
| 170 |
-
label: "SmolLM 2 135M",
|
| 171 |
-
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-SmolLM2-135M-Instruct",
|
| 172 |
-
hfFilePath: "model.shard-00001-of-00004.gguf",
|
| 173 |
-
fileSizeInMegabytes: 102,
|
| 174 |
-
},
|
| 175 |
"lfm2.5-230m": {
|
| 176 |
...createDefaultModelConfig(),
|
| 177 |
label: "LFM2.5 230M",
|
|
@@ -179,19 +209,13 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 179 |
hfFilePath: "LFM2.5-230M-Q4_K_XL-00001-of-00003.gguf",
|
| 180 |
fileSizeInMegabytes: 160,
|
| 181 |
},
|
| 182 |
-
"granite-4.0-350m": {
|
| 183 |
-
...createDefaultModelConfig(),
|
| 184 |
-
label: "Granite 4.0 350M",
|
| 185 |
-
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-granite-4.0-350m",
|
| 186 |
-
hfFilePath: "granite-4.0-350m-Q4_K_S-00001-of-00003.gguf",
|
| 187 |
-
fileSizeInMegabytes: 229,
|
| 188 |
-
},
|
| 189 |
"littlelamb-290m": {
|
| 190 |
...createDefaultModelConfig(),
|
| 191 |
label: "LittleLamb 290M",
|
| 192 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-LittleLamb",
|
| 193 |
hfFilePath: "LittleLamb-i1-Q4_K_S-00001-of-00003.gguf",
|
| 194 |
fileSizeInMegabytes: 231,
|
|
|
|
| 195 |
},
|
| 196 |
"gemma-3-270m": {
|
| 197 |
...createDefaultModelConfig(),
|
|
@@ -220,6 +244,15 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 220 |
hfRepoId: "Felladrin/gguf-sharded-UD-Q4_K_XL-Qwen3-0.6B",
|
| 221 |
hfFilePath: "model.shard-00001-of-00004.gguf",
|
| 222 |
fileSizeInMegabytes: 406,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 223 |
},
|
| 224 |
"lfm2.5-1.2b-instruct": {
|
| 225 |
...createDefaultModelConfig(),
|
|
@@ -249,13 +282,6 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 249 |
hfFilePath: "model.shard-00001-of-00003.gguf",
|
| 250 |
fileSizeInMegabytes: 781,
|
| 251 |
},
|
| 252 |
-
"pints-1.5b": {
|
| 253 |
-
...createDefaultModelConfig(),
|
| 254 |
-
label: "Pints 1.5B",
|
| 255 |
-
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-1.5-Pints-16K-v0.1",
|
| 256 |
-
hfFilePath: "model.shard-00001-of-00018.gguf",
|
| 257 |
-
fileSizeInMegabytes: 905,
|
| 258 |
-
},
|
| 259 |
"olmo-2-0425-1b": {
|
| 260 |
...createDefaultModelConfig(),
|
| 261 |
label: "OLMo 2 1B",
|
|
@@ -263,20 +289,6 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 263 |
hfFilePath: "model.shard-00001-of-00006.gguf",
|
| 264 |
fileSizeInMegabytes: 967,
|
| 265 |
},
|
| 266 |
-
"granite-4.0-1b": {
|
| 267 |
-
...createDefaultModelConfig(),
|
| 268 |
-
label: "Granite 4.0 1B",
|
| 269 |
-
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-granite-4.0-1b",
|
| 270 |
-
hfFilePath: "granite-4.0-1b-Q4_K_S-00001-of-00006.gguf",
|
| 271 |
-
fileSizeInMegabytes: 981,
|
| 272 |
-
},
|
| 273 |
-
"stablelm-2-zephyr-1.6b": {
|
| 274 |
-
...createDefaultModelConfig(),
|
| 275 |
-
label: "StableLM 2 Zephyr 1.6B",
|
| 276 |
-
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-stablelm-2-zephyr-1.6b",
|
| 277 |
-
hfFilePath: "model.shard-00001-of-00006.gguf",
|
| 278 |
-
fileSizeInMegabytes: 989,
|
| 279 |
-
},
|
| 280 |
"smollm2-1.7b": {
|
| 281 |
...createDefaultModelConfig(),
|
| 282 |
label: "SmolLM 2 1.7B",
|
|
@@ -304,6 +316,7 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 304 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-DeepSeek-R1-Distill-Qwen-1.5B",
|
| 305 |
hfFilePath: "model.shard-00001-of-00006.gguf",
|
| 306 |
fileSizeInMegabytes: 1070,
|
|
|
|
| 307 |
},
|
| 308 |
"internlm-2.5-1.8b": {
|
| 309 |
...createDefaultModelConfig(),
|
|
@@ -318,6 +331,7 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 318 |
hfRepoId: "Felladrin/gguf-sharded-UD-Q4_K_XL-Qwen3-1.7B",
|
| 319 |
hfFilePath: "model.shard-00001-of-00005.gguf",
|
| 320 |
fileSizeInMegabytes: 1140,
|
|
|
|
| 321 |
},
|
| 322 |
"granite-3.3-2b": {
|
| 323 |
...createDefaultModelConfig(),
|
|
@@ -353,7 +367,7 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 353 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_M-Spark-X2.6-1.7B",
|
| 354 |
hfFilePath: "Spark-X2.6-1.7B-Q4_K_M-00001-of-00004.gguf",
|
| 355 |
fileSizeInMegabytes: 1769,
|
| 356 |
-
reasoningBudgetTokens:
|
| 357 |
},
|
| 358 |
"smollm3-3b": {
|
| 359 |
...createDefaultModelConfig(),
|
|
@@ -389,7 +403,7 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 389 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-granite-4.2-3b",
|
| 390 |
hfFilePath: "granite-4.2-3b-Q4_K_S-00001-of-00005.gguf",
|
| 391 |
fileSizeInMegabytes: 2143,
|
| 392 |
-
reasoningBudgetTokens:
|
| 393 |
},
|
| 394 |
"phi-3.5-mini-3.8b": {
|
| 395 |
...createDefaultModelConfig(),
|
|
@@ -411,6 +425,7 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 411 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-Polaris-4B-Preview",
|
| 412 |
hfFilePath: "model.shard-00001-of-00008.gguf",
|
| 413 |
fileSizeInMegabytes: 2380,
|
|
|
|
| 414 |
},
|
| 415 |
"qwen-3.5-4b": {
|
| 416 |
...createDefaultModelConfig(),
|
|
@@ -418,7 +433,7 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 418 |
hfRepoId: "Felladrin/gguf-sharded-UD-Q3_K_XL-Qwen3.5-4B",
|
| 419 |
hfFilePath: "Qwen3.5-4B-UD-Q3_K_XL-00001-of-00005.gguf",
|
| 420 |
fileSizeInMegabytes: 2436,
|
| 421 |
-
reasoningBudgetTokens:
|
| 422 |
},
|
| 423 |
"gemma-4-e2b": {
|
| 424 |
...createDefaultModelConfig(),
|
|
@@ -426,7 +441,7 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 426 |
hfRepoId: "Felladrin/gguf-sharded-Q3_K_S-gemma-4-e2b-it",
|
| 427 |
hfFilePath: "gemma-4-e2b-it-Q3_K_S-00001-of-00005.gguf",
|
| 428 |
fileSizeInMegabytes: 2446,
|
| 429 |
-
reasoningBudgetTokens:
|
| 430 |
},
|
| 431 |
"phi-4-mini-reasoning": {
|
| 432 |
...createDefaultModelConfig(),
|
|
@@ -441,6 +456,6 @@ export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
| 441 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_M-Spark-X2.6-4B",
|
| 442 |
hfFilePath: "Spark-X2.6-4B-Q4_K_M-00001-of-00007.gguf",
|
| 443 |
fileSizeInMegabytes: 3543,
|
| 444 |
-
reasoningBudgetTokens:
|
| 445 |
},
|
| 446 |
};
|
|
|
|
| 1 |
import {
|
| 2 |
+
type CacheManager,
|
| 3 |
type LoadModelParams,
|
| 4 |
+
ModelManager,
|
| 5 |
type SamplingParams,
|
| 6 |
Wllama,
|
| 7 |
type WllamaConfig,
|
|
|
|
| 35 |
}
|
| 36 |
}
|
| 37 |
|
| 38 |
+
/**
|
| 39 |
+
* Deletes the cached shards of every model whose download did not finish.
|
| 40 |
+
* wllama 3.8.1 builds each cached model before checking it, and a missing
|
| 41 |
+
* shard makes that throw, so one interrupted download would otherwise stop
|
| 42 |
+
* every model from loading until the whole cache is cleared.
|
| 43 |
+
*/
|
| 44 |
+
export async function deleteIncompleteCachedModels(cacheManager: CacheManager) {
|
| 45 |
+
const cachedUrls = new Set(
|
| 46 |
+
(await cacheManager.list())
|
| 47 |
+
.map((file) => file.metadata.originalURL)
|
| 48 |
+
// wllama skips files saved without one, so they cannot block a load.
|
| 49 |
+
.filter(Boolean),
|
| 50 |
+
);
|
| 51 |
+
const incompleteUrls = new Set(
|
| 52 |
+
[...cachedUrls]
|
| 53 |
+
.map((url) => ModelManager.parseModelUrl(url))
|
| 54 |
+
.filter((shards) => shards.some((shard) => !cachedUrls.has(shard)))
|
| 55 |
+
.flat(),
|
| 56 |
+
);
|
| 57 |
+
if (incompleteUrls.size === 0) return;
|
| 58 |
+
|
| 59 |
+
addLogEntry(
|
| 60 |
+
`Deleting ${incompleteUrls.size} cached file(s) of models whose download did not finish`,
|
| 61 |
+
);
|
| 62 |
+
await cacheManager.deleteMany((file) =>
|
| 63 |
+
incompleteUrls.has(file.metadata.originalURL),
|
| 64 |
+
);
|
| 65 |
+
}
|
| 66 |
+
|
| 67 |
/**
|
| 68 |
* Initializes Wllama with a specific model from HuggingFace.
|
| 69 |
* Performs a single-threaded warmup run before loading with the full config.
|
|
|
|
| 78 |
try {
|
| 79 |
const wllama = await createWllamaInstance(config?.wllama);
|
| 80 |
|
| 81 |
+
await deleteIncompleteCachedModels(wllama.cacheManager);
|
| 82 |
+
|
| 83 |
await wllama.loadModelFromHF(
|
| 84 |
{ repo: hfRepoId, file: hfFilePath },
|
| 85 |
{
|
|
|
|
| 194 |
flash_attn: true,
|
| 195 |
contextSize: defaultContextSize,
|
| 196 |
shouldIncludeUrlsOnPrompt: true,
|
| 197 |
+
// Templates that read it (Granite 4.2, Spark X2.6) think less, so they
|
| 198 |
+
// reach the answer before `reasoningBudgetTokens` cuts them off; the others
|
| 199 |
+
// ignore it.
|
| 200 |
+
chatTemplateKwargs: { reasoning_effort: "low" },
|
| 201 |
getSampling: () => ({}),
|
| 202 |
});
|
| 203 |
|
| 204 |
export const wllamaModels: Readonly<Record<string, WllamaModel>> = {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 205 |
"lfm2.5-230m": {
|
| 206 |
...createDefaultModelConfig(),
|
| 207 |
label: "LFM2.5 230M",
|
|
|
|
| 209 |
hfFilePath: "LFM2.5-230M-Q4_K_XL-00001-of-00003.gguf",
|
| 210 |
fileSizeInMegabytes: 160,
|
| 211 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 212 |
"littlelamb-290m": {
|
| 213 |
...createDefaultModelConfig(),
|
| 214 |
label: "LittleLamb 290M",
|
| 215 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-LittleLamb",
|
| 216 |
hfFilePath: "LittleLamb-i1-Q4_K_S-00001-of-00003.gguf",
|
| 217 |
fileSizeInMegabytes: 231,
|
| 218 |
+
reasoningBudgetTokens: 1024,
|
| 219 |
},
|
| 220 |
"gemma-3-270m": {
|
| 221 |
...createDefaultModelConfig(),
|
|
|
|
| 244 |
hfRepoId: "Felladrin/gguf-sharded-UD-Q4_K_XL-Qwen3-0.6B",
|
| 245 |
hfFilePath: "model.shard-00001-of-00004.gguf",
|
| 246 |
fileSizeInMegabytes: 406,
|
| 247 |
+
reasoningBudgetTokens: 1024,
|
| 248 |
+
},
|
| 249 |
+
"qwen-3.5-0.8b": {
|
| 250 |
+
...createDefaultModelConfig(),
|
| 251 |
+
label: "Qwen 3.5 0.8B",
|
| 252 |
+
hfRepoId: "Felladrin/gguf-sharded-Q4_K_M-Qwen3.5-0.8B",
|
| 253 |
+
hfFilePath: "Qwen3.5-0.8B-Q4_K_M-00001-of-00003.gguf",
|
| 254 |
+
fileSizeInMegabytes: 533,
|
| 255 |
+
reasoningBudgetTokens: 1024,
|
| 256 |
},
|
| 257 |
"lfm2.5-1.2b-instruct": {
|
| 258 |
...createDefaultModelConfig(),
|
|
|
|
| 282 |
hfFilePath: "model.shard-00001-of-00003.gguf",
|
| 283 |
fileSizeInMegabytes: 781,
|
| 284 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 285 |
"olmo-2-0425-1b": {
|
| 286 |
...createDefaultModelConfig(),
|
| 287 |
label: "OLMo 2 1B",
|
|
|
|
| 289 |
hfFilePath: "model.shard-00001-of-00006.gguf",
|
| 290 |
fileSizeInMegabytes: 967,
|
| 291 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 292 |
"smollm2-1.7b": {
|
| 293 |
...createDefaultModelConfig(),
|
| 294 |
label: "SmolLM 2 1.7B",
|
|
|
|
| 316 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-DeepSeek-R1-Distill-Qwen-1.5B",
|
| 317 |
hfFilePath: "model.shard-00001-of-00006.gguf",
|
| 318 |
fileSizeInMegabytes: 1070,
|
| 319 |
+
reasoningBudgetTokens: 1024,
|
| 320 |
},
|
| 321 |
"internlm-2.5-1.8b": {
|
| 322 |
...createDefaultModelConfig(),
|
|
|
|
| 331 |
hfRepoId: "Felladrin/gguf-sharded-UD-Q4_K_XL-Qwen3-1.7B",
|
| 332 |
hfFilePath: "model.shard-00001-of-00005.gguf",
|
| 333 |
fileSizeInMegabytes: 1140,
|
| 334 |
+
reasoningBudgetTokens: 1024,
|
| 335 |
},
|
| 336 |
"granite-3.3-2b": {
|
| 337 |
...createDefaultModelConfig(),
|
|
|
|
| 367 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_M-Spark-X2.6-1.7B",
|
| 368 |
hfFilePath: "Spark-X2.6-1.7B-Q4_K_M-00001-of-00004.gguf",
|
| 369 |
fileSizeInMegabytes: 1769,
|
| 370 |
+
reasoningBudgetTokens: 1024,
|
| 371 |
},
|
| 372 |
"smollm3-3b": {
|
| 373 |
...createDefaultModelConfig(),
|
|
|
|
| 403 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-granite-4.2-3b",
|
| 404 |
hfFilePath: "granite-4.2-3b-Q4_K_S-00001-of-00005.gguf",
|
| 405 |
fileSizeInMegabytes: 2143,
|
| 406 |
+
reasoningBudgetTokens: 1024,
|
| 407 |
},
|
| 408 |
"phi-3.5-mini-3.8b": {
|
| 409 |
...createDefaultModelConfig(),
|
|
|
|
| 425 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_S-Polaris-4B-Preview",
|
| 426 |
hfFilePath: "model.shard-00001-of-00008.gguf",
|
| 427 |
fileSizeInMegabytes: 2380,
|
| 428 |
+
reasoningBudgetTokens: 1024,
|
| 429 |
},
|
| 430 |
"qwen-3.5-4b": {
|
| 431 |
...createDefaultModelConfig(),
|
|
|
|
| 433 |
hfRepoId: "Felladrin/gguf-sharded-UD-Q3_K_XL-Qwen3.5-4B",
|
| 434 |
hfFilePath: "Qwen3.5-4B-UD-Q3_K_XL-00001-of-00005.gguf",
|
| 435 |
fileSizeInMegabytes: 2436,
|
| 436 |
+
reasoningBudgetTokens: 1024,
|
| 437 |
},
|
| 438 |
"gemma-4-e2b": {
|
| 439 |
...createDefaultModelConfig(),
|
|
|
|
| 441 |
hfRepoId: "Felladrin/gguf-sharded-Q3_K_S-gemma-4-e2b-it",
|
| 442 |
hfFilePath: "gemma-4-e2b-it-Q3_K_S-00001-of-00005.gguf",
|
| 443 |
fileSizeInMegabytes: 2446,
|
| 444 |
+
reasoningBudgetTokens: 1024,
|
| 445 |
},
|
| 446 |
"phi-4-mini-reasoning": {
|
| 447 |
...createDefaultModelConfig(),
|
|
|
|
| 456 |
hfRepoId: "Felladrin/gguf-sharded-Q4_K_M-Spark-X2.6-4B",
|
| 457 |
hfFilePath: "Spark-X2.6-4B-Q4_K_M-00001-of-00007.gguf",
|
| 458 |
fileSizeInMegabytes: 3543,
|
| 459 |
+
reasoningBudgetTokens: 1024,
|
| 460 |
},
|
| 461 |
};
|
docs/ai-integration.md
CHANGED
|
@@ -20,11 +20,13 @@ Runs AI models entirely in the browser using WebAssembly or WebGPU. No data leav
|
|
| 20 |
Uses `@wllama/wllama` for browser-based inference. Automatically uses WebGPU when available, falling back to CPU via WebAssembly.
|
| 21 |
|
| 22 |
**Requirements:**
|
|
|
|
| 23 |
- Any modern browser
|
| 24 |
- ~300MB-2GB free RAM
|
| 25 |
- WebGPU optional (Chrome 113+, Edge 113+) - enables GPU acceleration
|
| 26 |
|
| 27 |
**How It Works:**
|
|
|
|
| 28 |
1. Checks WebGPU availability via `"gpu" in navigator`
|
| 29 |
2. Downloads model from HuggingFace (GGUF format, cached in the browser's Origin Private File System)
|
| 30 |
3. Loads model with `n_gpu_layers: 99999` if WebGPU is present, else `0` for CPU-only
|
|
@@ -32,22 +34,25 @@ Uses `@wllama/wllama` for browser-based inference. Automatically uses WebGPU whe
|
|
| 32 |
5. Streams tokens via the OAI-compatible `createChatCompletion` API
|
| 33 |
|
| 34 |
**Pre-configured Models:**
|
| 35 |
-
|
| 36 |
|
| 37 |
Two constraints decide what can join that list, both measured against the shipped wasm build:
|
|
|
|
| 38 |
- **Every model file must be a shard.** Chromium's OPFS storage silently truncates files past 2000 MiB, so a monolithic GGUF above that size arrives corrupted in the browser. Shards also keep progressive download working.
|
| 39 |
- **The wasm linear memory ceiling is 4 GiB.** Models at ~3.5 GB and above OOM during buffer reserve, so the practical catalog ceiling is around 3.7 GB of weights.
|
| 40 |
|
| 41 |
-
Models whose chat template thinks by default (Granite 4.2, Gemma 4, Spark X2.6, Qwen 3.5) keep thinking: wllama splits their chain of thought into a `reasoning_content` channel, and MiniSearch wraps it in the configured reasoning markers so the thinking UI shows it while the answer follows. Thinking spends the same token budget as the answer, so these models are slower. To keep a long-winded model from thinking its way out of the context window without ever answering, thinking models ship with `reasoningBudgetTokens` (wllama's `reasoning_budget_tokens`, currently
|
| 42 |
|
| 43 |
`node scripts/wllama-catalog-check.cjs` HEADs every shard of every catalog model on HuggingFace and verifies the claimed sizes; run it after adding or re-sharding a model.
|
| 44 |
|
| 45 |
**Configuration:**
|
|
|
|
| 46 |
- Settings โ Inference Type: `Browser`
|
| 47 |
- Settings โ Wllama Model: Select from dropdown
|
| 48 |
- Settings โ CPU threads: Relevant only when WebGPU is unavailable
|
| 49 |
|
| 50 |
**WebGPU Detection:**
|
|
|
|
| 51 |
```typescript
|
| 52 |
// client/modules/webGpu.ts
|
| 53 |
/** Whether the browser supports the WebGPU API. */
|
|
@@ -55,6 +60,7 @@ export const isWebGPUAvailable = "gpu" in navigator;
|
|
| 55 |
```
|
| 56 |
|
| 57 |
**Limitations:**
|
|
|
|
| 58 |
- First load requires model download (progressive via sharded files)
|
| 59 |
- Limited to smaller models (~4B params max due to browser memory)
|
| 60 |
- CPU mode is 2-5x slower than WebGPU
|
|
@@ -64,24 +70,28 @@ export const isWebGPUAvailable = "gpu" in navigator;
|
|
| 64 |
Uses OpenAI's API or any OpenAI-compatible service.
|
| 65 |
|
| 66 |
**Setup:**
|
|
|
|
| 67 |
1. Get API key from OpenAI or compatible provider
|
| 68 |
2. Settings โ Inference Type: `OpenAI`
|
| 69 |
3. Settings โ OpenAI API Key: Enter key
|
| 70 |
4. Settings โ OpenAI Model: Select or enter model ID
|
| 71 |
|
| 72 |
**Supported Providers:**
|
|
|
|
| 73 |
- OpenAI (gpt-4, gpt-3.5-turbo)
|
| 74 |
- Anthropic (via OpenAI-compatible endpoint)
|
| 75 |
- Google (Gemini via OpenAI-compatible endpoint)
|
| 76 |
- Any custom provider with OpenAI-compatible API
|
| 77 |
|
| 78 |
**Features:**
|
|
|
|
| 79 |
- Streaming responses, built on the Vercel AI SDK (`ai` + `@ai-sdk/openai-compatible`)
|
| 80 |
- Auto model selection (if blank)
|
| 81 |
- On a request error, automatically retries up to 5 times, switching to a different model fetched from the provider's `/models` endpoint each time (100ms ร attempt backoff between retries)
|
| 82 |
- Reasoning content support: reasoning/thinking output is wrapped between configurable `reasoningStartMarker`/`reasoningEndMarker` settings (default `<think>` / `</think>`)
|
| 83 |
|
| 84 |
**Configuration:**
|
|
|
|
| 85 |
```typescript
|
| 86 |
{
|
| 87 |
inferenceType: 'openai',
|
|
@@ -92,6 +102,7 @@ Uses OpenAI's API or any OpenAI-compatible service.
|
|
| 92 |
```
|
| 93 |
|
| 94 |
**Privacy Considerations:**
|
|
|
|
| 95 |
- Search queries and results sent to OpenAI
|
| 96 |
- Not suitable for sensitive data
|
| 97 |
- Consider internal API for private data
|
|
@@ -101,11 +112,13 @@ Uses OpenAI's API or any OpenAI-compatible service.
|
|
| 101 |
Uses aihorde.net, a distributed volunteer GPU network.
|
| 102 |
|
| 103 |
**Setup:**
|
|
|
|
| 104 |
1. Settings โ Inference Type: `AI Horde`
|
| 105 |
2. (Optional) Settings โ AI Horde API Key: Get from aihorde.net
|
| 106 |
3. Settings โ AI Horde Model: Select preferred model
|
| 107 |
|
| 108 |
**How It Works:**
|
|
|
|
| 109 |
1. Request sent to AI Horde API
|
| 110 |
2. Distributed to volunteer workers
|
| 111 |
3. Multiple workers may process in parallel
|
|
@@ -113,12 +126,14 @@ Uses aihorde.net, a distributed volunteer GPU network.
|
|
| 113 |
5. Results streamed back
|
| 114 |
|
| 115 |
**Features:**
|
|
|
|
| 116 |
- Free to use (anonymous or authenticated)
|
| 117 |
- Kudos-based priority system
|
| 118 |
- Large model selection (70B+ params available)
|
| 119 |
- No API key required (but recommended for priority)
|
| 120 |
|
| 121 |
**Configuration:**
|
|
|
|
| 122 |
```typescript
|
| 123 |
{
|
| 124 |
inferenceType: 'horde',
|
|
@@ -128,6 +143,7 @@ Uses aihorde.net, a distributed volunteer GPU network.
|
|
| 128 |
```
|
| 129 |
|
| 130 |
**Limitations:**
|
|
|
|
| 131 |
- Variable latency (depends on worker availability)
|
| 132 |
- Quality varies by worker
|
| 133 |
- May queue during high demand
|
|
@@ -138,11 +154,13 @@ Uses aihorde.net, a distributed volunteer GPU network.
|
|
| 138 |
Self-hosted OpenAI-compatible API for teams and compliance.
|
| 139 |
|
| 140 |
**Setup:**
|
|
|
|
| 141 |
1. Host an OpenAI-compatible API (e.g., vLLM, llama.cpp server, Ollama with OpenAI compat)
|
| 142 |
2. Configure environment variables (see `docs/configuration.md`)
|
| 143 |
3. Settings โ Inference Type: `Internal`
|
| 144 |
|
| 145 |
**Environment Variables:**
|
|
|
|
| 146 |
```bash
|
| 147 |
INTERNAL_OPENAI_COMPATIBLE_API_BASE_URL="https://llm.company.com/v1"
|
| 148 |
INTERNAL_OPENAI_COMPATIBLE_API_KEY="sk-internal-xxx"
|
|
@@ -152,12 +170,14 @@ INTERNAL_OPENAI_COMPATIBLE_API_NAME="Company LLM"
|
|
| 152 |
|
| 153 |
**Server-Side Proxy:**
|
| 154 |
The internal API uses a server-side proxy to:
|
|
|
|
| 155 |
- Hide API keys from client
|
| 156 |
- Add request logging/auditing
|
| 157 |
- Apply rate limiting
|
| 158 |
- Enable token-based authentication
|
| 159 |
|
| 160 |
**Endpoint:**
|
|
|
|
| 161 |
```
|
| 162 |
POST /inference?token=<argon2-hashed-search-token>
|
| 163 |
Content-Type: application/json
|
|
@@ -179,12 +199,14 @@ the first token, how many models were tried, and which model ids failed. See
|
|
| 179 |
the `/status` section of `docs/overview.md` for the field reference.
|
| 180 |
|
| 181 |
**Features:**
|
|
|
|
| 182 |
- Private data stays in your infrastructure
|
| 183 |
- Custom model selection
|
| 184 |
- Server-side logging
|
| 185 |
- Compatible with any OpenAI-compatible API
|
| 186 |
|
| 187 |
**Recommended Self-Hosted Options:**
|
|
|
|
| 188 |
- **vLLM**: High-performance, production-ready
|
| 189 |
- **llama.cpp server**: Single binary, easy setup
|
| 190 |
- **Ollama**: Simple, Docker-friendly
|
|
@@ -257,6 +279,7 @@ For non-browser inference types (`openai`, `internal`, `horde`), the state skips
|
|
| 257 |
### Response Throttling
|
| 258 |
|
| 259 |
Two PubSub update functions are throttled to 12 updates per second to ensure UI performance during high-frequency streaming:
|
|
|
|
| 260 |
- `updateResponse` → `responsePubSub`
|
| 261 |
- `updateReasoningContent` → `reasoningContentPubSub`
|
| 262 |
|
|
@@ -283,6 +306,7 @@ For browser inference, the system first loads the model from HuggingFace with a
|
|
| 283 |
- **Available for Context:** 75% (~3072 tokens)
|
| 284 |
|
| 285 |
**Allocation Priority:**
|
|
|
|
| 286 |
1. System prompt (with search results)
|
| 287 |
2. Conversation summary (if exists)
|
| 288 |
3. Recent chat messages (newest first)
|
|
@@ -291,6 +315,7 @@ For browser inference, the system first loads the model from HuggingFace with a
|
|
| 291 |
### Rolling Summaries
|
| 292 |
|
| 293 |
When conversation exceeds token budget:
|
|
|
|
| 294 |
1. **Detect Overflow:** Current tokens > 3072
|
| 295 |
2. **Generate Summary:** Call LLM with 800-token limit
|
| 296 |
3. **Store Summary:** Save in conversationSummaryPubSub
|
|
@@ -298,6 +323,7 @@ When conversation exceeds token budget:
|
|
| 298 |
5. **Continue:** Use summary + remaining messages for context
|
| 299 |
|
| 300 |
**Summary Prompt:**
|
|
|
|
| 301 |
```
|
| 302 |
Summarize this conversation in 3-5 sentences, preserving key facts
|
| 303 |
and user intent. Be concise but informative.
|
|
@@ -364,6 +390,7 @@ Wllama handles both WebGPU and CPU paths internally. If the model fails to load
|
|
| 364 |
### State Recovery
|
| 365 |
|
| 366 |
If generation fails mid-stream:
|
|
|
|
| 367 |
1. Set `textGenerationState` to `failed`
|
| 368 |
2. Preserve partial response in `responsePubSub`
|
| 369 |
3. Allow user to retry or modify query
|
|
@@ -383,6 +410,7 @@ Wllama model shards are cached in the browser's Origin Private File System after
|
|
| 383 |
### Progressive Model Loading
|
| 384 |
|
| 385 |
Wllama models are sharded (split into chunks):
|
|
|
|
| 386 |
1. Download metadata first (small, fast)
|
| 387 |
2. Download required shards progressively
|
| 388 |
3. Start inference when first shards available
|
|
|
|
| 20 |
Uses `@wllama/wllama` for browser-based inference. Automatically uses WebGPU when available, falling back to CPU via WebAssembly.
|
| 21 |
|
| 22 |
**Requirements:**
|
| 23 |
+
|
| 24 |
- Any modern browser
|
| 25 |
- ~300MB-2GB free RAM
|
| 26 |
- WebGPU optional (Chrome 113+, Edge 113+) - enables GPU acceleration
|
| 27 |
|
| 28 |
**How It Works:**
|
| 29 |
+
|
| 30 |
1. Checks WebGPU availability via `"gpu" in navigator`
|
| 31 |
2. Downloads model from HuggingFace (GGUF format, cached in the browser's Origin Private File System)
|
| 32 |
3. Loads model with `n_gpu_layers: 99999` if WebGPU is present, else `0` for CPU-only
|
|
|
|
| 34 |
5. Streams tokens via the OAI-compatible `createChatCompletion` API
|
| 35 |
|
| 36 |
**Pre-configured Models:**
|
| 37 |
+
Several, stored at `Felladrin/gguf-sharded-*` on HuggingFace. The full, authoritative list with HuggingFace repo IDs and sizes lives in `client/modules/wllama.ts` (`wllamaModels`) and is rendered in the Settings dropdown; refer to that file rather than a static table here, since models are added/swapped frequently.
|
| 38 |
|
| 39 |
Two constraints decide what can join that list, both measured against the shipped wasm build:
|
| 40 |
+
|
| 41 |
- **Every model file must be a shard.** Chromium's OPFS storage silently truncates files past 2000 MiB, so a monolithic GGUF above that size arrives corrupted in the browser. Shards also keep progressive download working.
|
| 42 |
- **The wasm linear memory ceiling is 4 GiB.** Models at ~3.5 GB and above OOM during buffer reserve, so the practical catalog ceiling is around 3.7 GB of weights.
|
| 43 |
|
| 44 |
+
Models whose chat template thinks by default (Granite 4.2, Gemma 4, Spark X2.6, Qwen 3 and 3.5, Polaris, DeepSeek R1 Distill, LittleLamb) keep thinking: wllama splits their chain of thought into a `reasoning_content` channel, and MiniSearch wraps it in the configured reasoning markers so the thinking UI shows it while the answer follows. Thinking spends the same token budget as the answer, so these models are slower. To keep a long-winded model from thinking its way out of the context window without ever answering, thinking models ship with `reasoningBudgetTokens` (wllama's `reasoning_budget_tokens`, currently 1024): once the chain of thought runs that long, the engine closes the reasoning channel and forces the final answer. A cut-off chain of thought degrades the answer, so the budget is a backstop, not a target: every model also gets `reasoning_effort: "low"` in its chat template variables, which templates that support it (Granite 4.2, Spark X2.6) answer by thinking less, and the others ignore.
|
| 45 |
|
| 46 |
`node scripts/wllama-catalog-check.cjs` HEADs every shard of every catalog model on HuggingFace and verifies the claimed sizes; run it after adding or re-sharding a model.
|
| 47 |
|
| 48 |
**Configuration:**
|
| 49 |
+
|
| 50 |
- Settings โ Inference Type: `Browser`
|
| 51 |
- Settings โ Wllama Model: Select from dropdown
|
| 52 |
- Settings โ CPU threads: Relevant only when WebGPU is unavailable
|
| 53 |
|
| 54 |
**WebGPU Detection:**
|
| 55 |
+
|
| 56 |
```typescript
|
| 57 |
// client/modules/webGpu.ts
|
| 58 |
/** Whether the browser supports the WebGPU API. */
|
|
|
|
| 60 |
```
|
| 61 |
|
| 62 |
**Limitations:**
|
| 63 |
+
|
| 64 |
- First load requires model download (progressive via sharded files)
|
| 65 |
- Limited to smaller models (~4B params max due to browser memory)
|
| 66 |
- CPU mode is 2-5x slower than WebGPU
|
|
|
|
| 70 |
Uses OpenAI's API or any OpenAI-compatible service.
|
| 71 |
|
| 72 |
**Setup:**
|
| 73 |
+
|
| 74 |
1. Get API key from OpenAI or compatible provider
|
| 75 |
2. Settings โ Inference Type: `OpenAI`
|
| 76 |
3. Settings โ OpenAI API Key: Enter key
|
| 77 |
4. Settings โ OpenAI Model: Select or enter model ID
|
| 78 |
|
| 79 |
**Supported Providers:**
|
| 80 |
+
|
| 81 |
- OpenAI (gpt-4, gpt-3.5-turbo)
|
| 82 |
- Anthropic (via OpenAI-compatible endpoint)
|
| 83 |
- Google (Gemini via OpenAI-compatible endpoint)
|
| 84 |
- Any custom provider with OpenAI-compatible API
|
| 85 |
|
| 86 |
**Features:**
|
| 87 |
+
|
| 88 |
- Streaming responses, built on the Vercel AI SDK (`ai` + `@ai-sdk/openai-compatible`)
|
| 89 |
- Auto model selection (if blank)
|
| 90 |
- On a request error, automatically retries up to 5 times, switching to a different model fetched from the provider's `/models` endpoint each time (100ms ร attempt backoff between retries)
|
| 91 |
- Reasoning content support: reasoning/thinking output is wrapped between configurable `reasoningStartMarker`/`reasoningEndMarker` settings (default `<think>` / `</think>`)
|
| 92 |
|
| 93 |
**Configuration:**
|
| 94 |
+
|
| 95 |
```typescript
|
| 96 |
{
|
| 97 |
inferenceType: 'openai',
|
|
|
|
| 102 |
```
|
| 103 |
|
| 104 |
**Privacy Considerations:**
|
| 105 |
+
|
| 106 |
- Search queries and results sent to OpenAI
|
| 107 |
- Not suitable for sensitive data
|
| 108 |
- Consider internal API for private data
|
|
|
|
| 112 |
Uses aihorde.net, a distributed volunteer GPU network.
|
| 113 |
|
| 114 |
**Setup:**
|
| 115 |
+
|
| 116 |
1. Settings โ Inference Type: `AI Horde`
|
| 117 |
2. (Optional) Settings โ AI Horde API Key: Get from aihorde.net
|
| 118 |
3. Settings โ AI Horde Model: Select preferred model
|
| 119 |
|
| 120 |
**How It Works:**
|
| 121 |
+
|
| 122 |
1. Request sent to AI Horde API
|
| 123 |
2. Distributed to volunteer workers
|
| 124 |
3. Multiple workers may process in parallel
|
|
|
|
| 126 |
5. Results streamed back
|
| 127 |
|
| 128 |
**Features:**
|
| 129 |
+
|
| 130 |
- Free to use (anonymous or authenticated)
|
| 131 |
- Kudos-based priority system
|
| 132 |
- Large model selection (70B+ params available)
|
| 133 |
- No API key required (but recommended for priority)
|
| 134 |
|
| 135 |
**Configuration:**
|
| 136 |
+
|
| 137 |
```typescript
|
| 138 |
{
|
| 139 |
inferenceType: 'horde',
|
|
|
|
| 143 |
```
|
| 144 |
|
| 145 |
**Limitations:**
|
| 146 |
+
|
| 147 |
- Variable latency (depends on worker availability)
|
| 148 |
- Quality varies by worker
|
| 149 |
- May queue during high demand
|
|
|
|
| 154 |
Self-hosted OpenAI-compatible API for teams and compliance.
|
| 155 |
|
| 156 |
**Setup:**
|
| 157 |
+
|
| 158 |
1. Host an OpenAI-compatible API (e.g., vLLM, llama.cpp server, Ollama with OpenAI compat)
|
| 159 |
2. Configure environment variables (see `docs/configuration.md`)
|
| 160 |
3. Settings โ Inference Type: `Internal`
|
| 161 |
|
| 162 |
**Environment Variables:**
|
| 163 |
+
|
| 164 |
```bash
|
| 165 |
INTERNAL_OPENAI_COMPATIBLE_API_BASE_URL="https://llm.company.com/v1"
|
| 166 |
INTERNAL_OPENAI_COMPATIBLE_API_KEY="sk-internal-xxx"
|
|
|
|
| 170 |
|
| 171 |
**Server-Side Proxy:**
|
| 172 |
The internal API uses a server-side proxy to:
|
| 173 |
+
|
| 174 |
- Hide API keys from client
|
| 175 |
- Add request logging/auditing
|
| 176 |
- Apply rate limiting
|
| 177 |
- Enable token-based authentication
|
| 178 |
|
| 179 |
**Endpoint:**
|
| 180 |
+
|
| 181 |
```
|
| 182 |
POST /inference?token=<argon2-hashed-search-token>
|
| 183 |
Content-Type: application/json
|
|
|
|
| 199 |
the `/status` section of `docs/overview.md` for the field reference.
|
| 200 |
|
| 201 |
**Features:**
|
| 202 |
+
|
| 203 |
- Private data stays in your infrastructure
|
| 204 |
- Custom model selection
|
| 205 |
- Server-side logging
|
| 206 |
- Compatible with any OpenAI-compatible API
|
| 207 |
|
| 208 |
**Recommended Self-Hosted Options:**
|
| 209 |
+
|
| 210 |
- **vLLM**: High-performance, production-ready
|
| 211 |
- **llama.cpp server**: Single binary, easy setup
|
| 212 |
- **Ollama**: Simple, Docker-friendly
|
|
|
|
| 279 |
### Response Throttling
|
| 280 |
|
| 281 |
Two PubSub update functions are throttled to 12 updates per second to ensure UI performance during high-frequency streaming:
|
| 282 |
+
|
| 283 |
- `updateResponse` → `responsePubSub`
|
| 284 |
- `updateReasoningContent` → `reasoningContentPubSub`
|
| 285 |
|
|
|
|
| 306 |
- **Available for Context:** 75% (~3072 tokens)
|
| 307 |
|
| 308 |
**Allocation Priority:**
|
| 309 |
+
|
| 310 |
1. System prompt (with search results)
|
| 311 |
2. Conversation summary (if exists)
|
| 312 |
3. Recent chat messages (newest first)
|
|
|
|
| 315 |
### Rolling Summaries
|
| 316 |
|
| 317 |
When conversation exceeds token budget:
|
| 318 |
+
|
| 319 |
1. **Detect Overflow:** Current tokens > 3072
|
| 320 |
2. **Generate Summary:** Call LLM with 800-token limit
|
| 321 |
3. **Store Summary:** Save in conversationSummaryPubSub
|
|
|
|
| 323 |
5. **Continue:** Use summary + remaining messages for context
|
| 324 |
|
| 325 |
**Summary Prompt:**
|
| 326 |
+
|
| 327 |
```
|
| 328 |
Summarize this conversation in 3-5 sentences, preserving key facts
|
| 329 |
and user intent. Be concise but informative.
|
|
|
|
| 390 |
### State Recovery
|
| 391 |
|
| 392 |
If generation fails mid-stream:
|
| 393 |
+
|
| 394 |
1. Set `textGenerationState` to `failed`
|
| 395 |
2. Preserve partial response in `responsePubSub`
|
| 396 |
3. Allow user to retry or modify query
|
|
|
|
| 410 |
### Progressive Model Loading
|
| 411 |
|
| 412 |
Wllama models are sharded (split into chunks):
|
| 413 |
+
|
| 414 |
1. Download metadata first (small, fast)
|
| 415 |
2. Download required shards progressively
|
| 416 |
3. Start inference when first shards available
|
docs/api.md
CHANGED
|
@@ -392,6 +392,7 @@ a restart with different environment variables takes effect on the next reload.
|
|
| 392 |
"accessKeysEnabled": false,
|
| 393 |
"accessKeyTimeoutHours": 0,
|
| 394 |
"wllamaDefaultModelId": "...",
|
|
|
|
| 395 |
"internalApiEnabled": false,
|
| 396 |
"internalApiName": "...",
|
| 397 |
"defaultInferenceType": "...",
|
|
|
|
| 392 |
"accessKeysEnabled": false,
|
| 393 |
"accessKeyTimeoutHours": 0,
|
| 394 |
"wllamaDefaultModelId": "...",
|
| 395 |
+
"wllamaDefaultGpuModelId": "...",
|
| 396 |
"internalApiEnabled": false,
|
| 397 |
"internalApiName": "...",
|
| 398 |
"defaultInferenceType": "...",
|
docs/coding-conventions.md
CHANGED
|
@@ -3,6 +3,7 @@
|
|
| 3 |
## Tech Stack
|
| 4 |
|
| 5 |
React 19 + TypeScript + Mantine UI v9 (`@mantine/core`, `@mantine/hooks`, `@mantine/form`, `@mantine/notifications`, `@mantine/code-highlight`, `@mantine/carousel`)
|
|
|
|
| 6 |
- **Build Tool**: Vite 8
|
| 7 |
- **Linting/Formatting**: Biome 2
|
| 8 |
- **Testing**: Vitest 5 + Testing Library
|
|
|
|
| 3 |
## Tech Stack
|
| 4 |
|
| 5 |
React 19 + TypeScript + Mantine UI v9 (`@mantine/core`, `@mantine/hooks`, `@mantine/form`, `@mantine/notifications`, `@mantine/code-highlight`, `@mantine/carousel`)
|
| 6 |
+
|
| 7 |
- **Build Tool**: Vite 8
|
| 8 |
- **Linting/Formatting**: Biome 2
|
| 9 |
- **Testing**: Vitest 5 + Testing Library
|
docs/configuration.md
CHANGED
|
@@ -12,6 +12,7 @@ All configuration is done via environment variables. Create a `.env` file in the
|
|
| 12 |
| `ACCESS_KEY_TIMEOUT_HOURS` | `24` | Hours to cache validated keys in browser. Set to `0` to require validation on every request |
|
| 13 |
|
| 14 |
**Example:**
|
|
|
|
| 15 |
```bash
|
| 16 |
ACCESS_KEYS="my-secret-key-1,my-secret-key-2"
|
| 17 |
ACCESS_KEY_TIMEOUT_HOURS="24"
|
|
@@ -21,9 +22,12 @@ ACCESS_KEY_TIMEOUT_HOURS="24"
|
|
| 21 |
|
| 22 |
Configure default models for different inference types:
|
| 23 |
|
| 24 |
-
| Variable
|
| 25 |
-
| ------------------------- | ----------------- | ---------------------------------------------------------------------------- |
|
| 26 |
-
| `WLLAMA_DEFAULT_MODEL_ID`
|
|
|
|
|
|
|
|
|
|
| 27 |
|
| 28 |
### Server Model Request Configuration
|
| 29 |
|
|
@@ -55,6 +59,7 @@ For self-hosted OpenAI-compatible APIs:
|
|
| 55 |
| `INTERNAL_OPENAI_COMPATIBLE_API_NAME` | `Internal API` | Display name shown in UI |
|
| 56 |
|
| 57 |
**Example:**
|
|
|
|
| 58 |
```bash
|
| 59 |
INTERNAL_OPENAI_COMPATIBLE_API_BASE_URL="https://llm.internal.company.com/v1"
|
| 60 |
INTERNAL_OPENAI_COMPATIBLE_API_KEY="sk-internal-xxx"
|
|
@@ -103,6 +108,7 @@ When SearXNG fails on a text search, the server can retry that search against se
|
|
| 103 |
`SEARCH_FALLBACK_PROVIDER` and `SEARCH_FALLBACK_API_KEY` are gone. If either is still set to a non-empty value while `SEARCH_FALLBACK_ENABLED` is on, the server stops at start with an error that names the variable to use instead, so an upgrade cannot silently drop a key: move the key to `SEARCH_FALLBACK_PARALLEL_API_KEY` or `SEARCH_FALLBACK_YOUCOM_API_KEY`, and the provider choice to `SEARCH_FALLBACK_PROVIDERS`.
|
| 104 |
|
| 105 |
Each provider has its own terms, and each operator is responsible for following them:
|
|
|
|
| 106 |
- **Parallel**: Parallel describes its keyless tier as for exploration and light use, and asks for an account and an API key for production use ([docs.parallel.ai](https://docs.parallel.ai/integrations/mcp/search-mcp)).
|
| 107 |
- **You.com**: the keyless `profile=free` endpoint allows 100 queries per day ([you.com](https://you.com/docs/build-with-agents/mcp-server)). Its API acceptable use policy asks integrations to honor 429 responses ([you.com](https://you.com/acceptable-use-policy-may-2026)).
|
| 108 |
|
|
@@ -142,7 +148,7 @@ Settings are stored in browser localStorage and can be changed via the Settings
|
|
| 142 |
| `inferenceType` | enum | `'browser'` | AI provider: `browser`, `openai`, `horde`, `internal` |
|
| 143 |
| `cpuThreads` | number | (auto) | Number of CPU threads for inference (Wllama), defaults to half the logical processors (`navigator.hardwareConcurrency / 2`), minimum 1 |
|
| 144 |
| `allowAiModelDownload` | boolean | `false` | Allow automatic AI model downloads |
|
| 145 |
-
| `wllamaModelId` | string |
|
| 146 |
| `hordeApiKey` | string | `'0000000000'` | AI Horde API key (default is anonymous) |
|
| 147 |
| `hordeModel` | string | `''` | Specific AI Horde model to request |
|
| 148 |
| `openAiApiBaseUrl` | string | `''` | Base URL for the OpenAI-compatible API |
|
|
@@ -153,17 +159,19 @@ Settings are stored in browser localStorage and can be changed via the Settings
|
|
| 153 |
### Model Selection
|
| 154 |
|
| 155 |
**Wllama Models:**
|
| 156 |
-
|
| 157 |
-
-
|
| 158 |
- All quantized to a Q3/Q4-class quant (Q4_K_S, Q4_K_M, Q3_K_S, UD-Q3_K_XL or UD-Q4_K_XL)
|
| 159 |
- Stored at: `Felladrin/gguf-sharded-*` on HuggingFace, always as sharded GGUFs
|
| 160 |
|
| 161 |
**OpenAI/Internal:**
|
|
|
|
| 162 |
- Any OpenAI-compatible API
|
| 163 |
- Auto-model detection if not specified
|
| 164 |
- Supports streaming and reasoning models
|
| 165 |
|
| 166 |
**AI Horde:**
|
|
|
|
| 167 |
- Uses aihorde.net distributed network
|
| 168 |
- Anonymous or authenticated access
|
| 169 |
- Parallel generation with race conditions
|
|
@@ -184,15 +192,15 @@ The default system prompt supports template placeholders populated at runtime:
|
|
| 184 |
|
| 185 |
```typescript
|
| 186 |
{
|
| 187 |
-
systemPrompt: `Answer using the search results below
|
| 188 |
|
| 189 |
-
Cite
|
| 190 |
|
| 191 |
-
|
| 192 |
|
| 193 |
-
|
| 194 |
|
| 195 |
-
|
| 196 |
|
| 197 |
Search results:
|
| 198 |
|
|
@@ -203,6 +211,7 @@ Search results:
|
|
| 203 |
```
|
| 204 |
|
| 205 |
**Placeholders:**
|
|
|
|
| 206 |
- `{{currentDate}}`: Current date injected at generation time
|
| 207 |
- `{{dateTime}}`: Alias for `{{currentDate}}` - both are replaced with the same current date value
|
| 208 |
- `{{searchResults}}`: Formatted search results from the web search, including the page excerpts when `enablePageContentFetch` is on
|
|
@@ -253,6 +262,7 @@ Same structure but without volume mounts and with pre-built assets.
|
|
| 253 |
### Dockerfile Environment
|
| 254 |
|
| 255 |
The Dockerfile sets up a single runtime stage:
|
|
|
|
| 256 |
- Node.js LTS
|
| 257 |
- Python 3 + SearXNG
|
| 258 |
|
|
@@ -350,6 +360,7 @@ services:
|
|
| 350 |
```
|
| 351 |
|
| 352 |
Two things about that command are required, not optional:
|
|
|
|
| 353 |
- **The file must be readable by uid 1000.** The container runs as `node`, and a file readable only by its owner on the host is invisible to it.
|
| 354 |
- **`secret_key` must be set to your own random value.** SearXNG refuses to start on the `ultrasecretkey` placeholder and logs `server.secret_key is not changed. Please use something else instead of ultrasecretkey.` The image's own file never carries the placeholder, because the Dockerfile replaces it at build time, but a file you write has to carry a real key.
|
| 355 |
|
|
@@ -388,6 +399,7 @@ Client-facing configuration (access keys, inference type, internal API settings)
|
|
| 388 |
"accessKeysEnabled": true,
|
| 389 |
"accessKeyTimeoutHours": 24,
|
| 390 |
"wllamaDefaultModelId": "littlelamb-290m",
|
|
|
|
| 391 |
"internalApiEnabled": true,
|
| 392 |
"internalApiName": "Internal API",
|
| 393 |
"defaultInferenceType": "browser",
|
|
@@ -404,6 +416,7 @@ Client-facing configuration (access keys, inference type, internal API settings)
|
|
| 404 |
| `ACCESS_KEYS` | Runtime | Read from `/api/config` |
|
| 405 |
| `ACCESS_KEY_TIMEOUT_HOURS` | Runtime | Read from `/api/config` |
|
| 406 |
| `WLLAMA_DEFAULT_MODEL_ID` | Runtime | Read from `/api/config` |
|
|
|
|
| 407 |
| `INTERNAL_OPENAI_COMPATIBLE_API_*` | Runtime | Read from `/api/config` (except `API_KEY` which is server-only) |
|
| 408 |
| `DEFAULT_INFERENCE_TYPE` | Runtime | Read from `/api/config` |
|
| 409 |
| Search token | Runtime | Read from `/api/config`, so a reload picks up the token of whichever server answers |
|
|
@@ -451,6 +464,7 @@ MiniSearch logs internal events to an in-app log panel (see the Logs section of
|
|
| 451 |
**Diagnosing empty search results:** the client distinguishes a search that returned no results (a no-results alert, state stays `completed`) from one that failed outright on an outage (a distinct failed state, with a retry on text searches). An empty response that names unresponsive engines is retried within the same budget as a 500 and counts as the second if it is still empty when the retries are spent, since a rate-limited, suspended or CAPTCHA-challenged engine is reported that way and not with a status code. The actual reason is printed to the server's console (via `server/webSearchService.ts`, always-on `debug` logging), either the SearXNG engines that failed (timeouts, suspensions, rate limits, from SearXNG's `unresponsive_engines` field) or a note that all returned results were discarded during processing (missing title, snippet, or media source). Check the server logs (`docker compose logs`) when troubleshooting failed searches.
|
| 452 |
|
| 453 |
Check effective configuration:
|
|
|
|
| 454 |
```typescript
|
| 455 |
// In browser console
|
| 456 |
console.log("Settings:", JSON.parse(localStorage.getItem("settings") || "{}"));
|
|
|
|
| 12 |
| `ACCESS_KEY_TIMEOUT_HOURS` | `24` | Hours to cache validated keys in browser. Set to `0` to require validation on every request |
|
| 13 |
|
| 14 |
**Example:**
|
| 15 |
+
|
| 16 |
```bash
|
| 17 |
ACCESS_KEYS="my-secret-key-1,my-secret-key-2"
|
| 18 |
ACCESS_KEY_TIMEOUT_HOURS="24"
|
|
|
|
| 22 |
|
| 23 |
Configure default models for different inference types:
|
| 24 |
|
| 25 |
+
| Variable | Default | Description |
|
| 26 |
+
| ----------------------------- | ----------------- | --------------------------------------------------------------------------------------------- |
|
| 27 |
+
| `WLLAMA_DEFAULT_MODEL_ID` | `littlelamb-290m` | Default Wllama model ID for browsers without WebGPU, where the model runs on the CPU |
|
| 28 |
+
| `WLLAMA_DEFAULT_GPU_MODEL_ID` | `gemma-3-4b` | Default Wllama model ID for browsers with WebGPU. It does not fall back to the variable above |
|
| 29 |
+
|
| 30 |
+
Both only seed a profile that has never saved settings; a user's own choice always wins. Detection is the same `"gpu" in navigator` check that decides whether the model's layers go to the GPU (`client/modules/webGpu.ts`).
|
| 31 |
|
| 32 |
### Server Model Request Configuration
|
| 33 |
|
|
|
|
| 59 |
| `INTERNAL_OPENAI_COMPATIBLE_API_NAME` | `Internal API` | Display name shown in UI |
|
| 60 |
|
| 61 |
**Example:**
|
| 62 |
+
|
| 63 |
```bash
|
| 64 |
INTERNAL_OPENAI_COMPATIBLE_API_BASE_URL="https://llm.internal.company.com/v1"
|
| 65 |
INTERNAL_OPENAI_COMPATIBLE_API_KEY="sk-internal-xxx"
|
|
|
|
| 108 |
`SEARCH_FALLBACK_PROVIDER` and `SEARCH_FALLBACK_API_KEY` are gone. If either is still set to a non-empty value while `SEARCH_FALLBACK_ENABLED` is on, the server stops at start with an error that names the variable to use instead, so an upgrade cannot silently drop a key: move the key to `SEARCH_FALLBACK_PARALLEL_API_KEY` or `SEARCH_FALLBACK_YOUCOM_API_KEY`, and the provider choice to `SEARCH_FALLBACK_PROVIDERS`.
|
| 109 |
|
| 110 |
Each provider has its own terms, and each operator is responsible for following them:
|
| 111 |
+
|
| 112 |
- **Parallel**: Parallel describes its keyless tier as for exploration and light use, and asks for an account and an API key for production use ([docs.parallel.ai](https://docs.parallel.ai/integrations/mcp/search-mcp)).
|
| 113 |
- **You.com**: the keyless `profile=free` endpoint allows 100 queries per day ([you.com](https://you.com/docs/build-with-agents/mcp-server)). Its API acceptable use policy asks integrations to honor 429 responses ([you.com](https://you.com/acceptable-use-policy-may-2026)).
|
| 114 |
|
|
|
|
| 148 |
| `inferenceType` | enum | `'browser'` | AI provider: `browser`, `openai`, `horde`, `internal` |
|
| 149 |
| `cpuThreads` | number | (auto) | Number of CPU threads for inference (Wllama), defaults to half the logical processors (`navigator.hardwareConcurrency / 2`), minimum 1 |
|
| 150 |
| `allowAiModelDownload` | boolean | `false` | Allow automatic AI model downloads |
|
| 151 |
+
| `wllamaModelId` | string | WebGPU or CPU default, see above. | Default Wllama model ID, chosen by whether the browser has WebGPU |
|
| 152 |
| `hordeApiKey` | string | `'0000000000'` | AI Horde API key (default is anonymous) |
|
| 153 |
| `hordeModel` | string | `''` | Specific AI Horde model to request |
|
| 154 |
| `openAiApiBaseUrl` | string | `''` | Base URL for the OpenAI-compatible API |
|
|
|
|
| 159 |
### Model Selection
|
| 160 |
|
| 161 |
**Wllama Models:**
|
| 162 |
+
|
| 163 |
+
- Several pre-configured models, listed in `client/modules/wllama.ts`
|
| 164 |
- All quantized to a Q3/Q4-class quant (Q4_K_S, Q4_K_M, Q3_K_S, UD-Q3_K_XL or UD-Q4_K_XL)
|
| 165 |
- Stored at: `Felladrin/gguf-sharded-*` on HuggingFace, always as sharded GGUFs
|
| 166 |
|
| 167 |
**OpenAI/Internal:**
|
| 168 |
+
|
| 169 |
- Any OpenAI-compatible API
|
| 170 |
- Auto-model detection if not specified
|
| 171 |
- Supports streaming and reasoning models
|
| 172 |
|
| 173 |
**AI Horde:**
|
| 174 |
+
|
| 175 |
- Uses aihorde.net distributed network
|
| 176 |
- Anonymous or authenticated access
|
| 177 |
- Parallel generation with race conditions
|
|
|
|
| 192 |
|
| 193 |
```typescript
|
| 194 |
{
|
| 195 |
+
systemPrompt: `Answer the question using the search results below. Reply in the same language as the question.
|
| 196 |
|
| 197 |
+
Cite each fact with a Markdown link right after it, using the site's domain as the link text. Example: [example.com](https://example.com/page).
|
| 198 |
|
| 199 |
+
If you answer from your own knowledge because the results do not cover it, say so.
|
| 200 |
|
| 201 |
+
Use only these Markdown elements: link, bold, italic, code, quote, table.
|
| 202 |
|
| 203 |
+
Today's date is {{currentDate}}. Use it for relative dates such as "yesterday".
|
| 204 |
|
| 205 |
Search results:
|
| 206 |
|
|
|
|
| 211 |
```
|
| 212 |
|
| 213 |
**Placeholders:**
|
| 214 |
+
|
| 215 |
- `{{currentDate}}`: Current date injected at generation time
|
| 216 |
- `{{dateTime}}`: Alias for `{{currentDate}}` - both are replaced with the same current date value
|
| 217 |
- `{{searchResults}}`: Formatted search results from the web search, including the page excerpts when `enablePageContentFetch` is on
|
|
|
|
| 262 |
### Dockerfile Environment
|
| 263 |
|
| 264 |
The Dockerfile sets up a single runtime stage:
|
| 265 |
+
|
| 266 |
- Node.js LTS
|
| 267 |
- Python 3 + SearXNG
|
| 268 |
|
|
|
|
| 360 |
```
|
| 361 |
|
| 362 |
Two things about that command are required, not optional:
|
| 363 |
+
|
| 364 |
- **The file must be readable by uid 1000.** The container runs as `node`, and a file readable only by its owner on the host is invisible to it.
|
| 365 |
- **`secret_key` must be set to your own random value.** SearXNG refuses to start on the `ultrasecretkey` placeholder and logs `server.secret_key is not changed. Please use something else instead of ultrasecretkey.` The image's own file never carries the placeholder, because the Dockerfile replaces it at build time, but a file you write has to carry a real key.
|
| 366 |
|
|
|
|
| 399 |
"accessKeysEnabled": true,
|
| 400 |
"accessKeyTimeoutHours": 24,
|
| 401 |
"wllamaDefaultModelId": "littlelamb-290m",
|
| 402 |
+
"wllamaDefaultGpuModelId": "gemma-3-4b",
|
| 403 |
"internalApiEnabled": true,
|
| 404 |
"internalApiName": "Internal API",
|
| 405 |
"defaultInferenceType": "browser",
|
|
|
|
| 416 |
| `ACCESS_KEYS` | Runtime | Read from `/api/config` |
|
| 417 |
| `ACCESS_KEY_TIMEOUT_HOURS` | Runtime | Read from `/api/config` |
|
| 418 |
| `WLLAMA_DEFAULT_MODEL_ID` | Runtime | Read from `/api/config` |
|
| 419 |
+
| `WLLAMA_DEFAULT_GPU_MODEL_ID` | Runtime | Read from `/api/config` |
|
| 420 |
| `INTERNAL_OPENAI_COMPATIBLE_API_*` | Runtime | Read from `/api/config` (except `API_KEY` which is server-only) |
|
| 421 |
| `DEFAULT_INFERENCE_TYPE` | Runtime | Read from `/api/config` |
|
| 422 |
| Search token | Runtime | Read from `/api/config`, so a reload picks up the token of whichever server answers |
|
|
|
|
| 464 |
**Diagnosing empty search results:** the client distinguishes a search that returned no results (a no-results alert, state stays `completed`) from one that failed outright on an outage (a distinct failed state, with a retry on text searches). An empty response that names unresponsive engines is retried within the same budget as a 500 and counts as the second if it is still empty when the retries are spent, since a rate-limited, suspended or CAPTCHA-challenged engine is reported that way and not with a status code. The actual reason is printed to the server's console (via `server/webSearchService.ts`, always-on `debug` logging), either the SearXNG engines that failed (timeouts, suspensions, rate limits, from SearXNG's `unresponsive_engines` field) or a note that all returned results were discarded during processing (missing title, snippet, or media source). Check the server logs (`docker compose logs`) when troubleshooting failed searches.
|
| 465 |
|
| 466 |
Check effective configuration:
|
| 467 |
+
|
| 468 |
```typescript
|
| 469 |
// In browser console
|
| 470 |
console.log("Settings:", JSON.parse(localStorage.getItem("settings") || "{}"));
|
docs/core-technologies.md
CHANGED
|
@@ -29,6 +29,7 @@
|
|
| 29 |
## Technology Selection Criteria
|
| 30 |
|
| 31 |
Technologies are chosen based on:
|
|
|
|
| 32 |
- **Agent legibility**: Easy for agents to model and reason about
|
| 33 |
- **Stability**: Mature, well-documented libraries
|
| 34 |
- **Composability**: Clean APIs that work well together
|
|
|
|
| 29 |
## Technology Selection Criteria
|
| 30 |
|
| 31 |
Technologies are chosen based on:
|
| 32 |
+
|
| 33 |
- **Agent legibility**: Easy for agents to model and reason about
|
| 34 |
- **Stability**: Mature, well-documented libraries
|
| 35 |
- **Composability**: Clean APIs that work well together
|
docs/development-commands.md
CHANGED
|
@@ -26,13 +26,47 @@ Failure-injection cases (dependency down, empty results, aborted stream) are cat
|
|
| 26 |
`eval/` is an offline eval that gives a regression signal for changes to the
|
| 27 |
reranker, the system prompt, the search-results formatting, or the model. See
|
| 28 |
`eval/README.md` for details and the LLM-judge environment variables.
|
|
|
|
| 29 |
- **`npm run eval`**: Run the full eval (retrieval + answer)
|
| 30 |
- **`npm run eval:retrieval`**: Retrieval eval (real ONNX reranker; local only, needs the model in `server/models/`)
|
| 31 |
- **`npm run eval:answer`**: Answer eval (LLM judge; gated on `EVAL_LLM_API_KEY`, skips without one)
|
| 32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
### Coverage Reports for AI Analysis
|
| 34 |
|
| 35 |
After running `test:coverage`, AI agents can analyze these JSON files:
|
|
|
|
| 36 |
- **`coverage/coverage-summary.json`**: Quick metrics view - overall percentages per file
|
| 37 |
|
| 38 |
```json
|
|
@@ -52,6 +86,7 @@ After running `test:coverage`, AI agents can analyze these JSON files:
|
|
| 52 |
- Maps statement/branch/function IDs to source locations
|
| 53 |
|
| 54 |
AI agents can parse these to identify:
|
|
|
|
| 55 |
- Files with coverage below thresholds
|
| 56 |
- Uncovered lines and branches
|
| 57 |
- Functions without tests
|
|
@@ -78,6 +113,7 @@ The repository uses nine GitHub Actions workflows for continuous integration, de
|
|
| 78 |
### Reusable Workflow (`reusable-check-docker.yml`)
|
| 79 |
|
| 80 |
Used by both `on-push-to-main` and `on-pull-request-to-main` to run the production container smoke test:
|
|
|
|
| 81 |
1. **check-docker-container** - Builds and starts the production container with `docker compose -f docker-compose.production.yml up -d`, then verifies the main page returns HTTP 200 within 60 seconds
|
| 82 |
|
| 83 |
### Docker Image Builds
|
|
|
|
| 26 |
`eval/` is an offline eval that gives a regression signal for changes to the
|
| 27 |
reranker, the system prompt, the search-results formatting, or the model. See
|
| 28 |
`eval/README.md` for details and the LLM-judge environment variables.
|
| 29 |
+
|
| 30 |
- **`npm run eval`**: Run the full eval (retrieval + answer)
|
| 31 |
- **`npm run eval:retrieval`**: Retrieval eval (real ONNX reranker; local only, needs the model in `server/models/`)
|
| 32 |
- **`npm run eval:answer`**: Answer eval (LLM judge; gated on `EVAL_LLM_API_KEY`, skips without one)
|
| 33 |
|
| 34 |
+
## wllama benchmark
|
| 35 |
+
|
| 36 |
+
`npm run bench:wllama` loads a wllama model in headless Chromium, times it on the questions in `scripts/wllama-bench/cases.json`, and scores the answers. It goes through MiniSearch's own code: the catalog entry in `client/modules/wllama.ts`, the warmup-then-load in `initializeWllama`, and the chat messages built from search results and page excerpts. So the numbers and answers are the ones a user would get. Use it to vet a model before it joins the catalog, or to compare catalog models. The scoring rules and the latest results are in [wllama-model-evaluation.md](wllama-model-evaluation.md).
|
| 37 |
+
|
| 38 |
+
```bash
|
| 39 |
+
npm run bench:wllama -- --model qwen-3.5-0.8b --model granite-4.2-3b --device webgpu
|
| 40 |
+
npm run bench:wllama -- --model owner/repo/Model-Q4_K_M-00001-of-00003.gguf --template-kwargs '{"enable_thinking":false}'
|
| 41 |
+
```
|
| 42 |
+
|
| 43 |
+
| Option | Default | Meaning |
|
| 44 |
+
| ---------------------- | -------------------- | -------------------------------------------------------------------------------------- |
|
| 45 |
+
| `--model` (repeatable) | required | A catalog id, or `owner/repo/file.gguf` on Hugging Face for a model not in the catalog |
|
| 46 |
+
| `--device` | `both` | `webgpu`, `cpu` (`n_gpu_layers: 0`, what MiniSearch does without WebGPU), or `both` |
|
| 47 |
+
| `--runs` | `1` | Repetitions per model and device |
|
| 48 |
+
| `--threads` | `cpuThreads` setting | CPU threads |
|
| 49 |
+
| `--template-kwargs` | catalog value | JSON for `chat_template_kwargs`; replaces the catalog's |
|
| 50 |
+
| `--reasoning-budget` | catalog value | `reasoning_budget_tokens`, applied at load time as the app does |
|
| 51 |
+
| `--chrome` | Playwright Chromium | Path to another Chromium binary |
|
| 52 |
+
| `--json` | none | Writes every result, including the full reasoning and answer, to this file |
|
| 53 |
+
| `--cases` | `cases.json` | Another question set with the same shape |
|
| 54 |
+
| `--case-timeout` | `300` | Seconds before an answer is cut off; the scorer counts the cut as a violation |
|
| 55 |
+
| `--profile` | see below | Browser profile directory; two runs at once need two, since Chromium locks one in use |
|
| 56 |
+
| `--system-prompt` | none | A file with another system prompt, to measure a change to it before shipping it |
|
| 57 |
+
| `--verbose` | off | Streams the page console, llama.cpp's log included |
|
| 58 |
+
|
| 59 |
+
Things to know:
|
| 60 |
+
|
| 61 |
+
- Run it on the host, not in the dev container, which has no browser or GPU. It needs the host's `npm install` and Playwright's Chromium (`npx playwright install chromium`).
|
| 62 |
+
- An old system Chrome is not a stand-in: Chrome 137 rejects the `subgroup_id` WGSL builtin that wllama's WebGPU shaders use, so every model aborts on WebGPU there.
|
| 63 |
+
- The browser profile lives in `<system temp>/minisearch-wllama-bench-profile`, so OPFS keeps each model after its first download. Delete that folder to reclaim the space.
|
| 64 |
+
- On a failure the script prints the page's last console lines; llama.cpp's reason for an abort is usually among them.
|
| 65 |
+
|
| 66 |
### Coverage Reports for AI Analysis
|
| 67 |
|
| 68 |
After running `test:coverage`, AI agents can analyze these JSON files:
|
| 69 |
+
|
| 70 |
- **`coverage/coverage-summary.json`**: Quick metrics view - overall percentages per file
|
| 71 |
|
| 72 |
```json
|
|
|
|
| 86 |
- Maps statement/branch/function IDs to source locations
|
| 87 |
|
| 88 |
AI agents can parse these to identify:
|
| 89 |
+
|
| 90 |
- Files with coverage below thresholds
|
| 91 |
- Uncovered lines and branches
|
| 92 |
- Functions without tests
|
|
|
|
| 113 |
### Reusable Workflow (`reusable-check-docker.yml`)
|
| 114 |
|
| 115 |
Used by both `on-push-to-main` and `on-pull-request-to-main` to run the production container smoke test:
|
| 116 |
+
|
| 117 |
1. **check-docker-container** - Builds and starts the production container with `docker compose -f docker-compose.production.yml up -d`, then verifies the main page returns HTTP 200 within 60 seconds
|
| 118 |
|
| 119 |
### Docker Image Builds
|
docs/glossary.md
CHANGED
|
@@ -7,6 +7,7 @@ Codebase-specific terms, jargon, and domain concepts used in MiniSearch.
|
|
| 7 |
### Search Token & Hash
|
| 8 |
|
| 9 |
A security mechanism used to authorize communication between the client and the internal search/AI endpoints.
|
|
|
|
| 10 |
- **Search Token**: A random string held by the running server, served to the client at runtime through `/api/config`. Used to verify that requests to the server come from a client that server handed the token to.
|
| 11 |
- **Search Token Hash**: To avoid exposing the raw token in all requests, the client generates a hash of the token. Managed via the `lastSearchTokenHashPubSub` channel.
|
| 12 |
- **Verification**: The server verifies these tokens to prevent unauthorized access to the search API. Stored in `server/verifiedTokens.ts` as an in-memory `Set<string>`.
|
|
@@ -25,6 +26,7 @@ MiniSearch supports multiple backends for Large Language Model (LLM) inference,
|
|
| 25 |
### PubSub (State Management)
|
| 26 |
|
| 27 |
Instead of a heavy state management library like Redux, MiniSearch uses a minimalist Publish-Subscribe pattern powered by the `create-pubsub` library.
|
|
|
|
| 28 |
- **Data Flow**: Components subscribe to "channels" (e.g., `queryPubSub`, `responsePubSub`)
|
| 29 |
- **Tuple Pattern**: Each channel is a 3-element tuple: `[update, subscribe, get]`
|
| 30 |
- **Persistence**: Some channels use `createLocalStoragePubSub` to automatically sync state with `localStorage`
|
|
@@ -33,6 +35,7 @@ Instead of a heavy state management library like Redux, MiniSearch uses a minima
|
|
| 33 |
### Reranker
|
| 34 |
|
| 35 |
A secondary search stage that takes initial results from SearXNG and re-orders them based on relevance to the query using a multilingual cross-encoder model (`mmarco-mMiniLMv2-L12-H384-v1`) running in-process via ONNX Runtime.
|
|
|
|
| 36 |
- **Implementation**: Loads the model's ONNX export with `onnxruntime-node`; no child process
|
| 37 |
- **Health Check**: Polls `/health` endpoint via `getRerankerStatus`
|
| 38 |
- **Scoring**: Results filtered using standard deviation thresholds (`kStandardDeviationFactor = 0.3`)
|
|
@@ -41,6 +44,7 @@ A secondary search stage that takes initial results from SearXNG and re-orders t
|
|
| 41 |
### Wllama
|
| 42 |
|
| 43 |
A WebAssembly (WASM) based integration of `llama.cpp` for running LLMs on the CPU in the browser.
|
|
|
|
| 44 |
- **Initialization**: Loads models from HuggingFace using `initializeWllama`
|
| 45 |
- **Warmup**: Includes a warmup phase with a single token completion using `n_threads: 1`
|
| 46 |
- **OPFS**: Uses the Origin Private File System via Wllama's cache manager to store model shards locally
|
|
@@ -49,6 +53,7 @@ A WebAssembly (WASM) based integration of `llama.cpp` for running LLMs on the CP
|
|
| 49 |
### AI Horde
|
| 50 |
|
| 51 |
A crowdsourced distributed cluster of workers providing AI inference. MiniSearch integrates with it using a polling strategy against the `/generate/text/status` endpoint.
|
|
|
|
| 52 |
- **Kudos**: Virtual currency used by the Horde. Default anonymous key is `0000000000`
|
| 53 |
- **Polling**: Requests sent to async API, status checked periodically until completion
|
| 54 |
- **Cancellation**: Can abort generation via `DELETE` on the status endpoint
|
|
@@ -56,6 +61,7 @@ A crowdsourced distributed cluster of workers providing AI inference. MiniSearch
|
|
| 56 |
### Conversation Memory & Rolling Summary
|
| 57 |
|
| 58 |
A mechanism to handle long chats that exceed the LLM context window.
|
|
|
|
| 59 |
- **Summarization**: When older messages are dropped, `createLlmSummary` asks the LLM to condense them under a limit of 800 tokens
|
| 60 |
- **Extractive Fallback**: If LLM summarization fails, `summarizeDroppedMessages` uses a token-counting extractive approach
|
| 61 |
- **Token Budget**: Computed based on `openAiContextLength` setting and current message count
|
|
@@ -73,6 +79,7 @@ GPT-Generated Unified Format. Binary format for storing LLM weights, optimized f
|
|
| 73 |
### Dexie
|
| 74 |
|
| 75 |
A minimalist wrapper for IndexedDB used for client-side persistence. MiniSearch uses two Dexie databases:
|
|
|
|
| 76 |
- **SearchCacheDatabase**: Temporary cache with TTL-based expiration
|
| 77 |
- **HistoryDatabase**: Long-term search history with retention policies
|
| 78 |
|
|
|
|
| 7 |
### Search Token & Hash
|
| 8 |
|
| 9 |
A security mechanism used to authorize communication between the client and the internal search/AI endpoints.
|
| 10 |
+
|
| 11 |
- **Search Token**: A random string held by the running server, served to the client at runtime through `/api/config`. Used to verify that requests to the server come from a client that server handed the token to.
|
| 12 |
- **Search Token Hash**: To avoid exposing the raw token in all requests, the client generates a hash of the token. Managed via the `lastSearchTokenHashPubSub` channel.
|
| 13 |
- **Verification**: The server verifies these tokens to prevent unauthorized access to the search API. Stored in `server/verifiedTokens.ts` as an in-memory `Set<string>`.
|
|
|
|
| 26 |
### PubSub (State Management)
|
| 27 |
|
| 28 |
Instead of a heavy state management library like Redux, MiniSearch uses a minimalist Publish-Subscribe pattern powered by the `create-pubsub` library.
|
| 29 |
+
|
| 30 |
- **Data Flow**: Components subscribe to "channels" (e.g., `queryPubSub`, `responsePubSub`)
|
| 31 |
- **Tuple Pattern**: Each channel is a 3-element tuple: `[update, subscribe, get]`
|
| 32 |
- **Persistence**: Some channels use `createLocalStoragePubSub` to automatically sync state with `localStorage`
|
|
|
|
| 35 |
### Reranker
|
| 36 |
|
| 37 |
A secondary search stage that takes initial results from SearXNG and re-orders them based on relevance to the query using a multilingual cross-encoder model (`mmarco-mMiniLMv2-L12-H384-v1`) running in-process via ONNX Runtime.
|
| 38 |
+
|
| 39 |
- **Implementation**: Loads the model's ONNX export with `onnxruntime-node`; no child process
|
| 40 |
- **Health Check**: Polls `/health` endpoint via `getRerankerStatus`
|
| 41 |
- **Scoring**: Results filtered using standard deviation thresholds (`kStandardDeviationFactor = 0.3`)
|
|
|
|
| 44 |
### Wllama
|
| 45 |
|
| 46 |
A WebAssembly (WASM) based integration of `llama.cpp` for running LLMs on the CPU in the browser.
|
| 47 |
+
|
| 48 |
- **Initialization**: Loads models from HuggingFace using `initializeWllama`
|
| 49 |
- **Warmup**: Includes a warmup phase with a single token completion using `n_threads: 1`
|
| 50 |
- **OPFS**: Uses the Origin Private File System via Wllama's cache manager to store model shards locally
|
|
|
|
| 53 |
### AI Horde
|
| 54 |
|
| 55 |
A crowdsourced distributed cluster of workers providing AI inference. MiniSearch integrates with it using a polling strategy against the `/generate/text/status` endpoint.
|
| 56 |
+
|
| 57 |
- **Kudos**: Virtual currency used by the Horde. Default anonymous key is `0000000000`
|
| 58 |
- **Polling**: Requests sent to async API, status checked periodically until completion
|
| 59 |
- **Cancellation**: Can abort generation via `DELETE` on the status endpoint
|
|
|
|
| 61 |
### Conversation Memory & Rolling Summary
|
| 62 |
|
| 63 |
A mechanism to handle long chats that exceed the LLM context window.
|
| 64 |
+
|
| 65 |
- **Summarization**: When older messages are dropped, `createLlmSummary` asks the LLM to condense them under a limit of 800 tokens
|
| 66 |
- **Extractive Fallback**: If LLM summarization fails, `summarizeDroppedMessages` uses a token-counting extractive approach
|
| 67 |
- **Token Budget**: Computed based on `openAiContextLength` setting and current message count
|
|
|
|
| 79 |
### Dexie
|
| 80 |
|
| 81 |
A minimalist wrapper for IndexedDB used for client-side persistence. MiniSearch uses two Dexie databases:
|
| 82 |
+
|
| 83 |
- **SearchCacheDatabase**: Temporary cache with TTL-based expiration
|
| 84 |
- **HistoryDatabase**: Long-term search history with retention policies
|
| 85 |
|
docs/overview.md
CHANGED
|
@@ -33,6 +33,7 @@ MiniSearch integrates multiple technology stacks within a unified deployment con
|
|
| 33 |
## Application Entry Points
|
| 34 |
|
| 35 |
The application has three primary entry points:
|
|
|
|
| 36 |
1. **Browser Entry**: `client/index.tsx` initializes the React application, mounting the root component and setting up error boundaries.
|
| 37 |
|
| 38 |
2. **Server Entry**: `vite.config.ts` configures the Vite development and preview servers, registering server hooks for search and inference endpoints.
|
|
@@ -42,6 +43,7 @@ The application has three primary entry points:
|
|
| 42 |
## Multi-Service Container Architecture
|
| 43 |
|
| 44 |
The Docker container runs three services concurrently:
|
|
|
|
| 45 |
- **SearXNG** - Privacy-focused metasearch engine
|
| 46 |
- **ONNX Runtime** - In-process inference for result reranking
|
| 47 |
- **Node.js application** - Main application server
|
|
@@ -57,6 +59,7 @@ PubSub channels are created using the create-pubsub package and provide type-saf
|
|
| 57 |
## Data Persistence Strategy
|
| 58 |
|
| 59 |
MiniSearch employs a dual-layer persistence approach:
|
|
|
|
| 60 |
- **IndexedDB** - Local storage for search history, settings, cached results, and saved AI transcripts
|
| 61 |
- **TTL-based caching** - 15-minute freshness cache for search results to minimize API calls, with a 24-hour stale retention as a fallback when the live search fails
|
| 62 |
|
|
@@ -112,6 +115,7 @@ The `textGeneration` module orchestrates the entire search-to-response flow, man
|
|
| 112 |
### Web Search Service Reliability
|
| 113 |
|
| 114 |
`server/webSearchService.ts` implements resilience patterns for SearXNG integration:
|
|
|
|
| 115 |
- **Circuit Breaker**: Opens after 5 consecutive failures, blocking requests for 60 seconds before attempting reset
|
| 116 |
- **Retry Logic**: Exponential backoff for HTTP 500 errors, up to 3 retries
|
| 117 |
- **Fail-Fast on Suspending Engines**: An empty response in which every unresponsive engine is under a long suspension (CAPTCHA, rate limit, access denied; SearXNG suspends them for an hour to a day) throws on the first attempt instead of spending the retry budget, which it could not possibly outlast. Transient engine errors (timeouts, server API errors) keep the full retry budget.
|
|
@@ -123,6 +127,7 @@ None of this changes which engines answer. That is decided inside SearXNG, and w
|
|
| 123 |
### Search Token Lifecycle
|
| 124 |
|
| 125 |
CSRF protection uses a token the server owns for its lifetime:
|
|
|
|
| 126 |
1. **Generation**: `regenerateSearchToken()` draws 32 random bytes on first use in a process and writes them to the token file; the file is a record, never read back as a source of truth
|
| 127 |
2. **Storage**: The token is held in memory for the life of the process, and recorded at `{os.tempdir()}/minisearch-token` with `0600` permissions
|
| 128 |
3. **Distribution**: Server serves the token to the client at runtime through `/api/config`, so a client always holds the token of the server answering it
|
|
@@ -140,6 +145,7 @@ MiniSearch uses a PubSub-based architecture where state flows through independen
|
|
| 140 |
### State Machine Transitions
|
| 141 |
|
| 142 |
**Text Generation States:**
|
|
|
|
| 143 |
- `idle` - No active generation
|
| 144 |
- `awaitingModelDownloadAllowance` - Waiting for user consent to download a browser model
|
| 145 |
- `loadingModel` - Downloading or initializing the browser (Wllama) model
|
|
@@ -151,6 +157,7 @@ MiniSearch uses a PubSub-based architecture where state flows through independen
|
|
| 151 |
- `failed` - Error occurred
|
| 152 |
|
| 153 |
**Search States:**
|
|
|
|
| 154 |
- `idle` - No active search
|
| 155 |
- `running` - Search in progress
|
| 156 |
- `completed` - Results received
|
|
@@ -206,6 +213,7 @@ MiniSearch implements all server-side logic as Vite plugin hooks. Each hook regi
|
|
| 206 |
| `rerankerServiceHook` | `server/rerankerServiceHook.ts` | Reranker model lifecycle management for result reranking |
|
| 207 |
|
| 208 |
Key server-side modules:
|
|
|
|
| 209 |
- **`server/webSearchService.ts`**: Integrates with SearXNG at `http://127.0.0.1:8888`. Implements a circuit breaker (opens after 5 failures, resets after 60s) and retry logic (up to 3 retries with exponential backoff, for 500s and for empty responses naming transiently unresponsive engines; an all-suspended set fails fast on the first attempt, since SearXNG suspensions last an hour to a day).
|
| 210 |
- **`server/pageContentService.ts`**: Reads result pages for answer grounding: SSRF-guarded fetches with a byte cap, readable-text extraction, and query-relevant passage selection. Reads run under a per-host circuit breaker (`server/pageReadHostBreaker.ts`) that skips a host for 5 minutes after 3 refusals in a row, then probes it with one read.
|
| 211 |
- **`server/searchToken.ts`**: Manages a token at `{os.tempdir()}/minisearch-token` used for CSRF protection on search requests.
|
|
@@ -437,6 +445,7 @@ MiniSearch uses a multi-layered client-side persistence strategy:
|
|
| 437 |
### IndexedDB Databases
|
| 438 |
|
| 439 |
Two separate Dexie databases handle different persistence needs:
|
|
|
|
| 440 |
1. **SearchCacheDatabase** (`client/modules/search.ts`): Temporary cache for search results with TTL-based expiration. Table schema:
|
| 441 |
- `textSearchHistory`: Keyed by hashed query, indexed by timestamp
|
| 442 |
- `imageSearchHistory`: Keyed by hashed query, indexed by timestamp
|
|
@@ -464,6 +473,7 @@ Two separate Dexie databases handle different persistence needs:
|
|
| 464 |
### localStorage Persistence
|
| 465 |
|
| 466 |
Lightweight state persisted across sessions via `createLocalStoragePubSub` pattern:
|
|
|
|
| 467 |
- `settings`: Application preferences (inference type, model, UI options)
|
| 468 |
- `querySuggestions`: Shuffled search suggestion pool
|
| 469 |
- `lastSearchTokenHash`: Cached security token hash
|
|
|
|
| 33 |
## Application Entry Points
|
| 34 |
|
| 35 |
The application has three primary entry points:
|
| 36 |
+
|
| 37 |
1. **Browser Entry**: `client/index.tsx` initializes the React application, mounting the root component and setting up error boundaries.
|
| 38 |
|
| 39 |
2. **Server Entry**: `vite.config.ts` configures the Vite development and preview servers, registering server hooks for search and inference endpoints.
|
|
|
|
| 43 |
## Multi-Service Container Architecture
|
| 44 |
|
| 45 |
The Docker container runs three services concurrently:
|
| 46 |
+
|
| 47 |
- **SearXNG** - Privacy-focused metasearch engine
|
| 48 |
- **ONNX Runtime** - In-process inference for result reranking
|
| 49 |
- **Node.js application** - Main application server
|
|
|
|
| 59 |
## Data Persistence Strategy
|
| 60 |
|
| 61 |
MiniSearch employs a dual-layer persistence approach:
|
| 62 |
+
|
| 63 |
- **IndexedDB** - Local storage for search history, settings, cached results, and saved AI transcripts
|
| 64 |
- **TTL-based caching** - 15-minute freshness cache for search results to minimize API calls, with a 24-hour stale retention as a fallback when the live search fails
|
| 65 |
|
|
|
|
| 115 |
### Web Search Service Reliability
|
| 116 |
|
| 117 |
`server/webSearchService.ts` implements resilience patterns for SearXNG integration:
|
| 118 |
+
|
| 119 |
- **Circuit Breaker**: Opens after 5 consecutive failures, blocking requests for 60 seconds before attempting reset
|
| 120 |
- **Retry Logic**: Exponential backoff for HTTP 500 errors, up to 3 retries
|
| 121 |
- **Fail-Fast on Suspending Engines**: An empty response in which every unresponsive engine is under a long suspension (CAPTCHA, rate limit, access denied; SearXNG suspends them for an hour to a day) throws on the first attempt instead of spending the retry budget, which it could not possibly outlast. Transient engine errors (timeouts, server API errors) keep the full retry budget.
|
|
|
|
| 127 |
### Search Token Lifecycle
|
| 128 |
|
| 129 |
CSRF protection uses a token the server owns for its lifetime:
|
| 130 |
+
|
| 131 |
1. **Generation**: `regenerateSearchToken()` draws 32 random bytes on first use in a process and writes them to the token file; the file is a record, never read back as a source of truth
|
| 132 |
2. **Storage**: The token is held in memory for the life of the process, and recorded at `{os.tempdir()}/minisearch-token` with `0600` permissions
|
| 133 |
3. **Distribution**: Server serves the token to the client at runtime through `/api/config`, so a client always holds the token of the server answering it
|
|
|
|
| 145 |
### State Machine Transitions
|
| 146 |
|
| 147 |
**Text Generation States:**
|
| 148 |
+
|
| 149 |
- `idle` - No active generation
|
| 150 |
- `awaitingModelDownloadAllowance` - Waiting for user consent to download a browser model
|
| 151 |
- `loadingModel` - Downloading or initializing the browser (Wllama) model
|
|
|
|
| 157 |
- `failed` - Error occurred
|
| 158 |
|
| 159 |
**Search States:**
|
| 160 |
+
|
| 161 |
- `idle` - No active search
|
| 162 |
- `running` - Search in progress
|
| 163 |
- `completed` - Results received
|
|
|
|
| 213 |
| `rerankerServiceHook` | `server/rerankerServiceHook.ts` | Reranker model lifecycle management for result reranking |
|
| 214 |
|
| 215 |
Key server-side modules:
|
| 216 |
+
|
| 217 |
- **`server/webSearchService.ts`**: Integrates with SearXNG at `http://127.0.0.1:8888`. Implements a circuit breaker (opens after 5 failures, resets after 60s) and retry logic (up to 3 retries with exponential backoff, for 500s and for empty responses naming transiently unresponsive engines; an all-suspended set fails fast on the first attempt, since SearXNG suspensions last an hour to a day).
|
| 218 |
- **`server/pageContentService.ts`**: Reads result pages for answer grounding: SSRF-guarded fetches with a byte cap, readable-text extraction, and query-relevant passage selection. Reads run under a per-host circuit breaker (`server/pageReadHostBreaker.ts`) that skips a host for 5 minutes after 3 refusals in a row, then probes it with one read.
|
| 219 |
- **`server/searchToken.ts`**: Manages a token at `{os.tempdir()}/minisearch-token` used for CSRF protection on search requests.
|
|
|
|
| 445 |
### IndexedDB Databases
|
| 446 |
|
| 447 |
Two separate Dexie databases handle different persistence needs:
|
| 448 |
+
|
| 449 |
1. **SearchCacheDatabase** (`client/modules/search.ts`): Temporary cache for search results with TTL-based expiration. Table schema:
|
| 450 |
- `textSearchHistory`: Keyed by hashed query, indexed by timestamp
|
| 451 |
- `imageSearchHistory`: Keyed by hashed query, indexed by timestamp
|
|
|
|
| 473 |
### localStorage Persistence
|
| 474 |
|
| 475 |
Lightweight state persisted across sessions via `createLocalStoragePubSub` pattern:
|
| 476 |
+
|
| 477 |
- `settings`: Application preferences (inference type, model, UI options)
|
| 478 |
- `querySuggestions`: Shuffled search suggestion pool
|
| 479 |
- `lastSearchTokenHash`: Cached security token hash
|
docs/project-structure.md
CHANGED
|
@@ -24,6 +24,7 @@
|
|
| 24 |
## Component Organization
|
| 25 |
|
| 26 |
Components are organized by feature domain. Each component folder contains:
|
|
|
|
| 27 |
- Component implementation
|
| 28 |
- Related styles
|
| 29 |
- Custom hooks
|
|
|
|
| 24 |
## Component Organization
|
| 25 |
|
| 26 |
Components are organized by feature domain. Each component folder contains:
|
| 27 |
+
|
| 28 |
- Component implementation
|
| 29 |
- Related styles
|
| 30 |
- Custom hooks
|
docs/pull-requests.md
CHANGED
|
@@ -21,11 +21,13 @@
|
|
| 21 |
## Quality Gates
|
| 22 |
|
| 23 |
Before any change, run this command in the development container:
|
|
|
|
| 24 |
```bash
|
| 25 |
docker compose exec development-server npm run lint
|
| 26 |
```
|
| 27 |
|
| 28 |
This comprehensive check includes:
|
|
|
|
| 29 |
- **Biome**: Code formatting and linting
|
| 30 |
- **TypeScript**: Type checking with strict mode
|
| 31 |
- **knip**: Dead code detection
|
|
@@ -40,12 +42,14 @@ This comprehensive check includes:
|
|
| 40 |
The GitHub Release body is built from `changelog.md`, not from GitHub's generated notes: the publish workflow takes every changelog line that was not there at the previous release tag, under its `## YYYY-MM-DD` heading. A change with no entry therefore ships with no mention in the Release.
|
| 41 |
|
| 42 |
The `changelog-guard` job on every PR to `main` fails when the PR changes a user-facing path and does not touch `changelog.md`. The trigger paths are:
|
|
|
|
| 43 |
- anything under `client/`, `server/`, or `shared/`
|
| 44 |
- `Dockerfile`, `docker-compose.yml`, `docker-compose.production.yml`, `searxng-settings.yml`
|
| 45 |
|
| 46 |
Add your entry to `changelog.md` under a `## YYYY-MM-DD` heading for today's date in UTC, most recent date first.
|
| 47 |
|
| 48 |
Two ways out:
|
|
|
|
| 49 |
- Add the `skip-changelog` label when the change touches a trigger path but is not notable for users - an internal refactor, a rename, a CI-only edit inside a trigger path.
|
| 50 |
- Renovate and Dependabot PRs pass automatically. Renovate bumps the pinned SearXNG commit in the `Dockerfile` and cannot write an entry.
|
| 51 |
|
|
@@ -60,6 +64,7 @@ The three-dot diff compares merge bases, so before the branch has its first comm
|
|
| 60 |
## Merge Philosophy
|
| 61 |
|
| 62 |
In high-throughput agent environments:
|
|
|
|
| 63 |
- Minimize blocking merge gates
|
| 64 |
- Keep PRs short-lived
|
| 65 |
- Corrections are cheap, waiting is expensive
|
|
@@ -69,6 +74,7 @@ In high-throughput agent environments:
|
|
| 69 |
## Commit Message Format
|
| 70 |
|
| 71 |
Use conventional commits:
|
|
|
|
| 72 |
- `feat:` for new features
|
| 73 |
- `fix:` for bug fixes
|
| 74 |
- `docs:` for documentation changes
|
|
|
|
| 21 |
## Quality Gates
|
| 22 |
|
| 23 |
Before any change, run this command in the development container:
|
| 24 |
+
|
| 25 |
```bash
|
| 26 |
docker compose exec development-server npm run lint
|
| 27 |
```
|
| 28 |
|
| 29 |
This comprehensive check includes:
|
| 30 |
+
|
| 31 |
- **Biome**: Code formatting and linting
|
| 32 |
- **TypeScript**: Type checking with strict mode
|
| 33 |
- **knip**: Dead code detection
|
|
|
|
| 42 |
The GitHub Release body is built from `changelog.md`, not from GitHub's generated notes: the publish workflow takes every changelog line that was not there at the previous release tag, under its `## YYYY-MM-DD` heading. A change with no entry therefore ships with no mention in the Release.
|
| 43 |
|
| 44 |
The `changelog-guard` job on every PR to `main` fails when the PR changes a user-facing path and does not touch `changelog.md`. The trigger paths are:
|
| 45 |
+
|
| 46 |
- anything under `client/`, `server/`, or `shared/`
|
| 47 |
- `Dockerfile`, `docker-compose.yml`, `docker-compose.production.yml`, `searxng-settings.yml`
|
| 48 |
|
| 49 |
Add your entry to `changelog.md` under a `## YYYY-MM-DD` heading for today's date in UTC, most recent date first.
|
| 50 |
|
| 51 |
Two ways out:
|
| 52 |
+
|
| 53 |
- Add the `skip-changelog` label when the change touches a trigger path but is not notable for users - an internal refactor, a rename, a CI-only edit inside a trigger path.
|
| 54 |
- Renovate and Dependabot PRs pass automatically. Renovate bumps the pinned SearXNG commit in the `Dockerfile` and cannot write an entry.
|
| 55 |
|
|
|
|
| 64 |
## Merge Philosophy
|
| 65 |
|
| 66 |
In high-throughput agent environments:
|
| 67 |
+
|
| 68 |
- Minimize blocking merge gates
|
| 69 |
- Keep PRs short-lived
|
| 70 |
- Corrections are cheap, waiting is expensive
|
|
|
|
| 74 |
## Commit Message Format
|
| 75 |
|
| 76 |
Use conventional commits:
|
| 77 |
+
|
| 78 |
- `feat:` for new features
|
| 79 |
- `fix:` for bug fixes
|
| 80 |
- `docs:` for documentation changes
|
docs/quick-start.md
CHANGED
|
@@ -23,6 +23,7 @@ docker compose up
|
|
| 23 |
Access the application at `http://localhost:7860` (main). During development, HMR is available on `http://localhost:7861` for instant code changes.
|
| 24 |
|
| 25 |
The development server includes:
|
|
|
|
| 26 |
- Hot Module Replacement (HMR) on port 7861 for instant code changes
|
| 27 |
- Full dev tools and source maps
|
| 28 |
- Live code watching with volume mounts
|
|
@@ -39,6 +40,7 @@ docker compose -f docker-compose.production.yml up --build
|
|
| 39 |
Access at `http://localhost:7860`
|
| 40 |
|
| 41 |
Production mode:
|
|
|
|
| 42 |
- Pre-built optimized assets
|
| 43 |
- No dev tools or HMR
|
| 44 |
- Optimized Docker layer caching
|
|
@@ -46,6 +48,7 @@ Production mode:
|
|
| 46 |
## Container Internals
|
| 47 |
|
| 48 |
The Docker container runs three services concurrently:
|
|
|
|
| 49 |
- **SearXNG** - Privacy-focused metasearch engine (starts in background)
|
| 50 |
- **ONNX Runtime** - In-process inference for result reranking
|
| 51 |
- **Node.js (Vite)** - Main application server
|
|
@@ -139,6 +142,7 @@ Access at `http://localhost:7860`
|
|
| 139 |
### Issue: Search returns no results
|
| 140 |
|
| 141 |
**Solution**: Verify SearXNG is running. Check container logs:
|
|
|
|
| 142 |
```bash
|
| 143 |
docker compose logs searxng
|
| 144 |
```
|
|
@@ -146,6 +150,7 @@ docker compose logs searxng
|
|
| 146 |
### Issue: AI response never loads
|
| 147 |
|
| 148 |
**Solution**: Check browser console for errors. Common causes:
|
|
|
|
| 149 |
- WebGPU not supported (app will automatically fall back to Wllama)
|
| 150 |
- Model download blocked by firewall or CORS
|
| 151 |
- Insufficient disk space for model caching
|
|
@@ -159,6 +164,7 @@ docker compose logs searxng
|
|
| 159 |
### Issue: Port already in use
|
| 160 |
|
| 161 |
**Solution**: Change ports in `docker-compose.yml`:
|
|
|
|
| 162 |
```yaml
|
| 163 |
ports:
|
| 164 |
- "7862:7861" # Use 7862 instead of 7860 (HMR port)
|
|
|
|
| 23 |
Access the application at `http://localhost:7860` (main). During development, HMR is available on `http://localhost:7861` for instant code changes.
|
| 24 |
|
| 25 |
The development server includes:
|
| 26 |
+
|
| 27 |
- Hot Module Replacement (HMR) on port 7861 for instant code changes
|
| 28 |
- Full dev tools and source maps
|
| 29 |
- Live code watching with volume mounts
|
|
|
|
| 40 |
Access at `http://localhost:7860`
|
| 41 |
|
| 42 |
Production mode:
|
| 43 |
+
|
| 44 |
- Pre-built optimized assets
|
| 45 |
- No dev tools or HMR
|
| 46 |
- Optimized Docker layer caching
|
|
|
|
| 48 |
## Container Internals
|
| 49 |
|
| 50 |
The Docker container runs three services concurrently:
|
| 51 |
+
|
| 52 |
- **SearXNG** - Privacy-focused metasearch engine (starts in background)
|
| 53 |
- **ONNX Runtime** - In-process inference for result reranking
|
| 54 |
- **Node.js (Vite)** - Main application server
|
|
|
|
| 142 |
### Issue: Search returns no results
|
| 143 |
|
| 144 |
**Solution**: Verify SearXNG is running. Check container logs:
|
| 145 |
+
|
| 146 |
```bash
|
| 147 |
docker compose logs searxng
|
| 148 |
```
|
|
|
|
| 150 |
### Issue: AI response never loads
|
| 151 |
|
| 152 |
**Solution**: Check browser console for errors. Common causes:
|
| 153 |
+
|
| 154 |
- WebGPU not supported (app will automatically fall back to Wllama)
|
| 155 |
- Model download blocked by firewall or CORS
|
| 156 |
- Insufficient disk space for model caching
|
|
|
|
| 164 |
### Issue: Port already in use
|
| 165 |
|
| 166 |
**Solution**: Change ports in `docker-compose.yml`:
|
| 167 |
+
|
| 168 |
```yaml
|
| 169 |
ports:
|
| 170 |
- "7862:7861" # Use 7862 instead of 7860 (HMR port)
|
docs/reranking.md
CHANGED
|
@@ -18,6 +18,7 @@ The reranking subsystem consists of four components:
|
|
| 18 |
### Startup
|
| 19 |
|
| 20 |
The `rerankerServiceHook` starts the reranker during server initialization:
|
|
|
|
| 21 |
1. Downloads the model and tokenizer from HuggingFace if not present (`cross-encoder/mmarco-mMiniLMv2-L12-H384-v1`)
|
| 22 |
2. Creates an ONNX Runtime inference session
|
| 23 |
3. Performs a warmup inference (`query: "test"`, `documents: ["test document"]`) to ensure the graph is initialized
|
|
@@ -79,6 +80,7 @@ Documents are sent to the reranker whole. Truncation happens by tokens inside th
|
|
| 79 |
### Scoring and Filtering
|
| 80 |
|
| 81 |
The reranker returns the classifier's raw relevance logit for each document. Scores are deliberately not passed through a sigmoid, because the filter below is calibrated against the raw scale. Results are then filtered using a two-stage statistical approach:
|
|
|
|
| 82 |
1. **Score Normalization**: Scores are shifted to positive range by adding the absolute value of the minimum score
|
| 83 |
2. **Standard Deviation Filter**: Results below `mean - kStandardDeviationFactor * standardDeviation` are filtered out
|
| 84 |
- `kStandardDeviationFactor = 0.3`
|
|
@@ -90,6 +92,7 @@ This filter is invariant to linear rescaling of the scores, since both the score
|
|
| 90 |
### Preserve Top Results Mode
|
| 91 |
|
| 92 |
When `preserveTopResults = true`, the ranking algorithm:
|
|
|
|
| 93 |
1. Keeps the original top result (first from SearXNG) at position 1
|
| 94 |
2. Filters remaining results by score
|
| 95 |
3. Takes up to 9 next-best results, sorted by reranker score
|
|
|
|
| 18 |
### Startup
|
| 19 |
|
| 20 |
The `rerankerServiceHook` starts the reranker during server initialization:
|
| 21 |
+
|
| 22 |
1. Downloads the model and tokenizer from HuggingFace if not present (`cross-encoder/mmarco-mMiniLMv2-L12-H384-v1`)
|
| 23 |
2. Creates an ONNX Runtime inference session
|
| 24 |
3. Performs a warmup inference (`query: "test"`, `documents: ["test document"]`) to ensure the graph is initialized
|
|
|
|
| 80 |
### Scoring and Filtering
|
| 81 |
|
| 82 |
The reranker returns the classifier's raw relevance logit for each document. Scores are deliberately not passed through a sigmoid, because the filter below is calibrated against the raw scale. Results are then filtered using a two-stage statistical approach:
|
| 83 |
+
|
| 84 |
1. **Score Normalization**: Scores are shifted to positive range by adding the absolute value of the minimum score
|
| 85 |
2. **Standard Deviation Filter**: Results below `mean - kStandardDeviationFactor * standardDeviation` are filtered out
|
| 86 |
- `kStandardDeviationFactor = 0.3`
|
|
|
|
| 92 |
### Preserve Top Results Mode
|
| 93 |
|
| 94 |
When `preserveTopResults = true`, the ranking algorithm:
|
| 95 |
+
|
| 96 |
1. Keeps the original top result (first from SearXNG) at position 1
|
| 97 |
2. Filters remaining results by score
|
| 98 |
3. Takes up to 9 next-best results, sorted by reranker score
|
docs/search-history.md
CHANGED
|
@@ -7,6 +7,7 @@ MiniSearch keeps a local-first record of every search so agents can resurface te
|
|
| 7 |
## Data Model and Storage
|
| 8 |
|
| 9 |
The `HistoryDatabase` Dexie instance stores three related tables:
|
|
|
|
| 10 |
1. `searches` โ canonical log of each query plus hydrated results payloads (text, image, run metadata, source, pin state).
|
| 11 |
2. `llmResponses` โ snapshot of every AI answer tied to its originating search run, enabling restoration of the assistant output.
|
| 12 |
3. `chatHistory` โ chronological record of chat turns (user/assistant roles) scoped by `conversationId` (the `searchRunId`).
|
|
@@ -16,6 +17,7 @@ All three tables share a `searchRunId` foreign key so restoring a specific run c
|
|
| 16 |
## Cleanup Guarantees
|
| 17 |
|
| 18 |
To keep IndexedDB lean, inserts trigger `performCleanup()`, which enforces:
|
|
|
|
| 19 |
- **Retention window** โ non-pinned entries older than `historyRetentionDays` are deleted in batches of 100.
|
| 20 |
- **Max entries** โ if the total exceeds `historyMaxEntries`, the oldest unpinned rows past the limit are removed.
|
| 21 |
- **Pin protection** โ pinned entries are skipped by both retention and max-entry sweeps.
|
|
@@ -25,6 +27,7 @@ Cleanup is gated by `historyAutoCleanup`, so advanced users can opt out. These k
|
|
| 25 |
## Hook API
|
| 26 |
|
| 27 |
`useSearchHistory` exposes the canonical API surface for components:
|
|
|
|
| 28 |
- Fetches recent searches (optionally paginated) and exposes filtered/grouped views.
|
| 29 |
- Provides fuzzy search, pin toggling, delete/clear operations, and error retry helpers.
|
| 30 |
- Keeps derived state such as grouped sections (Today/Yesterday/etc.) in sync with the settings flag `historyGroupByDate`.
|
|
@@ -51,12 +54,14 @@ The History Drawer Analytics tab renders the compact `SearchStats` cards scoped
|
|
| 51 |
Alongside the history database, MiniSearch uses a separate `SearchCacheDatabase` for temporary result caching:
|
| 52 |
|
| 53 |
**Schema:**
|
|
|
|
| 54 |
| Store | Key | Index | Entry Type |
|
| 55 |
| -------------------- | ------------ | --------- | ------------------ |
|
| 56 |
| `textSearchHistory` | hashed query | timestamp | `TextSearchCache` |
|
| 57 |
| `imageSearchHistory` | hashed query | timestamp | `ImageSearchCache` |
|
| 58 |
|
| 59 |
**Configuration:**
|
|
|
|
| 60 |
| Parameter | Value | Description |
|
| 61 |
| ------------------- | ---------- | ------------------------------------------------------------------------------------- |
|
| 62 |
| TTL | 15 minutes | Freshness window: a hit inside it skips the network |
|
|
|
|
| 7 |
## Data Model and Storage
|
| 8 |
|
| 9 |
The `HistoryDatabase` Dexie instance stores three related tables:
|
| 10 |
+
|
| 11 |
1. `searches` โ canonical log of each query plus hydrated results payloads (text, image, run metadata, source, pin state).
|
| 12 |
2. `llmResponses` โ snapshot of every AI answer tied to its originating search run, enabling restoration of the assistant output.
|
| 13 |
3. `chatHistory` โ chronological record of chat turns (user/assistant roles) scoped by `conversationId` (the `searchRunId`).
|
|
|
|
| 17 |
## Cleanup Guarantees
|
| 18 |
|
| 19 |
To keep IndexedDB lean, inserts trigger `performCleanup()`, which enforces:
|
| 20 |
+
|
| 21 |
- **Retention window** โ non-pinned entries older than `historyRetentionDays` are deleted in batches of 100.
|
| 22 |
- **Max entries** โ if the total exceeds `historyMaxEntries`, the oldest unpinned rows past the limit are removed.
|
| 23 |
- **Pin protection** โ pinned entries are skipped by both retention and max-entry sweeps.
|
|
|
|
| 27 |
## Hook API
|
| 28 |
|
| 29 |
`useSearchHistory` exposes the canonical API surface for components:
|
| 30 |
+
|
| 31 |
- Fetches recent searches (optionally paginated) and exposes filtered/grouped views.
|
| 32 |
- Provides fuzzy search, pin toggling, delete/clear operations, and error retry helpers.
|
| 33 |
- Keeps derived state such as grouped sections (Today/Yesterday/etc.) in sync with the settings flag `historyGroupByDate`.
|
|
|
|
| 54 |
Alongside the history database, MiniSearch uses a separate `SearchCacheDatabase` for temporary result caching:
|
| 55 |
|
| 56 |
**Schema:**
|
| 57 |
+
|
| 58 |
| Store | Key | Index | Entry Type |
|
| 59 |
| -------------------- | ------------ | --------- | ------------------ |
|
| 60 |
| `textSearchHistory` | hashed query | timestamp | `TextSearchCache` |
|
| 61 |
| `imageSearchHistory` | hashed query | timestamp | `ImageSearchCache` |
|
| 62 |
|
| 63 |
**Configuration:**
|
| 64 |
+
|
| 65 |
| Parameter | Value | Description |
|
| 66 |
| ------------------- | ---------- | ------------------------------------------------------------------------------------- |
|
| 67 |
| TTL | 15 minutes | Freshness window: a hit inside it skips the network |
|
docs/security.md
CHANGED
|
@@ -32,6 +32,7 @@ can reach the instance, so keep secrets out of that interface.
|
|
| 32 |
### Search Token Lifecycle
|
| 33 |
|
| 34 |
Every HTTP request from client to backend carries a `token` query parameter for CSRF protection:
|
|
|
|
| 35 |
1. **Token Generation**: On first use in a process, `regenerateSearchToken()` draws 32 random bytes and writes them to `{os.tempdir()}/minisearch-token` with `0600`. The file is never read back as a source of truth: a token that survives into a published image is one every container of that build shares, and anyone who pulls the image can read it out of the layer
|
| 36 |
2. **Client Distribution**: The server holds the token it generated for the life of the process and serves it as `searchToken` in `/api/config`. Each process generates its own, so a restart or a second instance means a new token, which is why a bookmarked or shared search URL stops working and lands on the expired-link page
|
| 37 |
3. **Per-Request Auth**: Client includes token as `?token=` parameter on all `/search/text`, `/search/images`, `/page-content`, `/thumbnail` and `/inference` requests
|
|
@@ -65,6 +66,7 @@ Every HTTP request from client to backend carries a `token` query parameter for
|
|
| 65 |
## Supply Chain: Native & WebAssembly Modules
|
| 66 |
|
| 67 |
The full trust model lives in [`.github/SECURITY.md`](../.github/SECURITY.md) ("Native & WebAssembly Module Trust Model"); the operational parts are mirrored here.
|
|
|
|
| 68 |
- **Exact pins**: the native/WASM dependencies โ `onnxruntime-node@1.29.0`, `@wllama/wllama@3.6.1`, `@huggingface/tokenizers@0.2.0`, `hash-wasm@4.12.0` and `@moonshine-ai/moonshine-wasm@0.1.5` โ are declared with single exact versions in `package.json` (no caret ranges), and `package-lock.json` pins each tarball's SHA-512 integrity hash
|
| 69 |
- **Install-script policy**: `package.json#allowScripts` carries an explicit decision for every package in the tree that ships an install script โ `onnxruntime-node` **denied** (its postinstall downloads CUDA 12 nupkgs unpacked by symlink-following `adm-zip`, GHSA-vwc7-r8mq-g2x9; both inference services run `executionProviders: ["cpu"]` and the CPU runtime ships in the tarball), `protobufjs` **denied** (its postinstall only prints a version-scheme advisory), `fsevents@2.3.3` **allowed** (the optional macOS file watcher must compile at install)
|
| 70 |
- **Strict mode**: `.npmrc` sets `strict-allow-scripts = true`, so an install script with no recorded decision fails `npm ci` instead of warning; the `.npmrc` layer itself can only allow, never deny, which is why the policy lives in `package.json`
|
|
|
|
| 32 |
### Search Token Lifecycle
|
| 33 |
|
| 34 |
Every HTTP request from client to backend carries a `token` query parameter for CSRF protection:
|
| 35 |
+
|
| 36 |
1. **Token Generation**: On first use in a process, `regenerateSearchToken()` draws 32 random bytes and writes them to `{os.tempdir()}/minisearch-token` with `0600`. The file is never read back as a source of truth: a token that survives into a published image is one every container of that build shares, and anyone who pulls the image can read it out of the layer
|
| 37 |
2. **Client Distribution**: The server holds the token it generated for the life of the process and serves it as `searchToken` in `/api/config`. Each process generates its own, so a restart or a second instance means a new token, which is why a bookmarked or shared search URL stops working and lands on the expired-link page
|
| 38 |
3. **Per-Request Auth**: Client includes token as `?token=` parameter on all `/search/text`, `/search/images`, `/page-content`, `/thumbnail` and `/inference` requests
|
|
|
|
| 66 |
## Supply Chain: Native & WebAssembly Modules
|
| 67 |
|
| 68 |
The full trust model lives in [`.github/SECURITY.md`](../.github/SECURITY.md) ("Native & WebAssembly Module Trust Model"); the operational parts are mirrored here.
|
| 69 |
+
|
| 70 |
- **Exact pins**: the native/WASM dependencies โ `onnxruntime-node@1.29.0`, `@wllama/wllama@3.6.1`, `@huggingface/tokenizers@0.2.0`, `hash-wasm@4.12.0` and `@moonshine-ai/moonshine-wasm@0.1.5` โ are declared with single exact versions in `package.json` (no caret ranges), and `package-lock.json` pins each tarball's SHA-512 integrity hash
|
| 71 |
- **Install-script policy**: `package.json#allowScripts` carries an explicit decision for every package in the tree that ships an install script โ `onnxruntime-node` **denied** (its postinstall downloads CUDA 12 nupkgs unpacked by symlink-following `adm-zip`, GHSA-vwc7-r8mq-g2x9; both inference services run `executionProviders: ["cpu"]` and the CPU runtime ships in the tarball), `protobufjs` **denied** (its postinstall only prints a version-scheme advisory), `fsevents@2.3.3` **allowed** (the optional macOS file watcher must compile at install)
|
| 72 |
- **Strict mode**: `.npmrc` sets `strict-allow-scripts = true`, so an install script with no recorded decision fails `npm ci` instead of warning; the `.npmrc` layer itself can only allow, never deny, which is why the policy lives in `package.json`
|
docs/ui-components.md
CHANGED
|
@@ -27,6 +27,7 @@ updateQuery("new query");
|
|
| 27 |
```
|
| 28 |
|
| 29 |
**Benefits:**
|
|
|
|
| 30 |
- **Decoupling** โ Modules (text generation, search, React components) read and write shared state without importing each other directly
|
| 31 |
- **No provider boilerplate** โ Unlike React Context or Redux, no `Provider` wrapper needed; any module imports a channel from `pubSub.ts`
|
| 32 |
- **Selective subscriptions** โ Components subscribe only to channels they use; a streaming token update throttling `responsePubSub` does not trigger re-renders in unrelated components
|
|
@@ -35,6 +36,7 @@ updateQuery("new query");
|
|
| 35 |
### localStorage Persistence
|
| 36 |
|
| 37 |
Some state must survive page reloads. The `createLocalStoragePubSub` helper wraps `createPubSub` with two behaviors:
|
|
|
|
| 38 |
1. **Hydration** โ On first call, reads existing value from `localStorage` via `localStorage.getItem(key)`. If a stored JSON string is found, it is parsed and used as the initial value; otherwise the default value is used
|
| 39 |
2. **Persistence** โ A subscriber is immediately registered on the inner PubSub that calls `localStorage.setItem` with the JSON-serialized new value on every state change
|
| 40 |
|
|
@@ -133,6 +135,7 @@ App
|
|
| 133 |
**Responsibility:** Application shell and routing
|
| 134 |
|
| 135 |
**Logic:**
|
|
|
|
| 136 |
```typescript
|
| 137 |
// App.tsx
|
| 138 |
// Access key validation is handled via useAccessKeyValidation hook
|
|
@@ -147,6 +150,7 @@ App
|
|
| 147 |
**Responsibility:** Query input and search initiation
|
| 148 |
|
| 149 |
**PubSub:**
|
|
|
|
| 150 |
- **Subscribes:** `queryPubSub`, `textSearchStatePubSub`
|
| 151 |
- **Updates:** `queryPubSub` (on type), triggers `searchAndRespond()` (on submit)
|
| 152 |
|
|
@@ -155,6 +159,7 @@ The query field is wrapped in its own `position: relative` element so
|
|
| 155 |
when the row of buttons below the field changes height.
|
| 156 |
|
| 157 |
**Logic:**
|
|
|
|
| 158 |
```typescript
|
| 159 |
function SearchForm() {
|
| 160 |
const [query, setQuery] = usePubSub(queryPubSub);
|
|
@@ -185,6 +190,7 @@ dictation session: the engine choice, the microphone lifecycle, and the splice
|
|
| 185 |
that appends each transcript without clobbering what the user typed meanwhile.
|
| 186 |
|
| 187 |
**PubSub:**
|
|
|
|
| 188 |
- **Subscribes:** `settingsPubSub` (`enableDictation`, `enableLocalDictationModel`)
|
| 189 |
|
| 190 |
**Props:**
|
|
@@ -208,9 +214,11 @@ effect never runs.
|
|
| 208 |
**Responsibility:** Display search results (text and images)
|
| 209 |
|
| 210 |
**PubSub:**
|
|
|
|
| 211 |
- **Subscribes:** `textSearchResultsPubSub`, `imageSearchResultsPubSub`, `textSearchStatePubSub`
|
| 212 |
|
| 213 |
**Logic:**
|
|
|
|
| 214 |
```typescript
|
| 215 |
function SearchResultsSection() {
|
| 216 |
const [textResults] = usePubSub(textSearchResultsPubSub);
|
|
@@ -234,6 +242,7 @@ function SearchResultsSection() {
|
|
| 234 |
**Responsibility:** AI response display and chat interface
|
| 235 |
|
| 236 |
**PubSub:**
|
|
|
|
| 237 |
- **Subscribes:** `responsePubSub`, `textGenerationStatePubSub`, `chatMessagesPubSub`
|
| 238 |
|
| 239 |
**State Machine (`textGenerationStatePubSub`):**
|
|
@@ -255,6 +264,7 @@ function SearchResultsSection() {
|
|
| 255 |
When models output internal thought processes, the UI extracts reasoning content bounded by `reasoningStartMarker` and `reasoningEndMarker` markers. Reasoning is displayed separately from the final response in a collapsible section.
|
| 256 |
|
| 257 |
**Logic:**
|
|
|
|
| 258 |
```typescript
|
| 259 |
function AiResponseSection() {
|
| 260 |
const [response] = usePubSub(responsePubSub);
|
|
@@ -289,6 +299,7 @@ function AiResponseSection() {
|
|
| 289 |
**Responsibility:** Application settings UI
|
| 290 |
|
| 291 |
**Sub-components:**
|
|
|
|
| 292 |
- **AISettingsForm:** Model selection, inference type, reasoning markers (sampling parameters such as temperature are hardcoded, not user-configurable)
|
| 293 |
- **SearchSettingsForm:** Result limits, image search toggle
|
| 294 |
- **InterfaceSettingsForm:** UI preferences
|
|
@@ -297,9 +308,11 @@ function AiResponseSection() {
|
|
| 297 |
- **ActionsForm:** Data management actions
|
| 298 |
|
| 299 |
**PubSub:**
|
|
|
|
| 300 |
- **Subscribes/Updates:** `settingsPubSub` (full settings object)
|
| 301 |
|
| 302 |
**Persistence:**
|
|
|
|
| 303 |
```typescript
|
| 304 |
// client/modules/pubSub.ts
|
| 305 |
export const settingsPubSub = createLocalStoragePubSub(
|
|
@@ -315,9 +328,11 @@ export const settingsPubSub = createLocalStoragePubSub(
|
|
| 315 |
**Responsibility:** Search history display and management
|
| 316 |
|
| 317 |
**PubSub:**
|
|
|
|
| 318 |
- **Subscribes:** History loaded from IndexedDB (not via PubSub, via custom hook)
|
| 319 |
|
| 320 |
**Hook:** `useSearchHistory()`
|
|
|
|
| 321 |
```typescript
|
| 322 |
const {
|
| 323 |
filteredSearches,
|
|
@@ -329,11 +344,13 @@ const {
|
|
| 329 |
```
|
| 330 |
|
| 331 |
Restoring a past search (re-running its query) is handled separately by `useHistoryRestore`, used in `SearchForm`:
|
|
|
|
| 332 |
```typescript
|
| 333 |
const { restoreSearch } = useHistoryRestore(updateQuery, textAreaRef);
|
| 334 |
```
|
| 335 |
|
| 336 |
**Features:**
|
|
|
|
| 337 |
- Fuzzy search through history
|
| 338 |
- Date-based grouping (Today, Yesterday, Last Week, etc.)
|
| 339 |
- Pin/unpin searches
|
|
@@ -419,6 +436,7 @@ Saved to IndexedDB
|
|
| 419 |
### usePubSub
|
| 420 |
|
| 421 |
Subscribes to a PubSub channel:
|
|
|
|
| 422 |
```typescript
|
| 423 |
const [value, setValue] = usePubSub(channel);
|
| 424 |
```
|
|
@@ -426,6 +444,7 @@ const [value, setValue] = usePubSub(channel);
|
|
| 426 |
### useSearchHistory
|
| 427 |
|
| 428 |
Manages search history from IndexedDB:
|
|
|
|
| 429 |
```typescript
|
| 430 |
const { searchHistory, groupedSearches, deleteEntry, togglePin } =
|
| 431 |
useSearchHistory();
|
|
@@ -434,6 +453,7 @@ const { searchHistory, groupedSearches, deleteEntry, togglePin } =
|
|
| 434 |
### useDrawerState
|
| 435 |
|
| 436 |
Manages open/close state for a drawer with log entry tracking:
|
|
|
|
| 437 |
```typescript
|
| 438 |
const { isDrawerOpen, openDrawer, closeDrawer } = useDrawerState(
|
| 439 |
"User opened the menu",
|
|
@@ -446,6 +466,7 @@ const { isDrawerOpen, openDrawer, closeDrawer } = useDrawerState(
|
|
| 446 |
Holds the [Screen Wake Lock](https://developer.mozilla.org/en-US/docs/Web/API/Screen_Wake_Lock_API)
|
| 447 |
while the flag is true, so a phone left untouched during a long answer does not
|
| 448 |
lock while the text streams in:
|
|
|
|
| 449 |
```typescript
|
| 450 |
useScreenWakeLock(statesKeepingTheScreenAwake.includes(textGenerationState));
|
| 451 |
```
|
|
@@ -460,12 +481,14 @@ lock and no error.
|
|
| 460 |
**Framework:** Mantine UI v9
|
| 461 |
|
| 462 |
**Theme Configuration:**
|
|
|
|
| 463 |
```typescript
|
| 464 |
// client/components/App/App.tsx
|
| 465 |
<MantineProvider defaultColorScheme="dark">
|
| 466 |
```
|
| 467 |
|
| 468 |
**Dark Mode:**
|
|
|
|
| 469 |
- Default color scheme is dark
|
| 470 |
- All components support dark mode via Mantine
|
| 471 |
|
|
@@ -474,12 +497,14 @@ lock and no error.
|
|
| 474 |
**Standards:** WCAG 2.1 AA compliance
|
| 475 |
|
| 476 |
**Features:**
|
|
|
|
| 477 |
- All interactive elements keyboard accessible
|
| 478 |
- ARIA labels on select interactive elements (e.g. chat input, message list, history actions, logs modal)
|
| 479 |
- Focus management in drawers and modals
|
| 480 |
- Screen reader announcements for loading states
|
| 481 |
|
| 482 |
**Implementation:**
|
|
|
|
| 483 |
```typescript
|
| 484 |
// Using Mantine's accessibility props
|
| 485 |
<Button aria-label="Search the web">
|
|
|
|
| 27 |
```
|
| 28 |
|
| 29 |
**Benefits:**
|
| 30 |
+
|
| 31 |
- **Decoupling** โ Modules (text generation, search, React components) read and write shared state without importing each other directly
|
| 32 |
- **No provider boilerplate** โ Unlike React Context or Redux, no `Provider` wrapper needed; any module imports a channel from `pubSub.ts`
|
| 33 |
- **Selective subscriptions** โ Components subscribe only to channels they use; a streaming token update throttling `responsePubSub` does not trigger re-renders in unrelated components
|
|
|
|
| 36 |
### localStorage Persistence
|
| 37 |
|
| 38 |
Some state must survive page reloads. The `createLocalStoragePubSub` helper wraps `createPubSub` with two behaviors:
|
| 39 |
+
|
| 40 |
1. **Hydration** โ On first call, reads existing value from `localStorage` via `localStorage.getItem(key)`. If a stored JSON string is found, it is parsed and used as the initial value; otherwise the default value is used
|
| 41 |
2. **Persistence** โ A subscriber is immediately registered on the inner PubSub that calls `localStorage.setItem` with the JSON-serialized new value on every state change
|
| 42 |
|
|
|
|
| 135 |
**Responsibility:** Application shell and routing
|
| 136 |
|
| 137 |
**Logic:**
|
| 138 |
+
|
| 139 |
```typescript
|
| 140 |
// App.tsx
|
| 141 |
// Access key validation is handled via useAccessKeyValidation hook
|
|
|
|
| 150 |
**Responsibility:** Query input and search initiation
|
| 151 |
|
| 152 |
**PubSub:**
|
| 153 |
+
|
| 154 |
- **Subscribes:** `queryPubSub`, `textSearchStatePubSub`
|
| 155 |
- **Updates:** `queryPubSub` (on type), triggers `searchAndRespond()` (on submit)
|
| 156 |
|
|
|
|
| 159 |
when the row of buttons below the field changes height.
|
| 160 |
|
| 161 |
**Logic:**
|
| 162 |
+
|
| 163 |
```typescript
|
| 164 |
function SearchForm() {
|
| 165 |
const [query, setQuery] = usePubSub(queryPubSub);
|
|
|
|
| 190 |
that appends each transcript without clobbering what the user typed meanwhile.
|
| 191 |
|
| 192 |
**PubSub:**
|
| 193 |
+
|
| 194 |
- **Subscribes:** `settingsPubSub` (`enableDictation`, `enableLocalDictationModel`)
|
| 195 |
|
| 196 |
**Props:**
|
|
|
|
| 214 |
**Responsibility:** Display search results (text and images)
|
| 215 |
|
| 216 |
**PubSub:**
|
| 217 |
+
|
| 218 |
- **Subscribes:** `textSearchResultsPubSub`, `imageSearchResultsPubSub`, `textSearchStatePubSub`
|
| 219 |
|
| 220 |
**Logic:**
|
| 221 |
+
|
| 222 |
```typescript
|
| 223 |
function SearchResultsSection() {
|
| 224 |
const [textResults] = usePubSub(textSearchResultsPubSub);
|
|
|
|
| 242 |
**Responsibility:** AI response display and chat interface
|
| 243 |
|
| 244 |
**PubSub:**
|
| 245 |
+
|
| 246 |
- **Subscribes:** `responsePubSub`, `textGenerationStatePubSub`, `chatMessagesPubSub`
|
| 247 |
|
| 248 |
**State Machine (`textGenerationStatePubSub`):**
|
|
|
|
| 264 |
When models output internal thought processes, the UI extracts reasoning content bounded by `reasoningStartMarker` and `reasoningEndMarker` markers. Reasoning is displayed separately from the final response in a collapsible section.
|
| 265 |
|
| 266 |
**Logic:**
|
| 267 |
+
|
| 268 |
```typescript
|
| 269 |
function AiResponseSection() {
|
| 270 |
const [response] = usePubSub(responsePubSub);
|
|
|
|
| 299 |
**Responsibility:** Application settings UI
|
| 300 |
|
| 301 |
**Sub-components:**
|
| 302 |
+
|
| 303 |
- **AISettingsForm:** Model selection, inference type, reasoning markers (sampling parameters such as temperature are hardcoded, not user-configurable)
|
| 304 |
- **SearchSettingsForm:** Result limits, image search toggle
|
| 305 |
- **InterfaceSettingsForm:** UI preferences
|
|
|
|
| 308 |
- **ActionsForm:** Data management actions
|
| 309 |
|
| 310 |
**PubSub:**
|
| 311 |
+
|
| 312 |
- **Subscribes/Updates:** `settingsPubSub` (full settings object)
|
| 313 |
|
| 314 |
**Persistence:**
|
| 315 |
+
|
| 316 |
```typescript
|
| 317 |
// client/modules/pubSub.ts
|
| 318 |
export const settingsPubSub = createLocalStoragePubSub(
|
|
|
|
| 328 |
**Responsibility:** Search history display and management
|
| 329 |
|
| 330 |
**PubSub:**
|
| 331 |
+
|
| 332 |
- **Subscribes:** History loaded from IndexedDB (not via PubSub, via custom hook)
|
| 333 |
|
| 334 |
**Hook:** `useSearchHistory()`
|
| 335 |
+
|
| 336 |
```typescript
|
| 337 |
const {
|
| 338 |
filteredSearches,
|
|
|
|
| 344 |
```
|
| 345 |
|
| 346 |
Restoring a past search (re-running its query) is handled separately by `useHistoryRestore`, used in `SearchForm`:
|
| 347 |
+
|
| 348 |
```typescript
|
| 349 |
const { restoreSearch } = useHistoryRestore(updateQuery, textAreaRef);
|
| 350 |
```
|
| 351 |
|
| 352 |
**Features:**
|
| 353 |
+
|
| 354 |
- Fuzzy search through history
|
| 355 |
- Date-based grouping (Today, Yesterday, Last Week, etc.)
|
| 356 |
- Pin/unpin searches
|
|
|
|
| 436 |
### usePubSub
|
| 437 |
|
| 438 |
Subscribes to a PubSub channel:
|
| 439 |
+
|
| 440 |
```typescript
|
| 441 |
const [value, setValue] = usePubSub(channel);
|
| 442 |
```
|
|
|
|
| 444 |
### useSearchHistory
|
| 445 |
|
| 446 |
Manages search history from IndexedDB:
|
| 447 |
+
|
| 448 |
```typescript
|
| 449 |
const { searchHistory, groupedSearches, deleteEntry, togglePin } =
|
| 450 |
useSearchHistory();
|
|
|
|
| 453 |
### useDrawerState
|
| 454 |
|
| 455 |
Manages open/close state for a drawer with log entry tracking:
|
| 456 |
+
|
| 457 |
```typescript
|
| 458 |
const { isDrawerOpen, openDrawer, closeDrawer } = useDrawerState(
|
| 459 |
"User opened the menu",
|
|
|
|
| 466 |
Holds the [Screen Wake Lock](https://developer.mozilla.org/en-US/docs/Web/API/Screen_Wake_Lock_API)
|
| 467 |
while the flag is true, so a phone left untouched during a long answer does not
|
| 468 |
lock while the text streams in:
|
| 469 |
+
|
| 470 |
```typescript
|
| 471 |
useScreenWakeLock(statesKeepingTheScreenAwake.includes(textGenerationState));
|
| 472 |
```
|
|
|
|
| 481 |
**Framework:** Mantine UI v9
|
| 482 |
|
| 483 |
**Theme Configuration:**
|
| 484 |
+
|
| 485 |
```typescript
|
| 486 |
// client/components/App/App.tsx
|
| 487 |
<MantineProvider defaultColorScheme="dark">
|
| 488 |
```
|
| 489 |
|
| 490 |
**Dark Mode:**
|
| 491 |
+
|
| 492 |
- Default color scheme is dark
|
| 493 |
- All components support dark mode via Mantine
|
| 494 |
|
|
|
|
| 497 |
**Standards:** WCAG 2.1 AA compliance
|
| 498 |
|
| 499 |
**Features:**
|
| 500 |
+
|
| 501 |
- All interactive elements keyboard accessible
|
| 502 |
- ARIA labels on select interactive elements (e.g. chat input, message list, history actions, logs modal)
|
| 503 |
- Focus management in drawers and modals
|
| 504 |
- Screen reader announcements for loading states
|
| 505 |
|
| 506 |
**Implementation:**
|
| 507 |
+
|
| 508 |
```typescript
|
| 509 |
// Using Mantine's accessibility props
|
| 510 |
<Button aria-label="Search the web">
|
docs/wllama-model-evaluation.md
ADDED
|
@@ -0,0 +1,129 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Wllama Model Evaluation
|
| 2 |
+
|
| 3 |
+
How the in-browser models in `client/modules/wllama.ts` are measured, and how they scored. The catalog keeps a model unless it answers worse than LittleLamb 290M, the default for browsers without WebGPU; see [What the catalog keeps](#what-the-catalog-keeps).
|
| 4 |
+
|
| 5 |
+
## What it measures
|
| 6 |
+
|
| 7 |
+
MiniSearch's own task, not a general benchmark. `npm run bench:wllama` (see [development-commands.md](development-commands.md#wllama-benchmark)) loads each model in headless Chromium through the app's code: the catalog entry, `initializeWllama`, and the chat messages that `getFormattedSearchResults` and `getDefaultChatMessages` build from the default system prompt (`shared/defaultSystemPrompt.ts`), the search results, and the page excerpts. The model answers the way it would in the app, with the app's sampling, thinking budget and chat template settings.
|
| 8 |
+
|
| 9 |
+
The questions live in `scripts/wllama-bench/cases.json`. Each one carries its own search results, written for the test so the right answer is known, the facts a correct answer must contain, and the language it must be in. They cover what the system prompt asks of a model:
|
| 10 |
+
|
| 11 |
+
- **Answering in the question's language**, across eleven languages: English, Portuguese, Spanish, French, German, Chinese, Japanese, Korean, Arabic, Hindi and Russian. They are the languages most often listed as supported by the model families in the catalog and by multilingual benchmarks, with Russian added for its share of web text; see [Why these languages](#why-these-languages).
|
| 12 |
+
- **Answering in one language from results in another** (`pt-from-en`, `zh-from-en`).
|
| 13 |
+
- **Combining two results** (`de-synthesis`).
|
| 14 |
+
- **Picking the right result among similar ones** (`es-distractor`).
|
| 15 |
+
- **Saying so when the results do not answer** (`pt-not-covered`).
|
| 16 |
+
- **Citing a result with a link, and never inventing one.**
|
| 17 |
+
|
| 18 |
+
## How it scores
|
| 19 |
+
|
| 20 |
+
`scripts/wllama-bench/score.mjs` scores every answer out of 4 points:
|
| 21 |
+
|
| 22 |
+
| Points | For |
|
| 23 |
+
| ------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
|
| 24 |
+
| up to 2 | The required facts, split evenly. A wrong fact misleads more than a wrong language, so facts weigh double. |
|
| 25 |
+
| 1 | Answering in the question's language. Text copied from the results does not count, so a quoted snippet cannot pass for an answer written in that language. |
|
| 26 |
+
| 1 | Citing at least one of the results with a Markdown link. |
|
| 27 |
+
| โ1 each | A violation: a link to a URL that is not among the results, an echoed instruction from the prompt, a repeated sentence, a third script mixed in (Cyrillic in a Hindi answer), a number that appears in no result, a fact from the distractor result, or an answer cut off after 5 minutes. |
|
| 28 |
+
|
| 29 |
+
The score never goes below 0. English questions count once and the others twice, because MiniSearch takes queries in any language. A model's score is the sum over the questions, averaged over its runs.
|
| 30 |
+
|
| 31 |
+
### What the score cannot see
|
| 32 |
+
|
| 33 |
+
The checks are mechanical. Text that is garbled but in the right language and carrying the right facts passes them, and so does a correct answer that only restates the result it was given. Before the scorer was adopted, 176 answers were graded by hand as fluent and correct or not, and the scorer agreed with the hand grade on 149 of them (85%) with a pass mark of 2 points. Nearly all of the disagreements were about fluency. Treat a gap of a few points between two models as noise: the app samples, so a second run differs.
|
| 34 |
+
|
| 35 |
+
## Results
|
| 36 |
+
|
| 37 |
+
Measured on 2026-11-07 on an Apple M1 Max, in headless Chromium 151 on WebGPU, with wllama 3.8.1, the catalog's settings (a 1024-token thinking budget and `reasoning_effort: "low"`), and the app's sampling. Several models ran at once on the same GPU, so the run times are not comparable and are left out.
|
| 38 |
+
|
| 39 |
+
### The whole catalog, eleven core questions
|
| 40 |
+
|
| 41 |
+
Every model answered the first eleven questions (`en` to `ru`) once. LittleLamb and the models near its score ran three or four times, because the app samples and a single run can land 10 points either way: the spread between runs of one model averaged 5.5 points. Scores are out of 84.
|
| 42 |
+
|
| 43 |
+
| Rank | Model | Size | Runs | Score | en | pt-from-en | es | fr | de | zh | ja | ko | ar | hi | ru |
|
| 44 |
+
| ---: | ----------------------------------------------------------------------------- | ------: | ---: | ------: | --: | ---------: | --: | --: | --: | --: | --: | --: | --: | --: | --: |
|
| 45 |
+
| 1 | Gemma 3 4B | 2380 MB | 1 | 84.0/84 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 46 |
+
| 2 | Granite 3.3 2B | 1460 MB | 1 | 84.0/84 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 47 |
+
| 3 | Llama 3.2 3B | 1930 MB | 1 | 84.0/84 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 48 |
+
| 4 | Polaris 4B | 2380 MB | 1 | 82.0/84 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 49 |
+
| 5 | Spark X2.6 1.7B | 1769 MB | 1 | 82.0/84 | 4.0 | 3.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 50 |
+
| 6 | Gemma 2 2B | 1640 MB | 1 | 79.3/84 | 3.3 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 |
|
| 51 |
+
| 7 | Qwen 3.5 4B | 2436 MB | 1 | 77.0/84 | 3.0 | 3.0 | 3.0 | 4.0 | 3.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 52 |
+
| 8 | Falcon 3 3B | 1930 MB | 1 | 74.0/84 | 4.0 | 3.0 | 3.0 | 4.0 | 3.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 3.0 |
|
| 53 |
+
| 9 | Phi 3.5 Mini 3.8B | 2190 MB | 1 | 74.0/84 | 4.0 | 4.0 | 1.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 | 3.0 |
|
| 54 |
+
| 10 | LFM2.5 1.2B Instruct | 700 MB | 1 | 70.0/84 | 4.0 | 3.0 | 4.0 | 4.0 | 3.0 | 3.0 | 3.0 | 3.0 | 4.0 | 3.0 | 3.0 |
|
| 55 |
+
| 11 | Qwen 3.5 0.8B | 533 MB | 1 | 70.0/84 | 4.0 | 1.0 | 2.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 | 3.0 | 4.0 | 4.0 |
|
| 56 |
+
| 12 | EXAONE 3.5 2.4B | 1580 MB | 1 | 67.0/84 | 1.0 | 1.0 | 4.0 | 4.0 | 0.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 57 |
+
| 13 | Qwen 3 1.7B | 1140 MB | 1 | 66.0/84 | 4.0 | 3.0 | 4.0 | 3.0 | 3.0 | 3.0 | 3.0 | 3.0 | 3.0 | 3.0 | 3.0 |
|
| 58 |
+
| 14 | SmolLM 3 3B | 1820 MB | 1 | 65.0/84 | 3.0 | 4.0 | 3.0 | 4.0 | 3.0 | 3.0 | 3.0 | 3.0 | 2.0 | 3.0 | 3.0 |
|
| 59 |
+
| 15 | Qwen 3 0.6B | 406 MB | 1 | 63.0/84 | 3.0 | 2.0 | 4.0 | 3.0 | 3.0 | 3.0 | 4.0 | 3.0 | 2.0 | 3.0 | 3.0 |
|
| 60 |
+
| 16 | Granite 3.1 1B [400M] | 775 MB | 1 | 61.0/84 | 3.0 | 1.0 | 3.0 | 3.0 | 3.0 | 3.0 | 3.0 | 4.0 | 4.0 | 2.0 | 3.0 |
|
| 61 |
+
| 17 | AceInstruct 1.5B | 1070 MB | 1 | 60.3/84 | 2.3 | 1.0 | 3.0 | 3.0 | 3.0 | 3.0 | 4.0 | 4.0 | 3.0 | 3.0 | 2.0 |
|
| 62 |
+
| 18 | SmolLM 2 1.7B | 999 MB | 1 | 60.3/84 | 2.3 | 1.0 | 3.0 | 3.0 | 3.0 | 4.0 | 3.0 | 3.0 | 3.0 | 2.0 | 4.0 |
|
| 63 |
+
| 19 | Granite 3.1 3B [800M] | 1900 MB | 1 | 59.3/84 | 1.3 | 1.0 | 3.0 | 2.0 | 3.0 | 4.0 | 4.0 | 4.0 | 3.0 | 3.0 | 2.0 |
|
| 64 |
+
| 20 | Granite 4.2 3B | 2143 MB | 1 | 59.3/84 | 3.3 | 2.0 | 3.0 | 2.0 | 3.0 | 2.0 | 2.0 | 4.0 | 3.0 | 3.0 | 4.0 |
|
| 65 |
+
| 21 | Llama 3.2 1B | 776 MB | 3 | 59.2/84 | 2.6 | 3.3 | 3.0 | 3.3 | 3.0 | 2.3 | 2.0 | 2.0 | 3.0 | 3.7 | 2.7 |
|
| 66 |
+
| 22 | InternLM 2.5 1.8B | 1120 MB | 3 | 58.7/84 | 2.0 | 2.7 | 3.0 | 3.3 | 3.0 | 3.0 | 2.3 | 3.3 | 3.7 | 1.3 | 2.7 |
|
| 67 |
+
| 23 | Spark X2.6 4B | 3543 MB | 1 | 57.0/84 | 3.0 | 4.0 | 0.0 | 0.0 | 0.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 |
|
| 68 |
+
| 24 | ~~Granite 4.0 1B~~ (removed: scored on the CPU; on WebGPU it writes only `@`) | 981 MB | 1 | 56.0/84 | 4.0 | 3.0 | 3.0 | 4.0 | 2.0 | 3.0 | 1.0 | 2.0 | 4.0 | 0.0 | 4.0 |
|
| 69 |
+
| 25 | Megrez 3B | 1740 MB | 3 | 55.7/84 | 3.0 | 2.0 | 2.3 | 3.0 | 3.3 | 3.0 | 3.0 | 3.0 | 2.0 | 1.0 | 3.7 |
|
| 70 |
+
| 26 | Gemma 4 E2B | 2446 MB | 1 | 55.0/84 | 1.0 | 0.0 | 0.0 | 1.0 | 3.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 |
|
| 71 |
+
| 27 | Falcon 3 1B | 1020 MB | 3 | 54.3/84 | 2.3 | 1.7 | 3.7 | 2.7 | 3.0 | 1.3 | 2.0 | 3.0 | 3.3 | 3.0 | 2.3 |
|
| 72 |
+
| 28 | Phi 4 Mini Reasoning 3.8B | 2460 MB | 3 | 50.7/84 | 2.7 | 0.7 | 2.3 | 2.7 | 1.3 | 3.3 | 3.0 | 3.7 | 2.7 | 2.0 | 2.3 |
|
| 73 |
+
| 29 | Gemma 3 1B | 781 MB | 3 | 49.7/84 | 3.0 | 3.0 | 3.0 | 2.3 | 2.0 | 2.3 | 3.0 | 3.3 | 1.3 | 0.3 | 2.7 |
|
| 74 |
+
| 30 | LFM2.5 230M | 160 MB | 3 | 49.0/84 | 3.0 | 3.0 | 3.0 | 3.0 | 3.0 | 0.0 | 3.0 | 2.0 | 2.7 | 2.3 | 1.0 |
|
| 75 |
+
| 31 | DeepSeek R1 Distill Qwen 1.5B | 1070 MB | 3 | 45.4/84 | 2.1 | 2.0 | 3.3 | 2.3 | 2.0 | 3.7 | 2.7 | 3.0 | 1.0 | 1.0 | 0.7 |
|
| 76 |
+
| 32 | SmolLM 2 360M | 260 MB | 3 | 44.8/84 | 2.1 | 2.0 | 2.7 | 2.7 | 1.0 | 2.3 | 2.3 | 2.0 | 1.7 | 1.7 | 3.0 |
|
| 77 |
+
| 33 | MiniCPM4 0.5B | 267 MB | 3 | 41.6/84 | 2.2 | 0.3 | 2.7 | 1.7 | 0.7 | 2.0 | 3.3 | 2.3 | 1.7 | 2.0 | 3.0 |
|
| 78 |
+
| 34 | OLMo 2 1B | 967 MB | 3 | 41.6/84 | 0.9 | 0.0 | 2.0 | 2.7 | 1.0 | 2.3 | 1.7 | 2.3 | 3.0 | 3.3 | 2.0 |
|
| 79 |
+
| 35 | Gemma 3 270M | 250 MB | 3 | 40.1/84 | 2.1 | 2.0 | 2.0 | 2.0 | 0.0 | 3.0 | 0.0 | 3.0 | 3.0 | 1.0 | 3.0 |
|
| 80 |
+
| 36 | LittleLamb 290M | 231 MB | 4 | 38.0/84 | 2.5 | 1.5 | 2.5 | 2.3 | 3.0 | 2.5 | 2.0 | 1.3 | 1.0 | 0.3 | 1.5 |
|
| 81 |
+
| 37 | ~~Pints 1.5B~~ (removed) | 905 MB | 1 | 25.3/84 | 1.3 | 1.0 | 3.0 | 2.0 | 0.0 | 0.0 | 3.0 | 1.0 | 1.0 | 0.0 | 1.0 |
|
| 82 |
+
| 38 | ~~Granite 4.0 350M~~ (removed) | 229 MB | 1 | 25.0/84 | 3.0 | 2.0 | 2.0 | 2.0 | 0.0 | 0.0 | 0.0 | 2.0 | 1.0 | 1.0 | 1.0 |
|
| 83 |
+
| 39 | ~~SmolLM 2 135M~~ (removed) | 102 MB | 1 | 24.7/84 | 0.7 | 1.0 | 1.0 | 2.0 | 1.0 | 3.0 | 0.0 | 1.0 | 1.0 | 1.0 | 1.0 |
|
| 84 |
+
| 40 | ~~StableLM 2 Zephyr 1.6B~~ (removed) | 989 MB | 1 | 16.3/84 | 0.3 | 0.0 | 1.0 | 4.0 | 1.0 | 0.0 | 0.0 | 1.0 | 0.0 | 1.0 | 0.0 |
|
| 85 |
+
|
| 86 |
+
### What the catalog keeps
|
| 87 |
+
|
| 88 |
+
LittleLamb 290M is the catalog's floor: it is the default for browsers without WebGPU because it runs well on a CPU, and its answers are the least the catalog accepts. A model whose mean score falls below LittleLamb's (38.0, measured before LittleLamb got a thinking budget) leaves the catalog. A model that ties with it stays, because the score cannot see how a model writes, and some people prefer one model's style. On 2026-11-08 that removed StableLM 2 Zephyr 1.6B, SmolLM 2 135M, Granite 4.0 350M and Pints 1.5B, all far below the floor (16 to 25). Granite 4.0 1B scored above it on the CPU but was removed anyway, because on WebGPU it writes only `@` characters until it is stopped.
|
| 89 |
+
|
| 90 |
+
LittleLamb itself is the most erratic model measured (its four runs scored 49.0, 44.3, 33.7 and 25.0): it often answers in English, sometimes not at all, and its Arabic and Hindi answers are rarely usable. With the 1024-token thinking budget it scored 41.4 over three runs and left no answer empty.
|
| 91 |
+
|
| 92 |
+
### GPU default candidates, all fifteen questions
|
| 93 |
+
|
| 94 |
+
The eleven questions above are direct, and several models answered all of them, so they cannot rank the best ones. The twelve strongest candidates for the GPU default also answered the four harder questions (`de-synthesis`, `es-distractor`, `pt-not-covered`, `zh-from-en`), twice each. Scores are out of 116.
|
| 95 |
+
|
| 96 |
+
| Rank | Model | Size | Runs | Score | en | pt-from-en | es | fr | de | zh | ja | ko | ar | hi | ru | de-synthesis | es-distractor | pt-not-covered | zh-from-en |
|
| 97 |
+
| ---: | ----------------------- | ------: | ---: | --------: | --: | ---------: | --: | --: | --: | --: | --: | --: | --: | --: | --: | -----------: | ------------: | -------------: | ---------: |
|
| 98 |
+
| 1 | Gemma 3 4B | 2380 MB | 2 | 116.0/116 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 99 |
+
| 2 | Spark X2.6 1.7B | 1769 MB | 2 | 113.0/116 | 4.0 | 3.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.5 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
|
| 100 |
+
| 3 | Llama 3.2 3B | 1930 MB | 2 | 110.5/116 | 3.5 | 3.5 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 3.0 | 4.0 | 4.0 | 4.0 |
|
| 101 |
+
| 4 | Polaris 4B | 2380 MB | 2 | 106.0/116 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.5 | 4.0 | 1.0 | 3.5 |
|
| 102 |
+
| 5 | Granite 3.3 2B | 1460 MB | 2 | 105.0/116 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 2.0 | 4.0 | 0.5 | 4.0 |
|
| 103 |
+
| 6 | Gemma 2 2B | 1640 MB | 2 | 102.7/116 | 3.7 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 4.0 | 4.0 | 3.5 | 4.0 | 4.0 | 3.0 | 4.0 | 0.5 | 3.5 |
|
| 104 |
+
| 7 | Granite 4.2 3B | 2143 MB | 2 | 95.0/116 | 4.0 | 1.5 | 3.0 | 3.5 | 3.0 | 3.0 | 4.0 | 4.0 | 4.0 | 4.0 | 3.5 | 4.0 | 3.0 | 2.0 | 3.0 |
|
| 105 |
+
| 8 | Falcon 3 3B | 1930 MB | 2 | 93.5/116 | 3.5 | 2.5 | 3.5 | 4.0 | 3.0 | 4.0 | 4.0 | 4.0 | 2.5 | 2.5 | 4.0 | 2.5 | 4.0 | 0.5 | 4.0 |
|
| 106 |
+
| 9 | Phi 3.5 Mini 3.8B | 2190 MB | 2 | 93.5/116 | 2.5 | 3.5 | 1.5 | 3.0 | 2.5 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 3.5 | 3.0 | 3.5 | 2.5 | 3.5 |
|
| 107 |
+
| 10 | Qwen 3.5 4B (see below) | 2436 MB | 2 | 90.0/116 | 2.0 | 2.0 | 1.5 | 2.0 | 1.5 | 4.0 | 4.0 | 4.0 | 4.0 | 3.0 | 3.0 | 3.5 | 4.0 | 3.5 | 4.0 |
|
| 108 |
+
| 11 | Qwen 3.5 0.8B | 533 MB | 2 | 76.7/116 | 2.7 | 2.0 | 3.5 | 3.0 | 1.0 | 4.0 | 4.0 | 0.5 | 3.5 | 2.0 | 2.5 | 3.5 | 3.5 | 2.0 | 2.0 |
|
| 109 |
+
| 12 | LFM2.5 1.2B Instruct | 700 MB | 2 | 74.0/116 | 4.0 | 3.0 | 3.5 | 2.5 | 2.5 | 2.5 | 3.0 | 2.5 | 3.0 | 2.0 | 2.0 | 1.0 | 2.5 | 2.0 | 3.0 |
|
| 110 |
+
|
| 111 |
+
Gemma 3 4B, the GPU default, scored every point in both runs. The harder questions separate the rest: `pt-not-covered` is the one most models fail, inventing Curitiba's population instead of saying the results do not give it, and `de-synthesis` catches models that compute the difference between the two mountains wrong. Qwen 3.5 4B's first run shared the GPU with other runs and hit the 5-minute limit on six questions; its second run, alone, scored 108.
|
| 112 |
+
|
| 113 |
+
## Why these languages
|
| 114 |
+
|
| 115 |
+
The catalog's model families and the common multilingual benchmarks agree on a core:
|
| 116 |
+
|
| 117 |
+
- Llama 3.2 officially supports English, German, French, Italian, Portuguese, Hindi, Spanish and Thai ([model card](https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct)).
|
| 118 |
+
- OpenAI's MMMLU translates MMLU into Arabic, Bengali, German, Spanish, French, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, Swahili, Yoruba and Chinese ([dataset card](https://huggingface.co/datasets/openai/MMMLU/blob/main/README.md)).
|
| 119 |
+
- LFM2 lists English, Arabic, Chinese, French, German, Japanese, Korean and Spanish ([model card](https://huggingface.co/LiquidAI/LFM2-2.6B/blob/refs%2Fpr%2F7/README.md)).
|
| 120 |
+
- Russian is among the largest languages in Common Crawl after English, in the group that each makes up a few percent of its pages ([Common Crawl](https://en.wikipedia.org/wiki/Common_Crawl)).
|
| 121 |
+
|
| 122 |
+
## Running it
|
| 123 |
+
|
| 124 |
+
```bash
|
| 125 |
+
npm run bench:wllama -- --model littlelamb-290m --model qwen-3-0.6b --device webgpu --runs 2 --json results.json
|
| 126 |
+
node scripts/wllama-bench/score.mjs scripts/wllama-bench/cases.json results.json
|
| 127 |
+
```
|
| 128 |
+
|
| 129 |
+
The first command prints the score at the end; the second re-scores saved results, for example after a change to the scorer. To vet a model that is not in the catalog yet, pass `--model owner/repo/file-00001-of-0000N.gguf`.
|
e2e/main-thread-budget.spec.ts
CHANGED
|
@@ -231,6 +231,7 @@ function stubEndpoints(page: Page) {
|
|
| 231 |
accessKeysEnabled: false,
|
| 232 |
accessKeyTimeoutHours: 24,
|
| 233 |
wllamaDefaultModelId: "littlelamb-290m",
|
|
|
|
| 234 |
internalApiEnabled: false,
|
| 235 |
internalApiName: "Internal API",
|
| 236 |
defaultInferenceType: "openai",
|
|
|
|
| 231 |
accessKeysEnabled: false,
|
| 232 |
accessKeyTimeoutHours: 24,
|
| 233 |
wllamaDefaultModelId: "littlelamb-290m",
|
| 234 |
+
wllamaDefaultGpuModelId: "gemma-3-4b",
|
| 235 |
internalApiEnabled: false,
|
| 236 |
internalApiName: "Internal API",
|
| 237 |
defaultInferenceType: "openai",
|
eval/README.md
CHANGED
|
@@ -6,6 +6,7 @@ has two parts, both driven by a single fixed golden set (`goldenSet.ts`).
|
|
| 6 |
|
| 7 |
Some of it runs in the default `npm test` suite (and so in CI); the rest is a
|
| 8 |
local signal:
|
|
|
|
| 9 |
- `metrics.test.ts`, `goldenSet.test.ts`, `promptConstruction.test.ts` run in
|
| 10 |
the default suite: pure math, golden-set structure, and prompt construction.
|
| 11 |
No model, no API key.
|
|
@@ -19,6 +20,7 @@ change, not an absolute quality score.
|
|
| 19 |
## The golden set
|
| 20 |
|
| 21 |
`goldenSet.ts` is a fixed, hand-curated list of queries. Each entry has:
|
|
|
|
| 22 |
- `query`: the search query.
|
| 23 |
- `results`: realistic candidate results (title, snippet, url), mixed
|
| 24 |
relevant and irrelevant, the way a real search engine returns them.
|
|
|
|
| 6 |
|
| 7 |
Some of it runs in the default `npm test` suite (and so in CI); the rest is a
|
| 8 |
local signal:
|
| 9 |
+
|
| 10 |
- `metrics.test.ts`, `goldenSet.test.ts`, `promptConstruction.test.ts` run in
|
| 11 |
the default suite: pure math, golden-set structure, and prompt construction.
|
| 12 |
No model, no API key.
|
|
|
|
| 20 |
## The golden set
|
| 21 |
|
| 22 |
`goldenSet.ts` is a fixed, hand-curated list of queries. Each entry has:
|
| 23 |
+
|
| 24 |
- `query`: the search query.
|
| 25 |
- `results`: realistic candidate results (title, snippet, url), mixed
|
| 26 |
relevant and irrelevant, the way a real search engine returns them.
|
knip.json
CHANGED
|
@@ -8,6 +8,7 @@
|
|
| 8 |
"ignoreFiles": [
|
| 9 |
"client/postcss.config.cjs",
|
| 10 |
"vitest.integration.config.ts",
|
| 11 |
-
"scripts/npm-bundle-overlay.cjs"
|
|
|
|
| 12 |
]
|
| 13 |
}
|
|
|
|
| 8 |
"ignoreFiles": [
|
| 9 |
"client/postcss.config.cjs",
|
| 10 |
"vitest.integration.config.ts",
|
| 11 |
+
"scripts/npm-bundle-overlay.cjs",
|
| 12 |
+
"scripts/wllama-bench/bench.ts"
|
| 13 |
]
|
| 14 |
}
|
package-lock.json
CHANGED
|
@@ -96,13 +96,13 @@
|
|
| 96 |
"license": "MIT"
|
| 97 |
},
|
| 98 |
"node_modules/@ai-sdk/gateway": {
|
| 99 |
-
"version": "4.0.
|
| 100 |
-
"resolved": "https://registry.npmjs.org/@ai-sdk/gateway/-/gateway-4.0.
|
| 101 |
-
"integrity": "sha512-
|
| 102 |
"license": "Apache-2.0",
|
| 103 |
"dependencies": {
|
| 104 |
-
"@ai-sdk/provider": "4.0.
|
| 105 |
-
"@ai-sdk/provider-utils": "5.0.
|
| 106 |
"@vercel/oidc": "3.2.0"
|
| 107 |
},
|
| 108 |
"engines": {
|
|
@@ -112,14 +112,26 @@
|
|
| 112 |
"zod": "^3.25.76 || ^4.1.8"
|
| 113 |
}
|
| 114 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 115 |
"node_modules/@ai-sdk/openai-compatible": {
|
| 116 |
-
"version": "3.0.
|
| 117 |
-
"resolved": "https://registry.npmjs.org/@ai-sdk/openai-compatible/-/openai-compatible-3.0.
|
| 118 |
-
"integrity": "sha512-
|
| 119 |
"license": "Apache-2.0",
|
| 120 |
"dependencies": {
|
| 121 |
-
"@ai-sdk/provider": "4.0.
|
| 122 |
-
"@ai-sdk/provider-utils": "5.0.
|
| 123 |
},
|
| 124 |
"engines": {
|
| 125 |
"node": ">=22"
|
|
@@ -128,6 +140,18 @@
|
|
| 128 |
"zod": "^3.25.76 || ^4.1.8"
|
| 129 |
}
|
| 130 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 131 |
"node_modules/@ai-sdk/provider": {
|
| 132 |
"version": "4.0.35",
|
| 133 |
"resolved": "https://registry.npmjs.org/@ai-sdk/provider/-/provider-4.0.35.tgz",
|
|
@@ -141,16 +165,16 @@
|
|
| 141 |
}
|
| 142 |
},
|
| 143 |
"node_modules/@ai-sdk/provider-utils": {
|
| 144 |
-
"version": "5.0.
|
| 145 |
-
"resolved": "https://registry.npmjs.org/@ai-sdk/provider-utils/-/provider-utils-5.0.
|
| 146 |
-
"integrity": "sha512-
|
| 147 |
"license": "Apache-2.0",
|
| 148 |
"dependencies": {
|
| 149 |
-
"@ai-sdk/provider": "4.0.
|
| 150 |
"@standard-schema/spec": "^1.1.0",
|
| 151 |
"@workflow/serde": "4.1.0",
|
| 152 |
"eventsource-parser": "^3.0.8",
|
| 153 |
-
"undici": "^7.29.
|
| 154 |
},
|
| 155 |
"engines": {
|
| 156 |
"node": ">=22"
|
|
@@ -159,6 +183,18 @@
|
|
| 159 |
"zod": "^3.25.76 || ^4.1.8"
|
| 160 |
}
|
| 161 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 162 |
"node_modules/@ai-sdk/provider-utils/node_modules/undici": {
|
| 163 |
"version": "7.30.1",
|
| 164 |
"resolved": "https://registry.npmjs.org/undici/-/undici-7.30.1.tgz",
|
|
@@ -286,9 +322,9 @@
|
|
| 286 |
}
|
| 287 |
},
|
| 288 |
"node_modules/@biomejs/biome": {
|
| 289 |
-
"version": "2.6.
|
| 290 |
-
"resolved": "https://registry.npmjs.org/@biomejs/biome/-/biome-2.6.
|
| 291 |
-
"integrity": "sha512-
|
| 292 |
"dev": true,
|
| 293 |
"license": "MIT OR Apache-2.0",
|
| 294 |
"bin": {
|
|
@@ -302,22 +338,22 @@
|
|
| 302 |
"url": "https://opencollective.com/biome"
|
| 303 |
},
|
| 304 |
"optionalDependencies": {
|
| 305 |
-
"@biomejs/cli-android-arm64": "2.6.
|
| 306 |
-
"@biomejs/cli-android-x64": "2.6.
|
| 307 |
-
"@biomejs/cli-darwin-arm64": "2.6.
|
| 308 |
-
"@biomejs/cli-darwin-x64": "2.6.
|
| 309 |
-
"@biomejs/cli-linux-arm64": "2.6.
|
| 310 |
-
"@biomejs/cli-linux-arm64-musl": "2.6.
|
| 311 |
-
"@biomejs/cli-linux-x64": "2.6.
|
| 312 |
-
"@biomejs/cli-linux-x64-musl": "2.6.
|
| 313 |
-
"@biomejs/cli-win32-arm64": "2.6.
|
| 314 |
-
"@biomejs/cli-win32-x64": "2.6.
|
| 315 |
}
|
| 316 |
},
|
| 317 |
"node_modules/@biomejs/cli-android-arm64": {
|
| 318 |
-
"version": "2.6.
|
| 319 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-android-arm64/-/cli-android-arm64-2.6.
|
| 320 |
-
"integrity": "sha512-
|
| 321 |
"cpu": [
|
| 322 |
"arm64"
|
| 323 |
],
|
|
@@ -332,9 +368,9 @@
|
|
| 332 |
}
|
| 333 |
},
|
| 334 |
"node_modules/@biomejs/cli-android-x64": {
|
| 335 |
-
"version": "2.6.
|
| 336 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-android-x64/-/cli-android-x64-2.6.
|
| 337 |
-
"integrity": "sha512-
|
| 338 |
"cpu": [
|
| 339 |
"x64"
|
| 340 |
],
|
|
@@ -349,9 +385,9 @@
|
|
| 349 |
}
|
| 350 |
},
|
| 351 |
"node_modules/@biomejs/cli-darwin-arm64": {
|
| 352 |
-
"version": "2.6.
|
| 353 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-darwin-arm64/-/cli-darwin-arm64-2.6.
|
| 354 |
-
"integrity": "sha512-
|
| 355 |
"cpu": [
|
| 356 |
"arm64"
|
| 357 |
],
|
|
@@ -366,9 +402,9 @@
|
|
| 366 |
}
|
| 367 |
},
|
| 368 |
"node_modules/@biomejs/cli-darwin-x64": {
|
| 369 |
-
"version": "2.6.
|
| 370 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-darwin-x64/-/cli-darwin-x64-2.6.
|
| 371 |
-
"integrity": "sha512-
|
| 372 |
"cpu": [
|
| 373 |
"x64"
|
| 374 |
],
|
|
@@ -383,9 +419,9 @@
|
|
| 383 |
}
|
| 384 |
},
|
| 385 |
"node_modules/@biomejs/cli-linux-arm64": {
|
| 386 |
-
"version": "2.6.
|
| 387 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-linux-arm64/-/cli-linux-arm64-2.6.
|
| 388 |
-
"integrity": "sha512-
|
| 389 |
"cpu": [
|
| 390 |
"arm64"
|
| 391 |
],
|
|
@@ -403,9 +439,9 @@
|
|
| 403 |
}
|
| 404 |
},
|
| 405 |
"node_modules/@biomejs/cli-linux-arm64-musl": {
|
| 406 |
-
"version": "2.6.
|
| 407 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-linux-arm64-musl/-/cli-linux-arm64-musl-2.6.
|
| 408 |
-
"integrity": "sha512-
|
| 409 |
"cpu": [
|
| 410 |
"arm64"
|
| 411 |
],
|
|
@@ -423,9 +459,9 @@
|
|
| 423 |
}
|
| 424 |
},
|
| 425 |
"node_modules/@biomejs/cli-linux-x64": {
|
| 426 |
-
"version": "2.6.
|
| 427 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-linux-x64/-/cli-linux-x64-2.6.
|
| 428 |
-
"integrity": "sha512-
|
| 429 |
"cpu": [
|
| 430 |
"x64"
|
| 431 |
],
|
|
@@ -443,9 +479,9 @@
|
|
| 443 |
}
|
| 444 |
},
|
| 445 |
"node_modules/@biomejs/cli-linux-x64-musl": {
|
| 446 |
-
"version": "2.6.
|
| 447 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-linux-x64-musl/-/cli-linux-x64-musl-2.6.
|
| 448 |
-
"integrity": "sha512-
|
| 449 |
"cpu": [
|
| 450 |
"x64"
|
| 451 |
],
|
|
@@ -463,9 +499,9 @@
|
|
| 463 |
}
|
| 464 |
},
|
| 465 |
"node_modules/@biomejs/cli-win32-arm64": {
|
| 466 |
-
"version": "2.6.
|
| 467 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-win32-arm64/-/cli-win32-arm64-2.6.
|
| 468 |
-
"integrity": "sha512-
|
| 469 |
"cpu": [
|
| 470 |
"arm64"
|
| 471 |
],
|
|
@@ -480,9 +516,9 @@
|
|
| 480 |
}
|
| 481 |
},
|
| 482 |
"node_modules/@biomejs/cli-win32-x64": {
|
| 483 |
-
"version": "2.6.
|
| 484 |
-
"resolved": "https://registry.npmjs.org/@biomejs/cli-win32-x64/-/cli-win32-x64-2.6.
|
| 485 |
-
"integrity": "sha512-
|
| 486 |
"cpu": [
|
| 487 |
"x64"
|
| 488 |
],
|
|
@@ -822,13 +858,13 @@
|
|
| 822 |
"license": "MIT"
|
| 823 |
},
|
| 824 |
"node_modules/@mantine/carousel": {
|
| 825 |
-
"version": "9.7.
|
| 826 |
-
"resolved": "https://registry.npmjs.org/@mantine/carousel/-/carousel-9.7.
|
| 827 |
-
"integrity": "sha512-
|
| 828 |
"license": "MIT",
|
| 829 |
"peerDependencies": {
|
| 830 |
-
"@mantine/core": "9.7.
|
| 831 |
-
"@mantine/hooks": "9.7.
|
| 832 |
"embla-carousel": ">=8.0.0",
|
| 833 |
"embla-carousel-react": ">=8.0.0",
|
| 834 |
"react": "^19.2.0",
|
|
@@ -836,24 +872,24 @@
|
|
| 836 |
}
|
| 837 |
},
|
| 838 |
"node_modules/@mantine/code-highlight": {
|
| 839 |
-
"version": "9.7.
|
| 840 |
-
"resolved": "https://registry.npmjs.org/@mantine/code-highlight/-/code-highlight-9.7.
|
| 841 |
-
"integrity": "sha512-
|
| 842 |
"license": "MIT",
|
| 843 |
"dependencies": {
|
| 844 |
"clsx": "^2.1.1"
|
| 845 |
},
|
| 846 |
"peerDependencies": {
|
| 847 |
-
"@mantine/core": "9.7.
|
| 848 |
-
"@mantine/hooks": "9.7.
|
| 849 |
"react": "^19.2.0",
|
| 850 |
"react-dom": "^19.2.0"
|
| 851 |
}
|
| 852 |
},
|
| 853 |
"node_modules/@mantine/core": {
|
| 854 |
-
"version": "9.7.
|
| 855 |
-
"resolved": "https://registry.npmjs.org/@mantine/core/-/core-9.7.
|
| 856 |
-
"integrity": "sha512-
|
| 857 |
"license": "MIT",
|
| 858 |
"dependencies": {
|
| 859 |
"@floating-ui/react": "^0.27.20",
|
|
@@ -863,15 +899,15 @@
|
|
| 863 |
"type-fest": "^5.10.0"
|
| 864 |
},
|
| 865 |
"peerDependencies": {
|
| 866 |
-
"@mantine/hooks": "9.7.
|
| 867 |
"react": "^19.2.0",
|
| 868 |
"react-dom": "^19.2.0"
|
| 869 |
}
|
| 870 |
},
|
| 871 |
"node_modules/@mantine/form": {
|
| 872 |
-
"version": "9.7.
|
| 873 |
-
"resolved": "https://registry.npmjs.org/@mantine/form/-/form-9.7.
|
| 874 |
-
"integrity": "sha512-
|
| 875 |
"license": "MIT",
|
| 876 |
"dependencies": {
|
| 877 |
"@standard-schema/spec": "^1.1.0",
|
|
@@ -883,34 +919,34 @@
|
|
| 883 |
}
|
| 884 |
},
|
| 885 |
"node_modules/@mantine/hooks": {
|
| 886 |
-
"version": "9.7.
|
| 887 |
-
"resolved": "https://registry.npmjs.org/@mantine/hooks/-/hooks-9.7.
|
| 888 |
-
"integrity": "sha512-
|
| 889 |
"license": "MIT",
|
| 890 |
"peerDependencies": {
|
| 891 |
"react": "^19.2.0"
|
| 892 |
}
|
| 893 |
},
|
| 894 |
"node_modules/@mantine/notifications": {
|
| 895 |
-
"version": "9.7.
|
| 896 |
-
"resolved": "https://registry.npmjs.org/@mantine/notifications/-/notifications-9.7.
|
| 897 |
-
"integrity": "sha512-
|
| 898 |
"license": "MIT",
|
| 899 |
"dependencies": {
|
| 900 |
-
"@mantine/store": "9.7.
|
| 901 |
"react-transition-group": "4.4.5"
|
| 902 |
},
|
| 903 |
"peerDependencies": {
|
| 904 |
-
"@mantine/core": "9.7.
|
| 905 |
-
"@mantine/hooks": "9.7.
|
| 906 |
"react": "^19.2.0",
|
| 907 |
"react-dom": "^19.2.0"
|
| 908 |
}
|
| 909 |
},
|
| 910 |
"node_modules/@mantine/store": {
|
| 911 |
-
"version": "9.7.
|
| 912 |
-
"resolved": "https://registry.npmjs.org/@mantine/store/-/store-9.7.
|
| 913 |
-
"integrity": "sha512-
|
| 914 |
"license": "MIT",
|
| 915 |
"peerDependencies": {
|
| 916 |
"react": "^19.2.0"
|
|
@@ -2914,15 +2950,15 @@
|
|
| 2914 |
}
|
| 2915 |
},
|
| 2916 |
"node_modules/@vitest/coverage-v8": {
|
| 2917 |
-
"version": "5.
|
| 2918 |
-
"resolved": "https://registry.npmjs.org/@vitest/coverage-v8/-/coverage-v8-5.0.
|
| 2919 |
-
"integrity": "sha512-
|
| 2920 |
"dev": true,
|
| 2921 |
"license": "MIT",
|
| 2922 |
"dependencies": {
|
| 2923 |
"@bcoe/v8-coverage": "^1.0.2",
|
| 2924 |
-
"@vitest/istanbul-lib-coverage": "^1.0
|
| 2925 |
-
"@vitest/istanbul-lib-report": "^1.0
|
| 2926 |
"ast-v8-to-istanbul": "^1.0.7",
|
| 2927 |
"magicast": "^0.5.5",
|
| 2928 |
"obug": "^3.0.0",
|
|
@@ -2933,8 +2969,8 @@
|
|
| 2933 |
"url": "https://opencollective.com/vitest"
|
| 2934 |
},
|
| 2935 |
"peerDependencies": {
|
| 2936 |
-
"@vitest/browser": "5.
|
| 2937 |
-
"vitest": "5.
|
| 2938 |
},
|
| 2939 |
"peerDependenciesMeta": {
|
| 2940 |
"@vitest/browser": {
|
|
@@ -2943,9 +2979,9 @@
|
|
| 2943 |
}
|
| 2944 |
},
|
| 2945 |
"node_modules/@vitest/istanbul-lib-coverage": {
|
| 2946 |
-
"version": "1.0
|
| 2947 |
-
"resolved": "https://registry.npmjs.org/@vitest/istanbul-lib-coverage/-/istanbul-lib-coverage-1.0.
|
| 2948 |
-
"integrity": "sha512-
|
| 2949 |
"dev": true,
|
| 2950 |
"license": "MIT",
|
| 2951 |
"engines": {
|
|
@@ -2953,27 +2989,27 @@
|
|
| 2953 |
}
|
| 2954 |
},
|
| 2955 |
"node_modules/@vitest/istanbul-lib-report": {
|
| 2956 |
-
"version": "1.0
|
| 2957 |
-
"resolved": "https://registry.npmjs.org/@vitest/istanbul-lib-report/-/istanbul-lib-report-1.0.
|
| 2958 |
-
"integrity": "sha512-
|
| 2959 |
"dev": true,
|
| 2960 |
"license": "MIT",
|
| 2961 |
"dependencies": {
|
| 2962 |
-
"@vitest/istanbul-lib-coverage": "1.0
|
| 2963 |
},
|
| 2964 |
"engines": {
|
| 2965 |
"node": ">=22"
|
| 2966 |
}
|
| 2967 |
},
|
| 2968 |
"node_modules/@vitest/mocker": {
|
| 2969 |
-
"version": "5.
|
| 2970 |
-
"resolved": "https://registry.npmjs.org/@vitest/mocker/-/mocker-5.0.
|
| 2971 |
-
"integrity": "sha512-
|
| 2972 |
"dev": true,
|
| 2973 |
"license": "MIT",
|
| 2974 |
"dependencies": {
|
| 2975 |
"@jridgewell/trace-mapping": "0.3.31",
|
| 2976 |
-
"@vitest/spy": "5.
|
| 2977 |
"estree-walker": "^3.0.3",
|
| 2978 |
"magic-string": "^1.4.3"
|
| 2979 |
},
|
|
@@ -2994,9 +3030,9 @@
|
|
| 2994 |
}
|
| 2995 |
},
|
| 2996 |
"node_modules/@vitest/spy": {
|
| 2997 |
-
"version": "5.
|
| 2998 |
-
"resolved": "https://registry.npmjs.org/@vitest/spy/-/spy-5.0.
|
| 2999 |
-
"integrity": "sha512-
|
| 3000 |
"dev": true,
|
| 3001 |
"license": "MIT",
|
| 3002 |
"funding": {
|
|
@@ -3025,14 +3061,14 @@
|
|
| 3025 |
}
|
| 3026 |
},
|
| 3027 |
"node_modules/ai": {
|
| 3028 |
-
"version": "7.0.
|
| 3029 |
-
"resolved": "https://registry.npmjs.org/ai/-/ai-7.0.
|
| 3030 |
-
"integrity": "sha512-
|
| 3031 |
"license": "Apache-2.0",
|
| 3032 |
"dependencies": {
|
| 3033 |
-
"@ai-sdk/gateway": "4.0.
|
| 3034 |
-
"@ai-sdk/provider": "4.0.
|
| 3035 |
-
"@ai-sdk/provider-utils": "5.0.
|
| 3036 |
},
|
| 3037 |
"engines": {
|
| 3038 |
"node": ">=22"
|
|
@@ -3041,6 +3077,18 @@
|
|
| 3041 |
"zod": "^3.25.76 || ^4.1.8"
|
| 3042 |
}
|
| 3043 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3044 |
"node_modules/ansi-regex": {
|
| 3045 |
"version": "5.0.1",
|
| 3046 |
"resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-5.0.1.tgz",
|
|
@@ -6719,9 +6767,9 @@
|
|
| 6719 |
}
|
| 6720 |
},
|
| 6721 |
"node_modules/rate-limiter-flexible": {
|
| 6722 |
-
"version": "11.2.
|
| 6723 |
-
"resolved": "https://registry.npmjs.org/rate-limiter-flexible/-/rate-limiter-flexible-11.2.
|
| 6724 |
-
"integrity": "sha512-
|
| 6725 |
"license": "ISC"
|
| 6726 |
},
|
| 6727 |
"node_modules/react": {
|
|
@@ -7074,9 +7122,9 @@
|
|
| 7074 |
}
|
| 7075 |
},
|
| 7076 |
"node_modules/rollup-plugin-visualizer": {
|
| 7077 |
-
"version": "7.1.
|
| 7078 |
-
"resolved": "https://registry.npmjs.org/rollup-plugin-visualizer/-/rollup-plugin-visualizer-7.1.
|
| 7079 |
-
"integrity": "sha512-
|
| 7080 |
"license": "MIT",
|
| 7081 |
"dependencies": {
|
| 7082 |
"open": "^11.0.0",
|
|
@@ -7518,21 +7566,21 @@
|
|
| 7518 |
}
|
| 7519 |
},
|
| 7520 |
"node_modules/tldts": {
|
| 7521 |
-
"version": "7.
|
| 7522 |
-
"resolved": "https://registry.npmjs.org/tldts/-/tldts-7.
|
| 7523 |
-
"integrity": "sha512-
|
| 7524 |
"license": "MIT",
|
| 7525 |
"dependencies": {
|
| 7526 |
-
"tldts-core": "^7.
|
| 7527 |
},
|
| 7528 |
"bin": {
|
| 7529 |
"tldts": "bin/cli.js"
|
| 7530 |
}
|
| 7531 |
},
|
| 7532 |
"node_modules/tldts-core": {
|
| 7533 |
-
"version": "7.
|
| 7534 |
-
"resolved": "https://registry.npmjs.org/tldts-core/-/tldts-core-7.
|
| 7535 |
-
"integrity": "sha512-
|
| 7536 |
"license": "MIT"
|
| 7537 |
},
|
| 7538 |
"node_modules/tough-cookie": {
|
|
@@ -7923,14 +7971,14 @@
|
|
| 7923 |
}
|
| 7924 |
},
|
| 7925 |
"node_modules/vitest": {
|
| 7926 |
-
"version": "5.
|
| 7927 |
-
"resolved": "https://registry.npmjs.org/vitest/-/vitest-5.0.
|
| 7928 |
-
"integrity": "sha512-
|
| 7929 |
"dev": true,
|
| 7930 |
"license": "MIT",
|
| 7931 |
"dependencies": {
|
| 7932 |
"@types/chai": "^5.2.2",
|
| 7933 |
-
"@vitest/mocker": "5.
|
| 7934 |
"chai": "^6.3.0",
|
| 7935 |
"es-module-lexer": "^2.3.2",
|
| 7936 |
"expect-type": "^1.4.0",
|
|
@@ -7956,12 +8004,12 @@
|
|
| 7956 |
"@edge-runtime/vm": "*",
|
| 7957 |
"@opentelemetry/api": "^1.9.0",
|
| 7958 |
"@types/node": "^22.0.0 || >=24.0.0",
|
| 7959 |
-
"@vitest/browser-playwright": "5.
|
| 7960 |
-
"@vitest/browser-preview": "5.
|
| 7961 |
"@vitest/browser-webdriverio": "^5.0.0-beta.5 || >=5.0.0",
|
| 7962 |
-
"@vitest/coverage-istanbul": "5.
|
| 7963 |
-
"@vitest/coverage-v8": "5.
|
| 7964 |
-
"@vitest/ui": "5.
|
| 7965 |
"happy-dom": "*",
|
| 7966 |
"jsdom": "*",
|
| 7967 |
"vite": "^6.4.0 || ^7.0.0 || ^8.0.0"
|
|
|
|
| 96 |
"license": "MIT"
|
| 97 |
},
|
| 98 |
"node_modules/@ai-sdk/gateway": {
|
| 99 |
+
"version": "4.0.136",
|
| 100 |
+
"resolved": "https://registry.npmjs.org/@ai-sdk/gateway/-/gateway-4.0.136.tgz",
|
| 101 |
+
"integrity": "sha512-P851NL80P85Ee6MhblRuuCzNlQdOYU5rKmdt/Wuo5dMtkzvIHhyVZY9wtfKG3wBJUUoxHxWKsI2aR6O75o8diA==",
|
| 102 |
"license": "Apache-2.0",
|
| 103 |
"dependencies": {
|
| 104 |
+
"@ai-sdk/provider": "4.0.36",
|
| 105 |
+
"@ai-sdk/provider-utils": "5.0.75",
|
| 106 |
"@vercel/oidc": "3.2.0"
|
| 107 |
},
|
| 108 |
"engines": {
|
|
|
|
| 112 |
"zod": "^3.25.76 || ^4.1.8"
|
| 113 |
}
|
| 114 |
},
|
| 115 |
+
"node_modules/@ai-sdk/gateway/node_modules/@ai-sdk/provider": {
|
| 116 |
+
"version": "4.0.36",
|
| 117 |
+
"resolved": "https://registry.npmjs.org/@ai-sdk/provider/-/provider-4.0.36.tgz",
|
| 118 |
+
"integrity": "sha512-GqvWZSZdBelCGughlBdOoupT258rgM3YMbaLAS9WSZJ0vlt8c9vm3My+ZT31mYhiReMCMCF2qF2axnGbILpk1A==",
|
| 119 |
+
"license": "Apache-2.0",
|
| 120 |
+
"dependencies": {
|
| 121 |
+
"json-schema": "^0.4.0"
|
| 122 |
+
},
|
| 123 |
+
"engines": {
|
| 124 |
+
"node": ">=22"
|
| 125 |
+
}
|
| 126 |
+
},
|
| 127 |
"node_modules/@ai-sdk/openai-compatible": {
|
| 128 |
+
"version": "3.0.85",
|
| 129 |
+
"resolved": "https://registry.npmjs.org/@ai-sdk/openai-compatible/-/openai-compatible-3.0.85.tgz",
|
| 130 |
+
"integrity": "sha512-vZx3MvZjc7h7Ud8GUEvS9Qngtoec+Zogv5Zxw8kVjXrZhp/OOHfEosXoqjScnokoCK/d5cQfurfNNUmTrbZOQQ==",
|
| 131 |
"license": "Apache-2.0",
|
| 132 |
"dependencies": {
|
| 133 |
+
"@ai-sdk/provider": "4.0.36",
|
| 134 |
+
"@ai-sdk/provider-utils": "5.0.75"
|
| 135 |
},
|
| 136 |
"engines": {
|
| 137 |
"node": ">=22"
|
|
|
|
| 140 |
"zod": "^3.25.76 || ^4.1.8"
|
| 141 |
}
|
| 142 |
},
|
| 143 |
+
"node_modules/@ai-sdk/openai-compatible/node_modules/@ai-sdk/provider": {
|
| 144 |
+
"version": "4.0.36",
|
| 145 |
+
"resolved": "https://registry.npmjs.org/@ai-sdk/provider/-/provider-4.0.36.tgz",
|
| 146 |
+
"integrity": "sha512-GqvWZSZdBelCGughlBdOoupT258rgM3YMbaLAS9WSZJ0vlt8c9vm3My+ZT31mYhiReMCMCF2qF2axnGbILpk1A==",
|
| 147 |
+
"license": "Apache-2.0",
|
| 148 |
+
"dependencies": {
|
| 149 |
+
"json-schema": "^0.4.0"
|
| 150 |
+
},
|
| 151 |
+
"engines": {
|
| 152 |
+
"node": ">=22"
|
| 153 |
+
}
|
| 154 |
+
},
|
| 155 |
"node_modules/@ai-sdk/provider": {
|
| 156 |
"version": "4.0.35",
|
| 157 |
"resolved": "https://registry.npmjs.org/@ai-sdk/provider/-/provider-4.0.35.tgz",
|
|
|
|
| 165 |
}
|
| 166 |
},
|
| 167 |
"node_modules/@ai-sdk/provider-utils": {
|
| 168 |
+
"version": "5.0.75",
|
| 169 |
+
"resolved": "https://registry.npmjs.org/@ai-sdk/provider-utils/-/provider-utils-5.0.75.tgz",
|
| 170 |
+
"integrity": "sha512-k4yAsFliGZwoTpyTpm3ptxwIeaPvRh5vGh/UWoqDiLLXKoyz4zuBGS7p3ygf8to5Bh4fAOHynotLBsWLIYdu2A==",
|
| 171 |
"license": "Apache-2.0",
|
| 172 |
"dependencies": {
|
| 173 |
+
"@ai-sdk/provider": "4.0.36",
|
| 174 |
"@standard-schema/spec": "^1.1.0",
|
| 175 |
"@workflow/serde": "4.1.0",
|
| 176 |
"eventsource-parser": "^3.0.8",
|
| 177 |
+
"undici": "^7.29.1"
|
| 178 |
},
|
| 179 |
"engines": {
|
| 180 |
"node": ">=22"
|
|
|
|
| 183 |
"zod": "^3.25.76 || ^4.1.8"
|
| 184 |
}
|
| 185 |
},
|
| 186 |
+
"node_modules/@ai-sdk/provider-utils/node_modules/@ai-sdk/provider": {
|
| 187 |
+
"version": "4.0.36",
|
| 188 |
+
"resolved": "https://registry.npmjs.org/@ai-sdk/provider/-/provider-4.0.36.tgz",
|
| 189 |
+
"integrity": "sha512-GqvWZSZdBelCGughlBdOoupT258rgM3YMbaLAS9WSZJ0vlt8c9vm3My+ZT31mYhiReMCMCF2qF2axnGbILpk1A==",
|
| 190 |
+
"license": "Apache-2.0",
|
| 191 |
+
"dependencies": {
|
| 192 |
+
"json-schema": "^0.4.0"
|
| 193 |
+
},
|
| 194 |
+
"engines": {
|
| 195 |
+
"node": ">=22"
|
| 196 |
+
}
|
| 197 |
+
},
|
| 198 |
"node_modules/@ai-sdk/provider-utils/node_modules/undici": {
|
| 199 |
"version": "7.30.1",
|
| 200 |
"resolved": "https://registry.npmjs.org/undici/-/undici-7.30.1.tgz",
|
|
|
|
| 322 |
}
|
| 323 |
},
|
| 324 |
"node_modules/@biomejs/biome": {
|
| 325 |
+
"version": "2.6.5",
|
| 326 |
+
"resolved": "https://registry.npmjs.org/@biomejs/biome/-/biome-2.6.5.tgz",
|
| 327 |
+
"integrity": "sha512-n0sp74QbSVEkWXlH659TKv4oJNylwAvDeZ8K8vt/PbcF1xneuNWe57z/G0Bxfh00cufhWXF6DiStLM8+jcLUpg==",
|
| 328 |
"dev": true,
|
| 329 |
"license": "MIT OR Apache-2.0",
|
| 330 |
"bin": {
|
|
|
|
| 338 |
"url": "https://opencollective.com/biome"
|
| 339 |
},
|
| 340 |
"optionalDependencies": {
|
| 341 |
+
"@biomejs/cli-android-arm64": "2.6.5",
|
| 342 |
+
"@biomejs/cli-android-x64": "2.6.5",
|
| 343 |
+
"@biomejs/cli-darwin-arm64": "2.6.5",
|
| 344 |
+
"@biomejs/cli-darwin-x64": "2.6.5",
|
| 345 |
+
"@biomejs/cli-linux-arm64": "2.6.5",
|
| 346 |
+
"@biomejs/cli-linux-arm64-musl": "2.6.5",
|
| 347 |
+
"@biomejs/cli-linux-x64": "2.6.5",
|
| 348 |
+
"@biomejs/cli-linux-x64-musl": "2.6.5",
|
| 349 |
+
"@biomejs/cli-win32-arm64": "2.6.5",
|
| 350 |
+
"@biomejs/cli-win32-x64": "2.6.5"
|
| 351 |
}
|
| 352 |
},
|
| 353 |
"node_modules/@biomejs/cli-android-arm64": {
|
| 354 |
+
"version": "2.6.5",
|
| 355 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-android-arm64/-/cli-android-arm64-2.6.5.tgz",
|
| 356 |
+
"integrity": "sha512-fkLzc1WRU7wRIUNmAm7cxILb4dd6eESz/VycWvVFKVn6tXx0gFQKdUbPnwT42xys8AEvIJQpFGcK00kURwqorg==",
|
| 357 |
"cpu": [
|
| 358 |
"arm64"
|
| 359 |
],
|
|
|
|
| 368 |
}
|
| 369 |
},
|
| 370 |
"node_modules/@biomejs/cli-android-x64": {
|
| 371 |
+
"version": "2.6.5",
|
| 372 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-android-x64/-/cli-android-x64-2.6.5.tgz",
|
| 373 |
+
"integrity": "sha512-fkZ9dPnDAzAcaPtPCiwqtkub5C9FQK/oGCGaU9sq+DysTaq13pE9FxBIMcEZ3ydn8/FfcMdXi/DDzKDP+qyn2Q==",
|
| 374 |
"cpu": [
|
| 375 |
"x64"
|
| 376 |
],
|
|
|
|
| 385 |
}
|
| 386 |
},
|
| 387 |
"node_modules/@biomejs/cli-darwin-arm64": {
|
| 388 |
+
"version": "2.6.5",
|
| 389 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-darwin-arm64/-/cli-darwin-arm64-2.6.5.tgz",
|
| 390 |
+
"integrity": "sha512-bKd0EOREgnVjeZm4s9knQ+WdPOkWS3JHB+eWGHlx+Pq66bqQA1/KybwU0O+zuyAyXiphSCBcrLruZt3YtnJoaQ==",
|
| 391 |
"cpu": [
|
| 392 |
"arm64"
|
| 393 |
],
|
|
|
|
| 402 |
}
|
| 403 |
},
|
| 404 |
"node_modules/@biomejs/cli-darwin-x64": {
|
| 405 |
+
"version": "2.6.5",
|
| 406 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-darwin-x64/-/cli-darwin-x64-2.6.5.tgz",
|
| 407 |
+
"integrity": "sha512-UkT9BECNOXSiW6SGddjyaRACy362Sc9n3vDaTFkccL0bC2gW7JzM16i4+JKPbEaDn1Sa+oETHV8KungnI/y40Q==",
|
| 408 |
"cpu": [
|
| 409 |
"x64"
|
| 410 |
],
|
|
|
|
| 419 |
}
|
| 420 |
},
|
| 421 |
"node_modules/@biomejs/cli-linux-arm64": {
|
| 422 |
+
"version": "2.6.5",
|
| 423 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-linux-arm64/-/cli-linux-arm64-2.6.5.tgz",
|
| 424 |
+
"integrity": "sha512-H9cVMWBqW/lkFK9NHJdt0bpM2q7PacUDEjQoff9+m0dCV09H/jmBykfTCkbDthHf8+jCuWDJWiQFJ7bDkiGOcA==",
|
| 425 |
"cpu": [
|
| 426 |
"arm64"
|
| 427 |
],
|
|
|
|
| 439 |
}
|
| 440 |
},
|
| 441 |
"node_modules/@biomejs/cli-linux-arm64-musl": {
|
| 442 |
+
"version": "2.6.5",
|
| 443 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-linux-arm64-musl/-/cli-linux-arm64-musl-2.6.5.tgz",
|
| 444 |
+
"integrity": "sha512-kfc3AePW1a0/d/gUpcIz95nf3Ce8YY2/xL0Fk3fLqAhKsotqUi6PN6BaVWss4p6ijWbiuacZ9NQ4NiQtpjbYpw==",
|
| 445 |
"cpu": [
|
| 446 |
"arm64"
|
| 447 |
],
|
|
|
|
| 459 |
}
|
| 460 |
},
|
| 461 |
"node_modules/@biomejs/cli-linux-x64": {
|
| 462 |
+
"version": "2.6.5",
|
| 463 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-linux-x64/-/cli-linux-x64-2.6.5.tgz",
|
| 464 |
+
"integrity": "sha512-e6/WMSN/+FhH+ZOoBevps/euxFQmv6BXqYNuTn1+ITR6O4XYFjHF5PeVSNZnjkgSYmvhcDkfepSX5lXczMijGA==",
|
| 465 |
"cpu": [
|
| 466 |
"x64"
|
| 467 |
],
|
|
|
|
| 479 |
}
|
| 480 |
},
|
| 481 |
"node_modules/@biomejs/cli-linux-x64-musl": {
|
| 482 |
+
"version": "2.6.5",
|
| 483 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-linux-x64-musl/-/cli-linux-x64-musl-2.6.5.tgz",
|
| 484 |
+
"integrity": "sha512-0mHVpLNt7/FugfZ2VPugUIEIiooWIyHGGJUqU9StxamBln9hz+maXRFIECU50jaf5/ZB93veGIViN5VKgw2AEA==",
|
| 485 |
"cpu": [
|
| 486 |
"x64"
|
| 487 |
],
|
|
|
|
| 499 |
}
|
| 500 |
},
|
| 501 |
"node_modules/@biomejs/cli-win32-arm64": {
|
| 502 |
+
"version": "2.6.5",
|
| 503 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-win32-arm64/-/cli-win32-arm64-2.6.5.tgz",
|
| 504 |
+
"integrity": "sha512-uBbfpdR/cH5up1iRJFGAPP97xKMzrgMs685CSrY7NM+Q4Unw7chY0o3JulEvJdmU6c6hGQxwqotMq7DDH0DzxA==",
|
| 505 |
"cpu": [
|
| 506 |
"arm64"
|
| 507 |
],
|
|
|
|
| 516 |
}
|
| 517 |
},
|
| 518 |
"node_modules/@biomejs/cli-win32-x64": {
|
| 519 |
+
"version": "2.6.5",
|
| 520 |
+
"resolved": "https://registry.npmjs.org/@biomejs/cli-win32-x64/-/cli-win32-x64-2.6.5.tgz",
|
| 521 |
+
"integrity": "sha512-07LaCeN9vvmbAAMOMsUjMfL5UbsCDkvVeyMJIgwNRc0xYqhizBGkjVzx05LYEwCXl2AiHcdzqVPR4IOYAsI1QQ==",
|
| 522 |
"cpu": [
|
| 523 |
"x64"
|
| 524 |
],
|
|
|
|
| 858 |
"license": "MIT"
|
| 859 |
},
|
| 860 |
"node_modules/@mantine/carousel": {
|
| 861 |
+
"version": "9.7.5",
|
| 862 |
+
"resolved": "https://registry.npmjs.org/@mantine/carousel/-/carousel-9.7.5.tgz",
|
| 863 |
+
"integrity": "sha512-CVIMYhddQhS355bb4fKa8S0Dml9Frpg6qfgeakhyuQzRVov49vt8Zvr1KDTV0tKs6ktV0I9x5+M87PfUOPaRVg==",
|
| 864 |
"license": "MIT",
|
| 865 |
"peerDependencies": {
|
| 866 |
+
"@mantine/core": "9.7.5",
|
| 867 |
+
"@mantine/hooks": "9.7.5",
|
| 868 |
"embla-carousel": ">=8.0.0",
|
| 869 |
"embla-carousel-react": ">=8.0.0",
|
| 870 |
"react": "^19.2.0",
|
|
|
|
| 872 |
}
|
| 873 |
},
|
| 874 |
"node_modules/@mantine/code-highlight": {
|
| 875 |
+
"version": "9.7.5",
|
| 876 |
+
"resolved": "https://registry.npmjs.org/@mantine/code-highlight/-/code-highlight-9.7.5.tgz",
|
| 877 |
+
"integrity": "sha512-a9YhELhBNuqsVvbH2E40xVeyGrsV1lAZNj5wIaQYUc6MSNnu4Iq7fJVYk6Qt33G1+MCbfujYo3UGMoXoovoahw==",
|
| 878 |
"license": "MIT",
|
| 879 |
"dependencies": {
|
| 880 |
"clsx": "^2.1.1"
|
| 881 |
},
|
| 882 |
"peerDependencies": {
|
| 883 |
+
"@mantine/core": "9.7.5",
|
| 884 |
+
"@mantine/hooks": "9.7.5",
|
| 885 |
"react": "^19.2.0",
|
| 886 |
"react-dom": "^19.2.0"
|
| 887 |
}
|
| 888 |
},
|
| 889 |
"node_modules/@mantine/core": {
|
| 890 |
+
"version": "9.7.5",
|
| 891 |
+
"resolved": "https://registry.npmjs.org/@mantine/core/-/core-9.7.5.tgz",
|
| 892 |
+
"integrity": "sha512-NiOTHpAZssi1Z5ISZBvcXEA4XDn0BWcxwpmmUthFx/eL0CsJ18FXMn/ydRCJmoeMGcIKmRE+pjt90KM70GOsUw==",
|
| 893 |
"license": "MIT",
|
| 894 |
"dependencies": {
|
| 895 |
"@floating-ui/react": "^0.27.20",
|
|
|
|
| 899 |
"type-fest": "^5.10.0"
|
| 900 |
},
|
| 901 |
"peerDependencies": {
|
| 902 |
+
"@mantine/hooks": "9.7.5",
|
| 903 |
"react": "^19.2.0",
|
| 904 |
"react-dom": "^19.2.0"
|
| 905 |
}
|
| 906 |
},
|
| 907 |
"node_modules/@mantine/form": {
|
| 908 |
+
"version": "9.7.5",
|
| 909 |
+
"resolved": "https://registry.npmjs.org/@mantine/form/-/form-9.7.5.tgz",
|
| 910 |
+
"integrity": "sha512-DbYsbSak59cwxFGRGBScWMzzVWKSUNPAJLMS8FDEyzpAJvywj4Bx07syFx1DgUJaIeMfjIERAVraPe3T2WxdJg==",
|
| 911 |
"license": "MIT",
|
| 912 |
"dependencies": {
|
| 913 |
"@standard-schema/spec": "^1.1.0",
|
|
|
|
| 919 |
}
|
| 920 |
},
|
| 921 |
"node_modules/@mantine/hooks": {
|
| 922 |
+
"version": "9.7.5",
|
| 923 |
+
"resolved": "https://registry.npmjs.org/@mantine/hooks/-/hooks-9.7.5.tgz",
|
| 924 |
+
"integrity": "sha512-+3cNisLOUkzF77SjRKkG/V+svZii9WDGvcwlaxFH58b14MNohz1zWbV+WfjZt9qfzrIPPUOedks/z6PMo5rAFA==",
|
| 925 |
"license": "MIT",
|
| 926 |
"peerDependencies": {
|
| 927 |
"react": "^19.2.0"
|
| 928 |
}
|
| 929 |
},
|
| 930 |
"node_modules/@mantine/notifications": {
|
| 931 |
+
"version": "9.7.5",
|
| 932 |
+
"resolved": "https://registry.npmjs.org/@mantine/notifications/-/notifications-9.7.5.tgz",
|
| 933 |
+
"integrity": "sha512-S7jApMXZhXmKLlwvvp5yfQ3AqfxQN9Y+GQK82f5vNWKLZVgQ3HdicpeSh1DmsNWdB4sJr0oIhWliCUPAC8ctyA==",
|
| 934 |
"license": "MIT",
|
| 935 |
"dependencies": {
|
| 936 |
+
"@mantine/store": "9.7.5",
|
| 937 |
"react-transition-group": "4.4.5"
|
| 938 |
},
|
| 939 |
"peerDependencies": {
|
| 940 |
+
"@mantine/core": "9.7.5",
|
| 941 |
+
"@mantine/hooks": "9.7.5",
|
| 942 |
"react": "^19.2.0",
|
| 943 |
"react-dom": "^19.2.0"
|
| 944 |
}
|
| 945 |
},
|
| 946 |
"node_modules/@mantine/store": {
|
| 947 |
+
"version": "9.7.5",
|
| 948 |
+
"resolved": "https://registry.npmjs.org/@mantine/store/-/store-9.7.5.tgz",
|
| 949 |
+
"integrity": "sha512-A+5/Y7+2EkkB8ByWSIxUpaXA0O7jFlhBKNrB8IAAkl0WADEhDh6UxDV/88p9fUfGP17vbWskEyMLT9t5vWmu4g==",
|
| 950 |
"license": "MIT",
|
| 951 |
"peerDependencies": {
|
| 952 |
"react": "^19.2.0"
|
|
|
|
| 2950 |
}
|
| 2951 |
},
|
| 2952 |
"node_modules/@vitest/coverage-v8": {
|
| 2953 |
+
"version": "5.1.0",
|
| 2954 |
+
"resolved": "https://registry.npmjs.org/@vitest/coverage-v8/-/coverage-v8-5.1.0.tgz",
|
| 2955 |
+
"integrity": "sha512-OY9aShBQG5KOIw2zCo9fDoCHI9RHliNHJ//nM7m6jGg41Hyke9JOGWmeMlK74xYErsxMoK8mIB3gAeEyFVA7LA==",
|
| 2956 |
"dev": true,
|
| 2957 |
"license": "MIT",
|
| 2958 |
"dependencies": {
|
| 2959 |
"@bcoe/v8-coverage": "^1.0.2",
|
| 2960 |
+
"@vitest/istanbul-lib-coverage": "^2.1.0",
|
| 2961 |
+
"@vitest/istanbul-lib-report": "^2.1.0",
|
| 2962 |
"ast-v8-to-istanbul": "^1.0.7",
|
| 2963 |
"magicast": "^0.5.5",
|
| 2964 |
"obug": "^3.0.0",
|
|
|
|
| 2969 |
"url": "https://opencollective.com/vitest"
|
| 2970 |
},
|
| 2971 |
"peerDependencies": {
|
| 2972 |
+
"@vitest/browser": "5.1.0",
|
| 2973 |
+
"vitest": "5.1.0"
|
| 2974 |
},
|
| 2975 |
"peerDependenciesMeta": {
|
| 2976 |
"@vitest/browser": {
|
|
|
|
| 2979 |
}
|
| 2980 |
},
|
| 2981 |
"node_modules/@vitest/istanbul-lib-coverage": {
|
| 2982 |
+
"version": "2.1.0",
|
| 2983 |
+
"resolved": "https://registry.npmjs.org/@vitest/istanbul-lib-coverage/-/istanbul-lib-coverage-2.1.0.tgz",
|
| 2984 |
+
"integrity": "sha512-WOjuS+xcjSIy7ruYBQ+9BUcV7XZQ0FyVcbWbWxW9TRx/kLGx5SQaeFRIFDJxjNplb6YsT7SpKo2zuXXUo91OIQ==",
|
| 2985 |
"dev": true,
|
| 2986 |
"license": "MIT",
|
| 2987 |
"engines": {
|
|
|
|
| 2989 |
}
|
| 2990 |
},
|
| 2991 |
"node_modules/@vitest/istanbul-lib-report": {
|
| 2992 |
+
"version": "2.1.0",
|
| 2993 |
+
"resolved": "https://registry.npmjs.org/@vitest/istanbul-lib-report/-/istanbul-lib-report-2.1.0.tgz",
|
| 2994 |
+
"integrity": "sha512-+ZFta25nWtGCOBG9/IET8w1RO5Oawsd6H8Uf6AJ9EA/StFuKVtKAEJEJ6uRt84OH+T4YkI0ASmIHAhRA+0KVxw==",
|
| 2995 |
"dev": true,
|
| 2996 |
"license": "MIT",
|
| 2997 |
"dependencies": {
|
| 2998 |
+
"@vitest/istanbul-lib-coverage": "2.1.0"
|
| 2999 |
},
|
| 3000 |
"engines": {
|
| 3001 |
"node": ">=22"
|
| 3002 |
}
|
| 3003 |
},
|
| 3004 |
"node_modules/@vitest/mocker": {
|
| 3005 |
+
"version": "5.1.0",
|
| 3006 |
+
"resolved": "https://registry.npmjs.org/@vitest/mocker/-/mocker-5.1.0.tgz",
|
| 3007 |
+
"integrity": "sha512-5umB9/jlWcqVS6fsrFmv5Gdnpu6wAUio/lEOwAmvB7v+lFo/IW2B9FhAD0p19GCULTgMzLCdUmuCDCnAzWgppQ==",
|
| 3008 |
"dev": true,
|
| 3009 |
"license": "MIT",
|
| 3010 |
"dependencies": {
|
| 3011 |
"@jridgewell/trace-mapping": "0.3.31",
|
| 3012 |
+
"@vitest/spy": "5.1.0",
|
| 3013 |
"estree-walker": "^3.0.3",
|
| 3014 |
"magic-string": "^1.4.3"
|
| 3015 |
},
|
|
|
|
| 3030 |
}
|
| 3031 |
},
|
| 3032 |
"node_modules/@vitest/spy": {
|
| 3033 |
+
"version": "5.1.0",
|
| 3034 |
+
"resolved": "https://registry.npmjs.org/@vitest/spy/-/spy-5.1.0.tgz",
|
| 3035 |
+
"integrity": "sha512-2DPO5OMHWKrBngC+fCmTQqemNF0gidRhp3lxbBmRhW+M1XJrfgPpqNqWVevXtn2f2ufD5ZaR1B6B641aAlcX5A==",
|
| 3036 |
"dev": true,
|
| 3037 |
"license": "MIT",
|
| 3038 |
"funding": {
|
|
|
|
| 3061 |
}
|
| 3062 |
},
|
| 3063 |
"node_modules/ai": {
|
| 3064 |
+
"version": "7.0.163",
|
| 3065 |
+
"resolved": "https://registry.npmjs.org/ai/-/ai-7.0.163.tgz",
|
| 3066 |
+
"integrity": "sha512-nm8+qn36NQXO4Hjfj7SlbcNRhUyNXKtdoS9xhXr/z92lvdGFuPkzHmA2tFzhaUQy13IitNYrcw2QSNpEJC/iSA==",
|
| 3067 |
"license": "Apache-2.0",
|
| 3068 |
"dependencies": {
|
| 3069 |
+
"@ai-sdk/gateway": "4.0.136",
|
| 3070 |
+
"@ai-sdk/provider": "4.0.36",
|
| 3071 |
+
"@ai-sdk/provider-utils": "5.0.75"
|
| 3072 |
},
|
| 3073 |
"engines": {
|
| 3074 |
"node": ">=22"
|
|
|
|
| 3077 |
"zod": "^3.25.76 || ^4.1.8"
|
| 3078 |
}
|
| 3079 |
},
|
| 3080 |
+
"node_modules/ai/node_modules/@ai-sdk/provider": {
|
| 3081 |
+
"version": "4.0.36",
|
| 3082 |
+
"resolved": "https://registry.npmjs.org/@ai-sdk/provider/-/provider-4.0.36.tgz",
|
| 3083 |
+
"integrity": "sha512-GqvWZSZdBelCGughlBdOoupT258rgM3YMbaLAS9WSZJ0vlt8c9vm3My+ZT31mYhiReMCMCF2qF2axnGbILpk1A==",
|
| 3084 |
+
"license": "Apache-2.0",
|
| 3085 |
+
"dependencies": {
|
| 3086 |
+
"json-schema": "^0.4.0"
|
| 3087 |
+
},
|
| 3088 |
+
"engines": {
|
| 3089 |
+
"node": ">=22"
|
| 3090 |
+
}
|
| 3091 |
+
},
|
| 3092 |
"node_modules/ansi-regex": {
|
| 3093 |
"version": "5.0.1",
|
| 3094 |
"resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-5.0.1.tgz",
|
|
|
|
| 6767 |
}
|
| 6768 |
},
|
| 6769 |
"node_modules/rate-limiter-flexible": {
|
| 6770 |
+
"version": "11.2.4",
|
| 6771 |
+
"resolved": "https://registry.npmjs.org/rate-limiter-flexible/-/rate-limiter-flexible-11.2.4.tgz",
|
| 6772 |
+
"integrity": "sha512-z28uTq33lBUAszaKtX9Vq4Fwemf6Lfr06tw3jo5pkqjsxfFyqk4yx5T4bEHAeLr39ZS6VeVtfaIu35QhVNKAvA==",
|
| 6773 |
"license": "ISC"
|
| 6774 |
},
|
| 6775 |
"node_modules/react": {
|
|
|
|
| 7122 |
}
|
| 7123 |
},
|
| 7124 |
"node_modules/rollup-plugin-visualizer": {
|
| 7125 |
+
"version": "7.1.2",
|
| 7126 |
+
"resolved": "https://registry.npmjs.org/rollup-plugin-visualizer/-/rollup-plugin-visualizer-7.1.2.tgz",
|
| 7127 |
+
"integrity": "sha512-+yrEz9LJ7fxd7T0yj4GPvcKHKfJNzTzKT7FHiF3f/US+7EaWlgB8PoGbUOlobD4Al8v6TeCCl/Fgbov5FV/DaA==",
|
| 7128 |
"license": "MIT",
|
| 7129 |
"dependencies": {
|
| 7130 |
"open": "^11.0.0",
|
|
|
|
| 7566 |
}
|
| 7567 |
},
|
| 7568 |
"node_modules/tldts": {
|
| 7569 |
+
"version": "7.4.19",
|
| 7570 |
+
"resolved": "https://registry.npmjs.org/tldts/-/tldts-7.4.19.tgz",
|
| 7571 |
+
"integrity": "sha512-48dH6aSpQnpbzDUc4vIQZ1VjB8tkK+e8c1yx05/SZK+rkssgjEOlxDd/yU8/a4vGYQ9vRgJr9dwMq9TJ0b3X2Q==",
|
| 7572 |
"license": "MIT",
|
| 7573 |
"dependencies": {
|
| 7574 |
+
"tldts-core": "^7.4.19"
|
| 7575 |
},
|
| 7576 |
"bin": {
|
| 7577 |
"tldts": "bin/cli.js"
|
| 7578 |
}
|
| 7579 |
},
|
| 7580 |
"node_modules/tldts-core": {
|
| 7581 |
+
"version": "7.4.20",
|
| 7582 |
+
"resolved": "https://registry.npmjs.org/tldts-core/-/tldts-core-7.4.20.tgz",
|
| 7583 |
+
"integrity": "sha512-C630TWXZCZmLq0F2bW4QqwnaaxCQ/svpAqvJwnC4JvgEM5ohhBnGsEzTLSOXcN+xjz+kEaHS5uoWIw2klnqd0A==",
|
| 7584 |
"license": "MIT"
|
| 7585 |
},
|
| 7586 |
"node_modules/tough-cookie": {
|
|
|
|
| 7971 |
}
|
| 7972 |
},
|
| 7973 |
"node_modules/vitest": {
|
| 7974 |
+
"version": "5.1.0",
|
| 7975 |
+
"resolved": "https://registry.npmjs.org/vitest/-/vitest-5.1.0.tgz",
|
| 7976 |
+
"integrity": "sha512-MzmNFnrwa8v8lF1bLxw3UJ4EREwVOVb75VqJZmUDRjJ6EluK/x9th7qTbaeDk41WxE1GxWqmTXcdsfobfH5sOA==",
|
| 7977 |
"dev": true,
|
| 7978 |
"license": "MIT",
|
| 7979 |
"dependencies": {
|
| 7980 |
"@types/chai": "^5.2.2",
|
| 7981 |
+
"@vitest/mocker": "5.1.0",
|
| 7982 |
"chai": "^6.3.0",
|
| 7983 |
"es-module-lexer": "^2.3.2",
|
| 7984 |
"expect-type": "^1.4.0",
|
|
|
|
| 8004 |
"@edge-runtime/vm": "*",
|
| 8005 |
"@opentelemetry/api": "^1.9.0",
|
| 8006 |
"@types/node": "^22.0.0 || >=24.0.0",
|
| 8007 |
+
"@vitest/browser-playwright": "5.1.0",
|
| 8008 |
+
"@vitest/browser-preview": "5.1.0",
|
| 8009 |
"@vitest/browser-webdriverio": "^5.0.0-beta.5 || >=5.0.0",
|
| 8010 |
+
"@vitest/coverage-istanbul": "5.1.0",
|
| 8011 |
+
"@vitest/coverage-v8": "5.1.0",
|
| 8012 |
+
"@vitest/ui": "5.1.0",
|
| 8013 |
"happy-dom": "*",
|
| 8014 |
"jsdom": "*",
|
| 8015 |
"vite": "^6.4.0 || ^7.0.0 || ^8.0.0"
|
package.json
CHANGED
|
@@ -33,7 +33,8 @@
|
|
| 33 |
"changelog-guard": "node scripts/changelog-guard.cjs",
|
| 34 |
"license-check": "node scripts/license-check.cjs",
|
| 35 |
"native-module-check": "node scripts/native-module-integrity.cjs",
|
| 36 |
-
"wllama-catalog-check": "node scripts/wllama-catalog-check.cjs"
|
|
|
|
| 37 |
},
|
| 38 |
"allowScripts": {
|
| 39 |
"onnxruntime-node": false,
|
|
|
|
| 33 |
"changelog-guard": "node scripts/changelog-guard.cjs",
|
| 34 |
"license-check": "node scripts/license-check.cjs",
|
| 35 |
"native-module-check": "node scripts/native-module-integrity.cjs",
|
| 36 |
+
"wllama-catalog-check": "node scripts/wllama-catalog-check.cjs",
|
| 37 |
+
"bench:wllama": "node scripts/wllama-bench/run.mjs"
|
| 38 |
},
|
| 39 |
"allowScripts": {
|
| 40 |
"onnxruntime-node": false,
|