diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/.pytest_cache/.gitignore devai_phase4/.pytest_cache/.gitignore --- devai_phase4_orig/.pytest_cache/.gitignore 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/.pytest_cache/.gitignore 2026-07-06 10:28:54.061919928 +0000 @@ -0,0 +1,2 @@ +# Created by pytest automatically. +* diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/.pytest_cache/CACHEDIR.TAG devai_phase4/.pytest_cache/CACHEDIR.TAG --- devai_phase4_orig/.pytest_cache/CACHEDIR.TAG 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/.pytest_cache/CACHEDIR.TAG 2026-07-06 10:28:54.061919928 +0000 @@ -0,0 +1,4 @@ +Signature: 8a477f597d28d172789f06886806bc55 +# This file is a cache directory tag created by pytest. +# For information about cache directory tags, see: +# https://bford.info/cachedir/spec.html diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/.pytest_cache/README.md devai_phase4/.pytest_cache/README.md --- devai_phase4_orig/.pytest_cache/README.md 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/.pytest_cache/README.md 2026-07-06 10:28:54.061919928 +0000 @@ -0,0 +1,8 @@ +# pytest cache directory # + +This directory contains data from the pytest's cache plugin, +which provides the `--lf` and `--ff` options, as well as the `cache` fixture. + +**Do not** commit this to version control. + +See [the docs](https://docs.pytest.org/en/stable/how-to/cache.html) for more information. diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/.pytest_cache/v/cache/nodeids devai_phase4/.pytest_cache/v/cache/nodeids --- devai_phase4_orig/.pytest_cache/v/cache/nodeids 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/.pytest_cache/v/cache/nodeids 2026-07-06 10:28:54.061919928 +0000 @@ -0,0 +1,5 @@ +[ + "tests/test_phase4_docs_presets.py::test_guides_exist_and_share_same_core_sections", + "tests/test_phase4_docs_presets.py::test_onboarding_and_ui_hooks_present", + "tests/test_phase4_docs_presets.py::test_provider_presets_catalog_exists" +] \ No newline at end of file diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/CHANGELOG_v4.7.md devai_phase4/CHANGELOG_v4.7.md --- devai_phase4_orig/CHANGELOG_v4.7.md 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/CHANGELOG_v4.7.md 2026-07-06 10:28:32.013919928 +0000 @@ -0,0 +1,24 @@ +# DevAI Studio v4.7 Combined Changelog + +## Phase 1 — BYOM foundation +- Added secure encrypted storage for custom provider API keys. +- Added custom router with provider-aware request building. +- Added quota/error handling for custom models. + +## Phase 2 — Management UI +- Added three-dot custom model management panel. +- Added add/edit/delete flows for custom models. +- Added routing rules editor and quota card visuals. + +## Phase 3 — Quota + fallback behavior +- Added quota exhaustion handling. +- Added local fallback / ask-for-new-key recovery paths. +- Added extended tests and screenshots for save/edit/rules flows. + +## Phase 4 — Documentation & polish +- Added `data/provider_presets.json` with curated providers and rule sets. +- Added `ui/onboarding_tour.py` with driver.js first-run tour. +- Added Hindi and English BYOM user guides. +- Added migration guide and searchable troubleshooting KB. +- Added About-tab docs hub and restart-tour control. +- Added release assets, extra screenshots, diff patch, and a demo video storyboard. diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/PHASE4_DIFF.patch devai_phase4/PHASE4_DIFF.patch --- devai_phase4_orig/PHASE4_DIFF.patch 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/PHASE4_DIFF.patch 2026-07-06 10:29:01.289919928 +0000 @@ -0,0 +1,64 @@ +diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/.pytest_cache/.gitignore devai_phase4/.pytest_cache/.gitignore +--- devai_phase4_orig/.pytest_cache/.gitignore 1970-01-01 00:00:00.000000000 +0000 ++++ devai_phase4/.pytest_cache/.gitignore 2026-07-06 10:28:54.061919928 +0000 +@@ -0,0 +1,2 @@ ++# Created by pytest automatically. ++* +diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/.pytest_cache/CACHEDIR.TAG devai_phase4/.pytest_cache/CACHEDIR.TAG +--- devai_phase4_orig/.pytest_cache/CACHEDIR.TAG 1970-01-01 00:00:00.000000000 +0000 ++++ devai_phase4/.pytest_cache/CACHEDIR.TAG 2026-07-06 10:28:54.061919928 +0000 +@@ -0,0 +1,4 @@ ++Signature: 8a477f597d28d172789f06886806bc55 ++# This file is a cache directory tag created by pytest. ++# For information about cache directory tags, see: ++# https://bford.info/cachedir/spec.html +diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/.pytest_cache/README.md devai_phase4/.pytest_cache/README.md +--- devai_phase4_orig/.pytest_cache/README.md 1970-01-01 00:00:00.000000000 +0000 ++++ devai_phase4/.pytest_cache/README.md 2026-07-06 10:28:54.061919928 +0000 +@@ -0,0 +1,8 @@ ++# pytest cache directory # ++ ++This directory contains data from the pytest's cache plugin, ++which provides the `--lf` and `--ff` options, as well as the `cache` fixture. ++ ++**Do not** commit this to version control. ++ ++See [the docs](https://docs.pytest.org/en/stable/how-to/cache.html) for more information. +diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/.pytest_cache/v/cache/nodeids devai_phase4/.pytest_cache/v/cache/nodeids +--- devai_phase4_orig/.pytest_cache/v/cache/nodeids 1970-01-01 00:00:00.000000000 +0000 ++++ devai_phase4/.pytest_cache/v/cache/nodeids 2026-07-06 10:28:54.061919928 +0000 +@@ -0,0 +1,5 @@ ++[ ++ "tests/test_phase4_docs_presets.py::test_guides_exist_and_share_same_core_sections", ++ "tests/test_phase4_docs_presets.py::test_onboarding_and_ui_hooks_present", ++ "tests/test_phase4_docs_presets.py::test_provider_presets_catalog_exists" ++] +\ No newline at end of file +diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/CHANGELOG_v4.7.md devai_phase4/CHANGELOG_v4.7.md +--- devai_phase4_orig/CHANGELOG_v4.7.md 1970-01-01 00:00:00.000000000 +0000 ++++ devai_phase4/CHANGELOG_v4.7.md 2026-07-06 10:28:32.013919928 +0000 +@@ -0,0 +1,24 @@ ++# DevAI Studio v4.7 Combined Changelog ++ ++## Phase 1 — BYOM foundation ++- Added secure encrypted storage for custom provider API keys. ++- Added custom router with provider-aware request building. ++- Added quota/error handling for custom models. ++ ++## Phase 2 — Management UI ++- Added three-dot custom model management panel. ++- Added add/edit/delete flows for custom models. ++- Added routing rules editor and quota card visuals. ++ ++## Phase 3 — Quota + fallback behavior ++- Added quota exhaustion handling. ++- Added local fallback / ask-for-new-key recovery paths. ++- Added extended tests and screenshots for save/edit/rules flows. ++ ++## Phase 4 — Documentation & polish ++- Added `data/provider_presets.json` with curated providers and rule sets. ++- Added `ui/onboarding_tour.py` with driver.js first-run tour. ++- Added Hindi and English BYOM user guides. ++- Added migration guide and searchable troubleshooting KB. ++- Added About-tab docs hub and restart-tour control. ++- Added release assets, extra screenshots, diff patch, and a demo video storyboard. diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/data/provider_presets.json devai_phase4/data/provider_presets.json --- devai_phase4_orig/data/provider_presets.json 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/data/provider_presets.json 2026-07-06 10:26:42.069919928 +0000 @@ -0,0 +1,286 @@ +{ + "presets": [ + { + "provider": "openrouter", + "display_name": "OpenRouter (100+ models)", + "description": "एक API key से 100+ models — Claude, GPT, Gemini, Llama सब", + "signup_url": "https://openrouter.ai/keys", + "docs_url": "https://openrouter.ai/docs", + "popular_models": [ + { + "model_id": "anthropic/claude-3.5-sonnet", + "display_name": "Claude 3.5 Sonnet", + "best_for": [ + "security", + "refactor", + "complex reasoning" + ], + "approx_cost_per_1k_tokens": { + "input": 0.003, + "output": 0.015 + }, + "context_window": 200000 + }, + { + "model_id": "anthropic/claude-3.5-haiku", + "display_name": "Claude 3.5 Haiku (fast+cheap)", + "best_for": [ + "bug_fix", + "test_gen" + ], + "approx_cost_per_1k_tokens": { + "input": 0.001, + "output": 0.005 + }, + "context_window": 200000 + }, + { + "model_id": "openai/gpt-4o", + "display_name": "GPT-4o", + "best_for": [ + "feature", + "new_app" + ], + "approx_cost_per_1k_tokens": { + "input": 0.0025, + "output": 0.01 + }, + "context_window": 128000 + }, + { + "model_id": "google/gemini-2.0-flash-exp:free", + "display_name": "Gemini 2.0 Flash (Free tier)", + "best_for": [ + "any" + ], + "approx_cost_per_1k_tokens": { + "input": 0, + "output": 0 + }, + "context_window": 1000000, + "notes": "Free but rate-limited to 15 req/min" + }, + { + "model_id": "deepseek/deepseek-r1", + "display_name": "DeepSeek R1 (reasoning)", + "best_for": [ + "refactor", + "security" + ], + "approx_cost_per_1k_tokens": { + "input": 0.00055, + "output": 0.00219 + }, + "context_window": 64000 + } + ] + }, + { + "provider": "anthropic", + "display_name": "Anthropic (direct)", + "description": "Claude models directly — lowest latency, no OpenRouter markup", + "signup_url": "https://console.anthropic.com/", + "docs_url": "https://docs.anthropic.com/", + "popular_models": [ + { + "model_id": "claude-3-5-sonnet-20241022", + "display_name": "Claude 3.5 Sonnet (latest)", + "best_for": [ + "security", + "refactor", + "complex reasoning" + ], + "approx_cost_per_1k_tokens": { + "input": 0.003, + "output": 0.015 + }, + "context_window": 200000 + }, + { + "model_id": "claude-3-5-haiku-20241022", + "display_name": "Claude 3.5 Haiku", + "best_for": [ + "bug_fix", + "test_gen" + ], + "approx_cost_per_1k_tokens": { + "input": 0.0008, + "output": 0.004 + }, + "context_window": 200000 + } + ] + }, + { + "provider": "openai", + "display_name": "OpenAI (direct)", + "signup_url": "https://platform.openai.com/api-keys", + "docs_url": "https://platform.openai.com/docs", + "popular_models": [ + { + "model_id": "gpt-4o", + "display_name": "GPT-4o", + "best_for": [ + "feature", + "new_app" + ], + "approx_cost_per_1k_tokens": { + "input": 0.0025, + "output": 0.01 + } + }, + { + "model_id": "gpt-4o-mini", + "display_name": "GPT-4o Mini", + "best_for": [ + "bug_fix" + ], + "approx_cost_per_1k_tokens": { + "input": 0.00015, + "output": 0.0006 + } + } + ] + }, + { + "provider": "groq", + "display_name": "Groq (fastest inference)", + "description": "Ultra-fast LPU inference — best for iterative dev loops", + "signup_url": "https://console.groq.com/keys", + "popular_models": [ + { + "model_id": "llama-3.3-70b-versatile", + "display_name": "Llama 3.3 70B", + "best_for": [ + "any" + ], + "approx_cost_per_1k_tokens": { + "input": 0.00059, + "output": 0.00079 + }, + "notes": "~500 tokens/sec — feels instant" + }, + { + "model_id": "deepseek-r1-distill-llama-70b", + "display_name": "DeepSeek R1 Distill", + "best_for": [ + "refactor" + ], + "approx_cost_per_1k_tokens": { + "input": 0.00075, + "output": 0.00099 + } + } + ] + }, + { + "provider": "together", + "display_name": "Together AI", + "signup_url": "https://api.together.xyz/settings/api-keys", + "popular_models": [ + { + "model_id": "Qwen/Qwen2.5-Coder-32B-Instruct", + "display_name": "Qwen 2.5 Coder 32B", + "best_for": [ + "bug_fix", + "feature" + ], + "approx_cost_per_1k_tokens": { + "input": 0.0008, + "output": 0.0008 + } + } + ] + }, + { + "provider": "deepinfra", + "display_name": "DeepInfra", + "signup_url": "https://deepinfra.com/dash/api_keys", + "popular_models": [ + { + "model_id": "meta-llama/Meta-Llama-3.1-405B-Instruct", + "display_name": "Llama 3.1 405B", + "best_for": [ + "complex reasoning" + ], + "approx_cost_per_1k_tokens": { + "input": 0.0027, + "output": 0.0027 + } + } + ] + } + ], + "example_rule_sets": [ + { + "name": "Cost-conscious (Free tier maximizer)", + "description": "जितना हो सके free models — सिर्फ़ जटिल tasks पर paid", + "rules": [ + { + "model": "google/gemini-2.0-flash-exp:free (OpenRouter)", + "task_types": [ + "any" + ], + "min_complexity": 0 + }, + { + "model": "anthropic/claude-3.5-sonnet (OpenRouter)", + "task_types": [ + "security", + "refactor" + ], + "min_complexity": 8 + } + ] + }, + { + "name": "Speed-focused (dev iteration)", + "description": "Groq की LPU speed का फायदा — instant feedback loop", + "rules": [ + { + "model": "llama-3.3-70b-versatile (Groq)", + "task_types": [ + "bug_fix", + "test_gen" + ], + "min_complexity": 0 + } + ] + }, + { + "name": "Quality-first (enterprise)", + "description": "हर task पर best-in-class model — cost secondary", + "rules": [ + { + "model": "claude-3-5-sonnet (Anthropic direct)", + "task_types": [ + "security", + "refactor", + "new_app" + ], + "always_use": true + }, + { + "model": "gpt-4o (OpenAI direct)", + "task_types": [ + "feature" + ], + "min_complexity": 5 + } + ] + }, + { + "name": "Privacy-hybrid", + "description": "Local first, cloud सिर्फ़ security audits पर", + "rules": [ + { + "model": "claude-3-5-sonnet", + "task_types": [ + "security" + ], + "min_complexity": 7 + } + ] + } + ] +} diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/demo/README_demo_script.md devai_phase4/demo/README_demo_script.md --- devai_phase4_orig/demo/README_demo_script.md 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/demo/README_demo_script.md 2026-07-06 10:28:33.917919928 +0000 @@ -0,0 +1,10 @@ +# Demo storyboard script + +Use the bundled MP4 as a silent walkthrough. Suggested live narration order: +1. Open Advanced options. +2. Start BYOM via the three-dot menu. +3. Choose provider preset and popular model. +4. Load an example rule set. +5. Test connection and save the model. +6. Show quota cards and fallback behavior. +7. Open About tab docs hub. Binary files devai_phase4_orig/demo/devai_studio_v4.7_phase4_demo.mp4 and devai_phase4/demo/devai_studio_v4.7_phase4_demo.mp4 differ diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/demo/slides.txt devai_phase4/demo/slides.txt --- devai_phase4_orig/demo/slides.txt 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/demo/slides.txt 2026-07-06 10:28:32.237919928 +0000 @@ -0,0 +1,29 @@ +file '/home/user/work/devai_phase4/demo/story_frames/01.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/02.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/03.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/04.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/05.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/06.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/07.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/08.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/09.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/10.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/11.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/12.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/13.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/14.png' +duration 22 +file '/home/user/work/devai_phase4/demo/story_frames/14.png' Binary files devai_phase4_orig/demo/story_frames/01.png and devai_phase4/demo/story_frames/01.png differ Binary files devai_phase4_orig/demo/story_frames/02.png and devai_phase4/demo/story_frames/02.png differ Binary files devai_phase4_orig/demo/story_frames/03.png and devai_phase4/demo/story_frames/03.png differ Binary files devai_phase4_orig/demo/story_frames/04.png and devai_phase4/demo/story_frames/04.png differ Binary files devai_phase4_orig/demo/story_frames/05.png and devai_phase4/demo/story_frames/05.png differ Binary files devai_phase4_orig/demo/story_frames/06.png and devai_phase4/demo/story_frames/06.png differ Binary files devai_phase4_orig/demo/story_frames/07.png and devai_phase4/demo/story_frames/07.png differ Binary files devai_phase4_orig/demo/story_frames/08.png and devai_phase4/demo/story_frames/08.png differ Binary files devai_phase4_orig/demo/story_frames/09.png and devai_phase4/demo/story_frames/09.png differ Binary files devai_phase4_orig/demo/story_frames/10.png and devai_phase4/demo/story_frames/10.png differ Binary files devai_phase4_orig/demo/story_frames/11.png and devai_phase4/demo/story_frames/11.png differ Binary files devai_phase4_orig/demo/story_frames/12.png and devai_phase4/demo/story_frames/12.png differ Binary files devai_phase4_orig/demo/story_frames/13.png and devai_phase4/demo/story_frames/13.png differ Binary files devai_phase4_orig/demo/story_frames/14.png and devai_phase4/demo/story_frames/14.png differ diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/docs/BYOM_USER_GUIDE_EN.md devai_phase4/docs/BYOM_USER_GUIDE_EN.md --- devai_phase4_orig/docs/BYOM_USER_GUIDE_EN.md 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/docs/BYOM_USER_GUIDE_EN.md 2026-07-06 10:28:32.013919928 +0000 @@ -0,0 +1,398 @@ +# 🎨 DevAI Studio v4.7 — Bring Your Own Model + +## Table of contents +1. [What is this feature](#1-what-is-this-feature) +2. [Setup — first model in 5 minutes](#2-setup--first-model-in-5-minutes) +3. [Choosing a provider](#3-choosing-a-provider) +4. [Task Rules — which model runs when](#4-task-rules--which-model-runs-when) +5. [Cost Management](#5-cost-management) +6. [Quota Exhaustion — what happens](#6-quota-exhaustion--what-happens) +7. [Security & Privacy](#7-security--privacy) +8. [Troubleshooting](#8-troubleshooting) +9. [Advanced: Custom endpoints](#9-advanced-custom-endpoints) +10. [FAQ](#10-faq) + +## 1. What is this feature? + +DevAI Studio v4.7 now has three model tiers: local, built-in cloud, and custom BYOM. This feature lets you attach your own API keys to production-grade models. + +### Tier summary +- **Local models** — free, offline, private, but speed depends on your hardware. +- **Built-in cloud models** — zero setup, quick start, but shared limits. +- **Custom models** — latest Claude, GPT, Gemini, Groq, Together, DeepInfra, or a custom endpoint; quota and billing belong to you. + +### Local models (always available) +- Qwen2.5-Coder-7B → coding tasks +- DeepSeek-R1-Distill-8B → complex reasoning +- InternLM3-8B → security audits +- Llama-3.2-3B → general chat +- MiniCPM-V-2.6 → vision (screenshot → code) +- Phi-3.5-mini → fast small tasks + +### What BYOM adds +- Latest flagship models without waiting for an app release. +- Personal quota instead of shared free-tier bottlenecks. +- Task-aware routing: bug fix, feature, refactor, security, tests, deployment, new app. +- Fallback control: pause and ask for a new key, or continue on a local model. + +## 2. Setup — first model in 5 minutes + +### Step 1 — Set a master key +Enter a strong passphrase in the Settings tab. It encrypts every stored API key. Save it in a password manager. If you lose it, re-add the models. + +### Step 2 — Pick a provider +- Fastest start: **OpenRouter** +- Best direct Claude latency: **Anthropic** +- Best multimodal generalist: **OpenAI GPT-4o** +- Fast iterative dev loops: **Groq** + +### Step 3 — Create an API key +Generate a key from the provider dashboard and verify billing or quota before pasting it into the app. + +### Step 4 — Studio > Advanced options > ⋮ > Add model +In the v4.7 UX you can choose a provider preset, then choose a popular model. The UI auto-fills model_id, base URL and a rough cost estimate. + +### Step 5 — Choose rules and fallback +- Task types: where the model is eligible. +- Min complexity: 0 for broad use, 8-10 for hard jobs only. +- Always use: overrides all routing rules. +- Fallback: ask for a new key or switch to a local model. + +### Step 6 — Test Connection +Once you see a green ✅, click Save Model and run a small validation task. + +### Recommended first prompt +```text +Audit this repository for SQL injection, XSS, hardcoded secrets, insecure dependencies and auth flaws. +``` + +## 3. Choosing a provider + +| Provider | Why choose it | Popular models | +|---|---|---| +| OpenRouter (100+ models) | एक API key से 100+ models — Claude, GPT, Gemini, Llama सब | Claude 3.5 Sonnet, Claude 3.5 Haiku (fast+cheap), GPT-4o | +| Anthropic (direct) | Claude models directly — lowest latency, no OpenRouter markup | Claude 3.5 Sonnet (latest), Claude 3.5 Haiku | +| OpenAI (direct) | OpenAI-compatible endpoint | GPT-4o, GPT-4o Mini | +| Groq (fastest inference) | Ultra-fast LPU inference — best for iterative dev loops | Llama 3.3 70B, DeepSeek R1 Distill | +| Together AI | OpenAI-compatible endpoint | Qwen 2.5 Coder 32B | +| DeepInfra | OpenAI-compatible endpoint | Llama 3.1 405B | + +### Quick recommendations +- Budget-sensitive solo developer: OpenRouter + free/cheap presets +- Security-review heavy team: Anthropic direct Sonnet +- UI + multimodal heavy team: GPT-4o +- Tight iteration loops: Groq Llama 3.3 70B +- Self-hosted compliance path: Custom endpoint / LiteLLM / vLLM + +## 4. Task Rules — which model runs when + +The rules engine evaluates task type, complexity score, and explicit overrides. If `always_use=true`, that model wins immediately. If multiple rules match, specificity and saved priority decide the winner. + +### Complexity scoring guide +- 0-2 → tiny edits, log cleanup, one-file tweaks +- 3-5 → medium bug fix, small feature, unit tests +- 6-7 → multi-file refactor, architectural reasoning +- 8-10 → security audits, migrations, large design work + +### Example rule sets shipped in v4.7 +- Cost-conscious (Free tier maximizer) +- Speed-focused (dev iteration) +- Quality-first (enterprise) +- Privacy-hybrid + +## 5. Cost Management + +### What the UI estimate means +The popular-model selector shows a rough estimate using **2K input / 500 output tokens**. It is not exact billing, but it is useful for sprint planning and model comparison. + +### Budgeting best practices +- Set hard spend caps in the provider dashboard. +- Keep a cheap model as the default and reserve premium models for complex tasks. +- Use task-specific routing so every small request does not hit Claude or GPT-4o. +- Use local models as fallback to reduce overage exposure. + +### Suggested budgets +| Profile | Monthly strategy | +|---|---| +| Solo hobby | Free-tier default + one premium audit model | +| Freelancer | Speed preset for iteration + Sonnet for client-critical reviews | +| Small team | Enterprise preset + provider spend caps + weekly dashboard review | + +## 6. Quota Exhaustion — what happens + +When the provider returns 402 or 429, v4.7 gives two options: +1. **ask_new_key** → pause and request a replacement key. +2. **local_model** → continue on the mapped local model. + +## 7. Security & Privacy + +- API keys are stored with **Fernet (AES-128 + HMAC-SHA256)** encryption at rest. +- The master key is not stored on disk. +- Secrets are masked in logs. +- Writes are atomic and file permissions stay restrictive. +- DevAI Studio does not add its own telemetry on top of your chosen provider path. + +## 8. Troubleshooting + +Below is a short version of the top issues. The full searchable knowledge base lives in `docs/TROUBLESHOOTING_KB.md`. + +### Issue #1: Invalid API key on Test Connection + +**Symptoms:** Red ❌ appears after clicking Test Connection. + +**Common causes** +- Extra spaces in the key +- Wrong provider selected +- Key revoked or expired + +**Recommended fix** +1. Regenerate the key from the provider dashboard +2. Paste it again without leading/trailing spaces +3. Confirm the provider dropdown matches the key type + +### Issue #2: Model replies in the wrong language + +**Symptoms:** Responses ignore the language of the prompt or system instruction. + +**Common causes** +- Model preset optimized for another language +- Task prompt lacks a language instruction +- Provider-side system prompt contamination + +**Recommended fix** +1. Add an explicit language instruction to the task +2. Prefer GPT-4o or Claude Sonnet for multilingual work +3. Test the model directly in the provider console + +### Issue #3: Fallback not triggering on 429 + +**Symptoms:** Pipeline pauses instead of moving to the chosen fallback behavior. + +**Common causes** +- Saved fallback policy is ask_new_key +- 429 body is returned as a generic error by a proxy +- Outdated custom model config + +**Recommended fix** +1. Re-open the model card and verify fallback policy +2. Retest using direct provider endpoint +3. Save the model again to refresh metadata + +### Issue #4: Pipeline hangs during Coder stage + +**Symptoms:** Status remains on coder or reviewer for too long. + +**Common causes** +- Provider latency spike +- Very large prompt or repository +- Provider queue saturation + +**Recommended fix** +1. Retry with a smaller ZIP +2. Switch to a faster preset like Groq or Claude Haiku +3. Use Private Mode for small/local tasks + +### Issue #5: Custom model is used even when disabled + +**Symptoms:** Router keeps picking a model after you thought it was inactive. + +**Common causes** +- always_use is enabled +- Task rules still match any task +- Priority order favors that model + +**Recommended fix** +1. Edit the model and disable always_use +2. Change task_types from any to specific categories +3. Adjust the rule priority matrix + +### Issue #6: Base URL looks correct but requests fail + +**Symptoms:** Connection test fails with network or 404 errors. + +**Common causes** +- Provider endpoint missing /v1 +- Anthropic direct uses /messages instead of /chat/completions +- Proxy strips auth headers + +**Recommended fix** +1. Verify the endpoint in provider docs +2. Use the built-in preset instead of manual typing +3. Try a direct provider endpoint to isolate the issue + +### Issue #7: Saved model disappears after restart + +**Symptoms:** A model is visible in-session but missing later. + +**Common causes** +- Master key changed +- Encrypted file corrupted +- Data directory permissions issue + +**Recommended fix** +1. Reuse the same master key +2. Restore data/custom_models.enc.json from backup +3. Ensure the process can write to the data folder + +### Issue #8: Quota bar never updates + +**Symptoms:** Model card always shows unknown quota. + +**Common causes** +- Provider does not expose headers +- First request never completed successfully +- Proxy removed rate-limit headers + +**Recommended fix** +1. Run one successful generation +2. Use a direct provider endpoint if quota visibility matters +3. Treat the bar as optional and rely on provider dashboard billing + +### Issue #9: Anthropic direct fails while OpenRouter works + +**Symptoms:** Claude via Anthropic shows HTTP errors. + +**Common causes** +- Anthropic key pasted into OpenRouter provider +- Account lacks billing +- Model name typo + +**Recommended fix** +1. Select Anthropic direct provider +2. Check billing and usage limits +3. Use a preset rather than a hand-typed model ID + +### Issue #10: OpenRouter free tier is rate-limited + +**Symptoms:** Requests work, then suddenly slow down or fail. + +**Common causes** +- Free tier cap reached +- Burst traffic from multiple runs +- No paid balance + +**Recommended fix** +1. Wait for the reset window +2. Use a paid key or alternate provider +3. Route simple tasks to a cheaper preset + +### Issue #11: Task Rules matrix saves unexpected priorities + +**Symptoms:** The top model in cards differs from the table order. + +**Common causes** +- Priority value duplicates +- Higher min complexity wins ties +- Model list sorted after save + +**Recommended fix** +1. Give each model a unique priority +2. Check min_complexity values +3. Refresh the panel after saving + +### Issue #12: Local fallback quality is poor + +**Symptoms:** Fallback works but output quality drops noticeably. + +**Common causes** +- Fallback switched to small local model +- Task too complex for local model +- Prompt too large for context + +**Recommended fix** +1. Use ask_new_key for critical work +2. Trim prompt size +3. Reserve local fallback for continuation, not best quality + +### Issue #13: Edit modal requires a new key unexpectedly + +**Symptoms:** Changing provider or model ID forces key replacement. + +**Common causes** +- Security design prevents silent target change +- Provider changed endpoints + +**Recommended fix** +1. Paste a new key when changing provider/model target +2. Keep provider/model unchanged to reuse the stored encrypted key + +### Issue #14: Model ID preset dropdown is empty + +**Symptoms:** Provider selected but no popular models appear. + +**Common causes** +- provider_presets.json missing +- Provider not curated +- JSON failed to load + +**Recommended fix** +1. Validate data/provider_presets.json +2. Choose a curated provider preset +3. Restart the app after restoring the file + +### Issue #15: Tour never starts + +**Symptoms:** First-run experience does not appear. + +**Common causes** +- localStorage already marks it complete +- Advanced options never opened +- CDN blocked + +**Recommended fix** +1. Use the Restart onboarding tour button in About +2. Open Advanced options once +3. Allow driver.js CDN or self-host the asset + +## 9. Advanced: Custom endpoints + +### Supported patterns +- **vLLM** — self-hosted OpenAI-compatible gateway. +- **LiteLLM proxy** — unified multi-provider gateway. +- **LM Studio HTTP server** — local desktop endpoint for experimentation. +- **Ollama via proxy bridge** — useful when a team wants one OpenAI-compatible URL. + +### Endpoint checklist +- Use the correct `/v1` base path. +- Verify auth header behavior. +- Manually test large context claims. +- Preserve upstream 401/403/429 status codes through proxies. + +## 10. FAQ + +### Do I need a separate API key for every provider? +Yes. Every direct provider has its own key. OpenRouter can expose many models through one key. + +### Can I recover stored keys if I forget the master key? +No. For safety, encrypted keys are not recoverable; re-add the models instead. + +### Do custom models replace local models? +No. They are a third tier. Local models remain available at all times. + +### Can I stay on free models only? +Yes. The cost-conscious preset is designed for that workflow. + +### Will local fallback produce the exact same answer? +Not always, but it preserves continuity when cloud quota is exhausted. + +### Are provider presets editable? +Yes. The JSON file is only a curated default; manual values still work. + +### Can the docs and KB be read offline? +Yes. The markdown files ship inside the release ZIP. + +### Are custom endpoints supported? +Yes, as long as the endpoint is OpenAI-compatible or you can supply a custom base URL. + +### Does billing happen inside the app? +No. Billing is handled by the provider; the app only shows an estimate. + +### Is this feature merge-safe for team repositories? +Yes. Storage is JSON-based and all existing v4.6 APIs stay unchanged. + +### Do I need a database migration? +No. The feature is additive and uses JSON-based storage. + +### Will screenshot-to-code flows use custom models? +Yes, when routing rules allow it and the provider is capable. + diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/docs/BYOM_USER_GUIDE_HI.md devai_phase4/docs/BYOM_USER_GUIDE_HI.md --- devai_phase4_orig/docs/BYOM_USER_GUIDE_HI.md 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/docs/BYOM_USER_GUIDE_HI.md 2026-07-06 10:28:32.013919928 +0000 @@ -0,0 +1,429 @@ +# 🎨 DevAI Studio v4.7 — अपना AI Model जोड़ें (Bring Your Own Model) + +## विषय-सूची +1. [यह feature क्या है](#1-यह-feature-क्या-है) +2. [Setup — पहला model 5 मिनट में](#2-setup--पहला-model-5-मिनट-में) +3. [Provider चुनना — कौन-सा best है](#3-provider-चुनना--कौन-सा-best-है) +4. [Task Rules — कब कौन-सा model](#4-task-rules--कब-कौन-सा-model) +5. [Cost Management](#5-cost-management) +6. [Quota Exhaustion — क्या होगा](#6-quota-exhaustion--क्या-होगा) +7. [Security & Privacy](#7-security--privacy) +8. [Troubleshooting](#8-troubleshooting) +9. [Advanced: Custom endpoints](#9-advanced-custom-endpoints) +10. [FAQ](#10-faq) + +## 1. यह Feature क्या है? + +DevAI Studio v4.7 में तीन model tiers हैं: local, built-in cloud, और custom BYOM. यह नया feature आपको अपने personal API keys के साथ production-grade models जोड़ने देता है। + +### Tier summary +- **Local models** — free, offline, private, लेकिन speed hardware पर depend करती है। +- **Built-in cloud models** — quick start, no setup, लेकिन shared limits हैं। +- **Custom models** — latest Claude, GPT, Gemini, Groq, Together, DeepInfra या custom endpoint; quota और billing आपकी। + +### Local models (हमेशा available) +- Qwen2.5-Coder-7B → coding tasks +- DeepSeek-R1-Distill-8B → complex reasoning +- InternLM3-8B → security audits +- Llama-3.2-3B → general chat +- MiniCPM-V-2.6 → vision (screenshot → code) +- Phi-3.5-mini → fast small tasks + +### Custom models feature से क्या मिलता है +- Latest flagship models बिना app update का इंतज़ार किए। +- आपकी personal quota — shared free-tier bottlenecks से बचाव। +- Task-aware routing: bug fix, feature, refactor, security, test generation, deployment, new app. +- Fallback control: pause and ask for a new key, या local model पर continue. + +## 2. Setup — पहला Model 5 मिनट में + +### Step 1 — Master Key set करें +Settings tab में strong passphrase डालें। यही आपकी API keys को encrypt करेगी। इसे password manager में save करें। भूल गए तो models फिर से जोड़ने पड़ेंगे। + +### Step 2 — Provider चुनें +- Fastest start: **OpenRouter** +- Best direct Claude latency: **Anthropic** +- Best general multimodal experience: **OpenAI GPT-4o** +- Fast iterative dev loops: **Groq** + +### Step 3 — API key बनाएँ +Provider dashboard में जाकर API key बनाइए और billing/quota verify कीजिए। OpenRouter free-tier भी दे सकता है, पर rate limits जल्दी hit हो सकती हैं। + +### Step 4 — Studio > Advanced options > ⋮ > Add model +नई v4.7 UX में provider preset चुनिए, फिर Popular models dropdown से common model pick कीजिए। इससे model_id, base URL और rough cost estimate अपने-आप भर जाते हैं। + +### Step 5 — Rules and fallback चुनें +- Task types: model किस category के कामों पर eligible होगा। +- Min complexity: 0 मतलब broad usage, 8-10 मतलब सिर्फ़ tough jobs। +- Always use: बाकी rules override करेगा। +- Fallback: नई key माँगे या local model पर switch करे। + +### Step 6 — Test Connection +Green ✅ आने के बाद Save Model दबाएँ। पहला run ideally छोटा security audit या bug-fix prompt होना चाहिए ताकि routing clearly verify हो जाए। + +### Recommended first prompt +```text +इस repo का security audit करो — SQL injection, XSS, hardcoded secrets, insecure dependencies और auth flaws check करो। +``` + +## 3. Provider चुनना — कौन-सा best है? + +| Provider | Why choose it | Popular models | +|---|---|---| +| OpenRouter (100+ models) | एक API key से 100+ models — Claude, GPT, Gemini, Llama सब | Claude 3.5 Sonnet, Claude 3.5 Haiku (fast+cheap), GPT-4o | +| Anthropic (direct) | Claude models directly — lowest latency, no OpenRouter markup | Claude 3.5 Sonnet (latest), Claude 3.5 Haiku | +| OpenAI (direct) | OpenAI-compatible endpoint | GPT-4o, GPT-4o Mini | +| Groq (fastest inference) | Ultra-fast LPU inference — best for iterative dev loops | Llama 3.3 70B, DeepSeek R1 Distill | +| Together AI | OpenAI-compatible endpoint | Qwen 2.5 Coder 32B | +| DeepInfra | OpenAI-compatible endpoint | Llama 3.1 405B | + +### Quick recommendations +- Budget-sensitive solo developer: OpenRouter + free/cheap presets +- Security review heavy team: Anthropic direct Sonnet +- UI + multimodal heavy team: GPT-4o +- Tight iteration loops: Groq Llama 3.3 70B +- Self-hosted compliance path: Custom endpoint / LiteLLM / vLLM + +## 4. Task Rules — कब कौन-सा model? + +Rules engine तीन चीज़ें देखता है: task type, complexity score, और explicit override. अगर `always_use=true` है तो वही model top priority ले लेता है। अगर multiple rules match करें तो higher specificity और saved priority matter करती है। + +### Complexity scoring practical guide +- 0-2 → tiny edits, log cleanup, one-file tweaks +- 3-5 → medium bug fix, small feature, unit tests +- 6-7 → multi-file refactor, architectural reasoning +- 8-10 → security audits, migrations, large design work + +### Example rule sets shipped in v4.7 +- Cost-conscious (Free tier maximizer) +- Speed-focused (dev iteration) +- Quality-first (enterprise) +- Privacy-hybrid + +### Recommended patterns +1. Cheap-first rule for `any` tasks with complexity 0. +2. Stronger model for `security` and `refactor` from complexity 7 or 8. +3. Local fallback for continuity, ask_new_key for quality-critical outputs. + +## 5. Cost Management + +### What the UI estimate means +Popular-model selection shows a rough estimate using **2K input / 500 output tokens**. यह exact billing नहीं है, पर sprint planning और model comparison के लिए काफी useful है। + +### Budgeting best practices +- हर month के लिए provider-side spend cap set करें। +- Free या cheap model को default रखें; premium rules सिर्फ़ complex work के लिए रखें। +- Task-type specific routing use करें ताकि GPT/Claude हर छोटी request पर न जाए। +- Local models को fallback में रखकर cloud overage कम करें। + +### Suggested budgets +| Profile | Monthly strategy | +|---|---| +| Solo hobby | Free-tier default + one premium audit model | +| Freelancer | Speed preset for iteration + Sonnet for client-critical reviews | +| Small team | Enterprise preset + provider spend caps + weekly dashboard review | + +## 6. Quota Exhaustion — क्या होगा? + +जब provider 402/429 दे, v4.7 का flow दो रास्ते देता है: +1. **ask_new_key** → pipeline pause, user से नई key माँगना। +2. **local_model** → task mapping के हिसाब से local model पर continue करना। + +### Recommended walkthrough +- A model card quota bar low दिखाती है। +- Run के दौरान provider limit hit होती है। +- Status में clear message आता है। +- User या तो key replace करता है या local fallback पर continue करता है। + +### What to document for teams +- कौन-से tasks पर cloud-only quality चाहिए। +- कौन-से tasks local fallback tolerate कर सकते हैं। +- किस provider का spend alert कौन monitor करेगा। + +## 7. Security & Privacy + +- API keys **Fernet (AES-128 + HMAC-SHA256)** से encrypted-at-rest रहती हैं। +- Master key disk पर store नहीं होती; session context में use होकर disappear हो जाती है। +- Secret values logs में mask रहते हैं। +- Atomic writes और restrictive permissions accidental leakage risk कम करते हैं। +- No vendor telemetry by DevAI Studio; billing और request flow आपके चुने provider तक सीमित रहता है। + +### Operational security checklist +- Personal and team keys अलग रखें। +- Revoked keys तुरंत replace करें। +- Shared demo environments में real production keys use न करें। +- Master key को password manager में रखें, plain text note में नहीं। + +## 8. Troubleshooting + +नीचे top issues का short version है। Full searchable knowledge base `docs/TROUBLESHOOTING_KB.md` में है। + +### समस्या #1: Invalid API key on Test Connection + +**Symptoms:** Red ❌ appears after clicking Test Connection. + +**Common causes** +- Extra spaces in the key +- Wrong provider selected +- Key revoked or expired + +**Recommended fix** +1. Regenerate the key from the provider dashboard +2. Paste it again without leading/trailing spaces +3. Confirm the provider dropdown matches the key type + +### समस्या #2: Model replies in the wrong language + +**Symptoms:** Responses ignore the language of the prompt or system instruction. + +**Common causes** +- Model preset optimized for another language +- Task prompt lacks a language instruction +- Provider-side system prompt contamination + +**Recommended fix** +1. Add an explicit language instruction to the task +2. Prefer GPT-4o or Claude Sonnet for multilingual work +3. Test the model directly in the provider console + +### समस्या #3: Fallback not triggering on 429 + +**Symptoms:** Pipeline pauses instead of moving to the chosen fallback behavior. + +**Common causes** +- Saved fallback policy is ask_new_key +- 429 body is returned as a generic error by a proxy +- Outdated custom model config + +**Recommended fix** +1. Re-open the model card and verify fallback policy +2. Retest using direct provider endpoint +3. Save the model again to refresh metadata + +### समस्या #4: Pipeline hangs during Coder stage + +**Symptoms:** Status remains on coder or reviewer for too long. + +**Common causes** +- Provider latency spike +- Very large prompt or repository +- Provider queue saturation + +**Recommended fix** +1. Retry with a smaller ZIP +2. Switch to a faster preset like Groq or Claude Haiku +3. Use Private Mode for small/local tasks + +### समस्या #5: Custom model is used even when disabled + +**Symptoms:** Router keeps picking a model after you thought it was inactive. + +**Common causes** +- always_use is enabled +- Task rules still match any task +- Priority order favors that model + +**Recommended fix** +1. Edit the model and disable always_use +2. Change task_types from any to specific categories +3. Adjust the rule priority matrix + +### समस्या #6: Base URL looks correct but requests fail + +**Symptoms:** Connection test fails with network or 404 errors. + +**Common causes** +- Provider endpoint missing /v1 +- Anthropic direct uses /messages instead of /chat/completions +- Proxy strips auth headers + +**Recommended fix** +1. Verify the endpoint in provider docs +2. Use the built-in preset instead of manual typing +3. Try a direct provider endpoint to isolate the issue + +### समस्या #7: Saved model disappears after restart + +**Symptoms:** A model is visible in-session but missing later. + +**Common causes** +- Master key changed +- Encrypted file corrupted +- Data directory permissions issue + +**Recommended fix** +1. Reuse the same master key +2. Restore data/custom_models.enc.json from backup +3. Ensure the process can write to the data folder + +### समस्या #8: Quota bar never updates + +**Symptoms:** Model card always shows unknown quota. + +**Common causes** +- Provider does not expose headers +- First request never completed successfully +- Proxy removed rate-limit headers + +**Recommended fix** +1. Run one successful generation +2. Use a direct provider endpoint if quota visibility matters +3. Treat the bar as optional and rely on provider dashboard billing + +### समस्या #9: Anthropic direct fails while OpenRouter works + +**Symptoms:** Claude via Anthropic shows HTTP errors. + +**Common causes** +- Anthropic key pasted into OpenRouter provider +- Account lacks billing +- Model name typo + +**Recommended fix** +1. Select Anthropic direct provider +2. Check billing and usage limits +3. Use a preset rather than a hand-typed model ID + +### समस्या #10: OpenRouter free tier is rate-limited + +**Symptoms:** Requests work, then suddenly slow down or fail. + +**Common causes** +- Free tier cap reached +- Burst traffic from multiple runs +- No paid balance + +**Recommended fix** +1. Wait for the reset window +2. Use a paid key or alternate provider +3. Route simple tasks to a cheaper preset + +### समस्या #11: Task Rules matrix saves unexpected priorities + +**Symptoms:** The top model in cards differs from the table order. + +**Common causes** +- Priority value duplicates +- Higher min complexity wins ties +- Model list sorted after save + +**Recommended fix** +1. Give each model a unique priority +2. Check min_complexity values +3. Refresh the panel after saving + +### समस्या #12: Local fallback quality is poor + +**Symptoms:** Fallback works but output quality drops noticeably. + +**Common causes** +- Fallback switched to small local model +- Task too complex for local model +- Prompt too large for context + +**Recommended fix** +1. Use ask_new_key for critical work +2. Trim prompt size +3. Reserve local fallback for continuation, not best quality + +### समस्या #13: Edit modal requires a new key unexpectedly + +**Symptoms:** Changing provider or model ID forces key replacement. + +**Common causes** +- Security design prevents silent target change +- Provider changed endpoints + +**Recommended fix** +1. Paste a new key when changing provider/model target +2. Keep provider/model unchanged to reuse the stored encrypted key + +### समस्या #14: Model ID preset dropdown is empty + +**Symptoms:** Provider selected but no popular models appear. + +**Common causes** +- provider_presets.json missing +- Provider not curated +- JSON failed to load + +**Recommended fix** +1. Validate data/provider_presets.json +2. Choose a curated provider preset +3. Restart the app after restoring the file + +### समस्या #15: Tour never starts + +**Symptoms:** First-run experience does not appear. + +**Common causes** +- localStorage already marks it complete +- Advanced options never opened +- CDN blocked + +**Recommended fix** +1. Use the Restart onboarding tour button in About +2. Open Advanced options once +3. Allow driver.js CDN or self-host the asset + +## 9. Advanced: Custom endpoints + +### Supported patterns +- **vLLM** — self-hosted OpenAI-compatible gateway for large models. +- **LiteLLM proxy** — multi-provider normalization layer with your own routing/billing policies. +- **LM Studio HTTP server** — local desktop endpoint for experimentation. +- **Ollama via proxy bridge** — useful when a team wants a unified OpenAI-compatible URL. + +### Endpoint checklist +- Base URL सही `/v1` prefix के साथ दें। +- Auth header behavior verify करें। +- Large context claims को manually test करें। +- 401/403 और 429 status codes proxy preserve कर रहा हो। + +### Example endpoint patterns +```text +https://api.openai.com/v1 +https://api.together.xyz/v1 +https://api.groq.com/openai/v1 +https://your-vllm-gateway.example.com/v1 +http://localhost:1234/v1 +``` + +## 10. FAQ + +### क्या मुझे हर provider के लिए अलग API key चाहिए? +हाँ। हर direct provider की अपनी key होती है। OpenRouter एक key से कई models दे सकता है। + +### क्या master key भूलने पर keys recover होंगी? +नहीं। सुरक्षा कारणों से encrypted keys दोबारा नहीं निकाली जा सकतीं; models फिर से जोड़ने पड़ेंगे। + +### क्या custom models local models को replace कर देंगे? +नहीं। यह तीसरी tier है; local models हमेशा available रहते हैं। + +### क्या मैं सिर्फ़ free models use कर सकता हूँ? +हाँ, cost-conscious preset इसी use case के लिए है। + +### क्या fallback local model हमेशा exact same answer देगा? +नहीं, पर pipeline continue रहेगी और context बचा रहेगा। + +### क्या provider presets editable हैं? +हाँ, JSON curated defaults है; आप manual values भी डाल सकते हैं। + +### क्या docs और KB offline पढ़े जा सकते हैं? +हाँ, release ZIP के docs folder में markdown files रहेंगी। + +### क्या custom endpoints supported हैं? +हाँ, अगर endpoint OpenAI-compatible है या आप custom base URL दे सकते हैं। + +### क्या billing app के अंदर होती है? +नहीं। billing provider dashboard पर होती है; app सिर्फ़ estimation दिखाता है। + +### क्या team repo में ये feature merge-safe है? +हाँ, क्योंकि storage JSON-based है और existing APIs unchanged हैं। + +### क्या कोई database migration चाहिए? +नहीं। Feature additive है और JSON-based storage use करता है। + +### क्या screenshots to code flow पर भी custom models apply होंगे? +हाँ, जहाँ routing allows और provider capable हो। + diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/docs/MIGRATION_v4.6_to_v4.7.md devai_phase4/docs/MIGRATION_v4.6_to_v4.7.md --- devai_phase4_orig/docs/MIGRATION_v4.6_to_v4.7.md 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/docs/MIGRATION_v4.6_to_v4.7.md 2026-07-06 10:28:32.013919928 +0000 @@ -0,0 +1,62 @@ +# DevAI Studio v4.6 → v4.7 Migration Guide + +## Table of contents +1. Scope +2. Compatibility guarantees +3. Pre-upgrade checklist +4. Deployment steps +5. Validation checklist +6. Rollback procedure +7. FAQ + +## 1. Scope +- This guide upgrades a working v4.6 deployment to v4.7 with BYOM support. +- The feature is additive and JSON-based. + +## 2. Compatibility guarantees +- All v4.6 APIs remain unchanged. +- Existing local-model and built-in cloud flows keep working. +- No database migration is required. +- No mandatory environment variables are introduced. +- Breaking changes: **zero**. + +## 3. Pre-upgrade checklist +- Keep a backup copy of the current v4.6 ZIP. +- Record the current commit/tag and release artifact name. +- Confirm `data/custom_models.enc.json` does not exist in v4.6 (expected, feature is new). +- If you have local edits, export a patch before replacing files. + +## 4. Deployment steps +1. Stop the running v4.6 app. +2. Extract the combined v4.7 package over a clean directory. +3. Preserve your existing environment file if you use one. +4. Start the app normally; no migration command is needed. +5. Open About tab and confirm the docs buttons render. +6. Open Advanced options and verify the three-dot BYOM entry point exists. + +## 5. Validation checklist +- App launches successfully. +- Existing v4.6 tasks run unchanged. +- BYOM provider presets load. +- Tour can be restarted from About. +- Troubleshooting KB opens from About. +- No data loss is observed in existing workspace exports or logs. + +## 6. Rollback procedure +1. Stop v4.7. +2. Restore the saved v4.6 ZIP into a clean directory. +3. Restore your previous environment file. +4. Delete any newly created BYOM files only if your deployment policy requires a pristine rollback: + - `data/provider_presets.json` + - `ui/onboarding_tour.py` + - `docs/BYOM_USER_GUIDE_HI.md` + - `docs/BYOM_USER_GUIDE_EN.md` + - `docs/MIGRATION_v4.6_to_v4.7.md` + - `docs/TROUBLESHOOTING_KB.md` +5. Start v4.6 and run the same smoke tests you used before the upgrade. + +## 7. FAQ +- **Do I need to convert old data?** No. +- **Do I need new env vars?** No required ones. +- **Is there any schema migration?** No. +- **Can I keep the docs folder on rollback?** Yes, if your packaging policy allows it. diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/docs/TROUBLESHOOTING_KB.md devai_phase4/docs/TROUBLESHOOTING_KB.md --- devai_phase4_orig/docs/TROUBLESHOOTING_KB.md 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/docs/TROUBLESHOOTING_KB.md 2026-07-06 10:28:32.013919928 +0000 @@ -0,0 +1,492 @@ +# DevAI Studio v4.7 Troubleshooting Knowledge Base + +## Table of contents +1. [Invalid API key on Test Connection](#issue-1-invalid-api-key-on-test-connection) +2. [Model replies in the wrong language](#issue-2-model-replies-in-the-wrong-language) +3. [Fallback not triggering on 429](#issue-3-fallback-not-triggering-on-429) +4. [Pipeline hangs during Coder stage](#issue-4-pipeline-hangs-during-coder-stage) +5. [Custom model is used even when disabled](#issue-5-custom-model-is-used-even-when-disabled) +6. [Base URL looks correct but requests fail](#issue-6-base-url-looks-correct-but-requests-fail) +7. [Saved model disappears after restart](#issue-7-saved-model-disappears-after-restart) +8. [Quota bar never updates](#issue-8-quota-bar-never-updates) +9. [Anthropic direct fails while OpenRouter works](#issue-9-anthropic-direct-fails-while-openrouter-works) +10. [OpenRouter free tier is rate-limited](#issue-10-openrouter-free-tier-is-rate-limited) +11. [Task Rules matrix saves unexpected priorities](#issue-11-task-rules-matrix-saves-unexpected-priorities) +12. [Local fallback quality is poor](#issue-12-local-fallback-quality-is-poor) +13. [Edit modal requires a new key unexpectedly](#issue-13-edit-modal-requires-a-new-key-unexpectedly) +14. [Model ID preset dropdown is empty](#issue-14-model-id-preset-dropdown-is-empty) +15. [Tour never starts](#issue-15-tour-never-starts) +16. [Tour restarts every session](#issue-16-tour-restarts-every-session) +17. [Security note mentions AES-128 but audit expects Fernet](#issue-17-security-note-mentions-aes-128-but-audit-expects-fernet) +18. [Provider preset cost seems too low](#issue-18-provider-preset-cost-seems-too-low) +19. [Migration from v4.6 loses nothing but docs are missing](#issue-19-migration-from-v4.6-loses-nothing-but-docs-are-missing) +20. [Troubleshooting search via Ctrl+F is hard](#issue-20-troubleshooting-search-via-ctrl+f-is-hard) +21. [Custom endpoint with LM Studio returns 404](#issue-21-custom-endpoint-with-lm-studio-returns-404) +22. [vLLM proxy rejects large context requests](#issue-22-vllm-proxy-rejects-large-context-requests) +23. [Self-hosted proxy masks 401 as 500](#issue-23-self-hosted-proxy-masks-401-as-500) +24. [API key copy/paste includes newline](#issue-24-api-key-copypaste-includes-newline) +25. [Budget control expectations are too high](#issue-25-budget-control-expectations-are-too-high) +26. [JSON file merge conflict on team repo](#issue-26-json-file-merge-conflict-on-team-repo) +27. [Rules mismatch because task is misclassified](#issue-27-rules-mismatch-because-task-is-misclassified) +28. [About tab links open a blank page](#issue-28-about-tab-links-open-a-blank-page) +29. [Demo video is present but silent](#issue-29-demo-video-is-present-but-silent) +30. [Gradio update changes DOM selectors](#issue-30-gradio-update-changes-dom-selectors) +31. [Quota exhaustion pauses without resume context](#issue-31-quota-exhaustion-pauses-without-resume-context) +32. [Enterprise reviewer asks for rollback plan](#issue-32-enterprise-reviewer-asks-for-rollback-plan) + +## How to use this document +- Use the table of contents for anchor links. +- Use Ctrl+F with the exact symptom phrase shown in bold. +- Cross-check the About tab docs links if a deployment strips local file routing. + + +### Issue #1: Invalid API key on Test Connection + +**Symptoms:** Red ❌ appears after clicking Test Connection. + +**Common causes** +- Extra spaces in the key +- Wrong provider selected +- Key revoked or expired + +**Recommended fix** +1. Regenerate the key from the provider dashboard +2. Paste it again without leading/trailing spaces +3. Confirm the provider dropdown matches the key type + + +### Issue #2: Model replies in the wrong language + +**Symptoms:** Responses ignore the language of the prompt or system instruction. + +**Common causes** +- Model preset optimized for another language +- Task prompt lacks a language instruction +- Provider-side system prompt contamination + +**Recommended fix** +1. Add an explicit language instruction to the task +2. Prefer GPT-4o or Claude Sonnet for multilingual work +3. Test the model directly in the provider console + + +### Issue #3: Fallback not triggering on 429 + +**Symptoms:** Pipeline pauses instead of moving to the chosen fallback behavior. + +**Common causes** +- Saved fallback policy is ask_new_key +- 429 body is returned as a generic error by a proxy +- Outdated custom model config + +**Recommended fix** +1. Re-open the model card and verify fallback policy +2. Retest using direct provider endpoint +3. Save the model again to refresh metadata + + +### Issue #4: Pipeline hangs during Coder stage + +**Symptoms:** Status remains on coder or reviewer for too long. + +**Common causes** +- Provider latency spike +- Very large prompt or repository +- Provider queue saturation + +**Recommended fix** +1. Retry with a smaller ZIP +2. Switch to a faster preset like Groq or Claude Haiku +3. Use Private Mode for small/local tasks + + +### Issue #5: Custom model is used even when disabled + +**Symptoms:** Router keeps picking a model after you thought it was inactive. + +**Common causes** +- always_use is enabled +- Task rules still match any task +- Priority order favors that model + +**Recommended fix** +1. Edit the model and disable always_use +2. Change task_types from any to specific categories +3. Adjust the rule priority matrix + + +### Issue #6: Base URL looks correct but requests fail + +**Symptoms:** Connection test fails with network or 404 errors. + +**Common causes** +- Provider endpoint missing /v1 +- Anthropic direct uses /messages instead of /chat/completions +- Proxy strips auth headers + +**Recommended fix** +1. Verify the endpoint in provider docs +2. Use the built-in preset instead of manual typing +3. Try a direct provider endpoint to isolate the issue + + +### Issue #7: Saved model disappears after restart + +**Symptoms:** A model is visible in-session but missing later. + +**Common causes** +- Master key changed +- Encrypted file corrupted +- Data directory permissions issue + +**Recommended fix** +1. Reuse the same master key +2. Restore data/custom_models.enc.json from backup +3. Ensure the process can write to the data folder + + +### Issue #8: Quota bar never updates + +**Symptoms:** Model card always shows unknown quota. + +**Common causes** +- Provider does not expose headers +- First request never completed successfully +- Proxy removed rate-limit headers + +**Recommended fix** +1. Run one successful generation +2. Use a direct provider endpoint if quota visibility matters +3. Treat the bar as optional and rely on provider dashboard billing + + +### Issue #9: Anthropic direct fails while OpenRouter works + +**Symptoms:** Claude via Anthropic shows HTTP errors. + +**Common causes** +- Anthropic key pasted into OpenRouter provider +- Account lacks billing +- Model name typo + +**Recommended fix** +1. Select Anthropic direct provider +2. Check billing and usage limits +3. Use a preset rather than a hand-typed model ID + + +### Issue #10: OpenRouter free tier is rate-limited + +**Symptoms:** Requests work, then suddenly slow down or fail. + +**Common causes** +- Free tier cap reached +- Burst traffic from multiple runs +- No paid balance + +**Recommended fix** +1. Wait for the reset window +2. Use a paid key or alternate provider +3. Route simple tasks to a cheaper preset + + +### Issue #11: Task Rules matrix saves unexpected priorities + +**Symptoms:** The top model in cards differs from the table order. + +**Common causes** +- Priority value duplicates +- Higher min complexity wins ties +- Model list sorted after save + +**Recommended fix** +1. Give each model a unique priority +2. Check min_complexity values +3. Refresh the panel after saving + + +### Issue #12: Local fallback quality is poor + +**Symptoms:** Fallback works but output quality drops noticeably. + +**Common causes** +- Fallback switched to small local model +- Task too complex for local model +- Prompt too large for context + +**Recommended fix** +1. Use ask_new_key for critical work +2. Trim prompt size +3. Reserve local fallback for continuation, not best quality + + +### Issue #13: Edit modal requires a new key unexpectedly + +**Symptoms:** Changing provider or model ID forces key replacement. + +**Common causes** +- Security design prevents silent target change +- Provider changed endpoints + +**Recommended fix** +1. Paste a new key when changing provider/model target +2. Keep provider/model unchanged to reuse the stored encrypted key + + +### Issue #14: Model ID preset dropdown is empty + +**Symptoms:** Provider selected but no popular models appear. + +**Common causes** +- provider_presets.json missing +- Provider not curated +- JSON failed to load + +**Recommended fix** +1. Validate data/provider_presets.json +2. Choose a curated provider preset +3. Restart the app after restoring the file + + +### Issue #15: Tour never starts + +**Symptoms:** First-run experience does not appear. + +**Common causes** +- localStorage already marks it complete +- Advanced options never opened +- CDN blocked + +**Recommended fix** +1. Use the Restart onboarding tour button in About +2. Open Advanced options once +3. Allow driver.js CDN or self-host the asset + + +### Issue #16: Tour restarts every session + +**Symptoms:** Users see the tour repeatedly. + +**Common causes** +- localStorage blocked in browser +- Incognito session +- Browser privacy extension + +**Recommended fix** +1. Use a standard browser profile +2. Disable aggressive storage cleaners for the app origin +3. Treat it as a browser setting rather than an app bug + + +### Issue #17: Security note mentions AES-128 but audit expects Fernet + +**Symptoms:** Documentation wording causes confusion. + +**Common causes** +- Fernet uses AES-128-CBC plus HMAC-SHA256 +- Reader expects a lower-level cipher description + +**Recommended fix** +1. Document the exact primitive and wrapper name together +2. Keep the wording consistent across guides and release notes + + +### Issue #18: Provider preset cost seems too low + +**Symptoms:** Estimated bug-fix cost differs from dashboard billing. + +**Common causes** +- Estimate assumes 2K input and 500 output tokens +- Provider bills cached tokens or additional overhead +- Round-off in UI + +**Recommended fix** +1. Treat it as a rough planning estimate +2. Check exact bill in provider console +3. Adjust your internal budgeting spreadsheet + + +### Issue #19: Migration from v4.6 loses nothing but docs are missing + +**Symptoms:** App works, but new guides are absent. + +**Common causes** +- Phase 4 files not deployed +- Incomplete ZIP extraction + +**Recommended fix** +1. Confirm docs/ folder exists in deployment package +2. Redeploy using the combined Phase 1-4 archive + + +### Issue #20: Troubleshooting search via Ctrl+F is hard + +**Symptoms:** Finding issues in KB takes too long. + +**Common causes** +- Heading text inconsistent +- No anchor-friendly titles + +**Recommended fix** +1. Use the TOC links first +2. Keep issue titles short and unique +3. Search the exact symptom phrase in quotes + + +### Issue #21: Custom endpoint with LM Studio returns 404 + +**Symptoms:** Local LM Studio server responds but chat route fails. + +**Common causes** +- Wrong port +- Different API path +- Server not started in OpenAI-compatible mode + +**Recommended fix** +1. Verify server URL in browser +2. Use the documented /v1 endpoint +3. Enable OpenAI compatibility in LM Studio settings + + +### Issue #22: vLLM proxy rejects large context requests + +**Symptoms:** Long prompts fail despite large advertised context. + +**Common causes** +- Server max model len lower than preset value +- Proxy truncation +- Model served with smaller rope scaling + +**Recommended fix** +1. Set a realistic context window in your infra +2. Use a smaller chunking strategy +3. Test maximum prompt size manually + + +### Issue #23: Self-hosted proxy masks 401 as 500 + +**Symptoms:** App shows generic error while creds are wrong. + +**Common causes** +- Reverse proxy rewrites status code +- Custom middleware + +**Recommended fix** +1. Inspect proxy logs +2. Prefer direct provider URL when debugging auth +3. Update middleware to preserve upstream status + + +### Issue #24: API key copy/paste includes newline + +**Symptoms:** Invisible characters break auth. + +**Common causes** +- Clipboard copied from terminal with newline +- Password manager inserted formatting + +**Recommended fix** +1. Paste into a plain-text editor first +2. Re-copy from provider dashboard + + +### Issue #25: Budget control expectations are too high + +**Symptoms:** User expects hard spend caps inside the app. + +**Common causes** +- Feature only estimates cost; billing lives with provider +- No provider budget API configured + +**Recommended fix** +1. Use provider-side spend limits +2. Treat app estimate as advisory + + +### Issue #26: JSON file merge conflict on team repo + +**Symptoms:** Two branches modify provider_presets.json differently. + +**Common causes** +- Manual edits to the same preset list +- Different ordering + +**Recommended fix** +1. Sort presets by provider name +2. Review both additions before merge + + +### Issue #27: Rules mismatch because task is misclassified + +**Symptoms:** Router picks a model for feature instead of refactor. + +**Common causes** +- Classifier confidence low +- Prompt mixes multiple intents + +**Recommended fix** +1. Rewrite the task to emphasize the primary intent +2. Use always_use temporarily for critical flows + + +### Issue #28: About tab links open a blank page + +**Symptoms:** Clicking docs buttons does not show files. + +**Common causes** +- Deployment blocks local file serving +- Reverse proxy strips /file route + +**Recommended fix** +1. Expose the docs directory in deployment config +2. Fall back to packaging docs in release ZIP + + +### Issue #29: Demo video is present but silent + +**Symptoms:** Stakeholder expects narration. + +**Common causes** +- Video is a storyboard walkthrough rather than a recorded voiceover + +**Recommended fix** +1. Use the screenshots plus release notes as the spoken script +2. Record narration later if needed + + +### Issue #30: Gradio update changes DOM selectors + +**Symptoms:** Tour highlights break after dependency upgrades. + +**Common causes** +- Button or accordion DOM changed +- driver.js selector stale + +**Recommended fix** +1. Revalidate selectors after any Gradio upgrade +2. Keep tour hooks tied to explicit elem_id values + + +### Issue #31: Quota exhaustion pauses without resume context + +**Symptoms:** User adds a new key and loses track of partial progress. + +**Common causes** +- Completed work context not surfaced in current UX + +**Recommended fix** +1. Persist task summary in status panel +2. Re-run with the same prompt after key update and compare output + + +### Issue #32: Enterprise reviewer asks for rollback plan + +**Symptoms:** Team needs a low-risk revert path. + +**Common causes** +- Change review process requires a documented rollback + +**Recommended fix** +1. Use the migration guide rollback steps +2. Keep the v4.6 ZIP backup alongside the v4.7 release package + diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/release_notes_v4.7.md devai_phase4/release_notes_v4.7.md --- devai_phase4_orig/release_notes_v4.7.md 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/release_notes_v4.7.md 2026-07-06 10:28:32.013919928 +0000 @@ -0,0 +1,19 @@ +# DevAI Studio v4.7 Release Notes + +## Highlights +- Bring Your Own Model (BYOM) is now production-ready. +- Curated provider presets remove the need to memorize base URLs and model IDs. +- First-run onboarding tour explains the three-dot workflow. +- Full Hindi and English user guides are bundled in the release. +- Migration remains zero-breaking and zero-schema-change. + +## Included artifacts +- Combined Phase 1 + 2 + 3 + 4 source tree +- `CHANGELOG_v4.7.md` +- `PHASE4_DIFF.patch` +- `screenshots/` with 12+ walkthrough images +- `demo/devai_studio_v4.7_phase4_demo.mp4` +- `docs/` with user guides, migration guide and troubleshooting KB + +## Upgrade risk +Low. All changes are additive. Existing v4.6 APIs remain unchanged. Binary files devai_phase4_orig/screenshots/09_provider_presets_overview.png and devai_phase4/screenshots/09_provider_presets_overview.png differ Binary files devai_phase4_orig/screenshots/10_cost_estimator.png and devai_phase4/screenshots/10_cost_estimator.png differ Binary files devai_phase4_orig/screenshots/11_example_rule_sets.png and devai_phase4/screenshots/11_example_rule_sets.png differ Binary files devai_phase4_orig/screenshots/12_onboarding_tour.png and devai_phase4/screenshots/12_onboarding_tour.png differ Binary files devai_phase4_orig/screenshots/13_docs_hub_about_tab.png and devai_phase4/screenshots/13_docs_hub_about_tab.png differ Binary files devai_phase4_orig/screenshots/14_quota_exhaustion_storyboard.png and devai_phase4/screenshots/14_quota_exhaustion_storyboard.png differ diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/tests/test_phase4_docs_presets.py devai_phase4/tests/test_phase4_docs_presets.py --- devai_phase4_orig/tests/test_phase4_docs_presets.py 1970-01-01 00:00:00.000000000 +0000 +++ devai_phase4/tests/test_phase4_docs_presets.py 2026-07-06 10:26:42.073919928 +0000 @@ -0,0 +1,26 @@ +from pathlib import Path +import json + +ROOT = Path(__file__).resolve().parent.parent + + +def test_provider_presets_catalog_exists(): + data = json.loads((ROOT / "data/provider_presets.json").read_text(encoding="utf-8")) + assert len(data["presets"]) >= 6 + assert len(data["example_rule_sets"]) >= 4 + + +def test_guides_exist_and_share_same_core_sections(): + hi = (ROOT / "docs/BYOM_USER_GUIDE_HI.md").read_text(encoding="utf-8") + en = (ROOT / "docs/BYOM_USER_GUIDE_EN.md").read_text(encoding="utf-8") + for marker in ["1.", "2.", "3.", "4.", "5.", "6.", "7.", "8.", "9.", "10."]: + assert marker in hi + assert marker in en + + +def test_onboarding_and_ui_hooks_present(): + tour = (ROOT / "ui/onboarding_tour.py").read_text(encoding="utf-8") + panel = (ROOT / "ui/custom_models_panel.py").read_text(encoding="utf-8") + assert "devai_tour_completed_v4.7" in tour + assert "add-model-btn" in panel + assert "fallback-radio" in panel diff -ruN '--exclude=__pycache__' '--exclude=*.pyc' devai_phase4_orig/ui/custom_models_panel.py devai_phase4/ui/custom_models_panel.py --- devai_phase4_orig/ui/custom_models_panel.py 2026-07-06 10:20:00.089919928 +0000 +++ devai_phase4/ui/custom_models_panel.py 2026-07-06 10:28:42.293919928 +0000 @@ -8,9 +8,10 @@ from __future__ import annotations import html +import json import tempfile from pathlib import Path -from typing import Any, List, Optional +from typing import Any, Dict, List, Optional import gradio as gr @@ -68,6 +69,93 @@ "number", "bool", "str", ] +PRESETS_PATH = Path(__file__).resolve().parent.parent / "data" / "provider_presets.json" + + +def _load_presets() -> Dict[str, Any]: + try: + return json.loads(PRESETS_PATH.read_text(encoding="utf-8")) + except Exception: + return {"presets": [], "example_rule_sets": []} + + +PRESET_DATA = _load_presets() +PROVIDER_PRESETS = {item.get("provider"): item for item in PRESET_DATA.get("presets", []) if item.get("provider")} +RULE_SET_PRESETS = {item.get("name"): item for item in PRESET_DATA.get("example_rule_sets", []) if item.get("name")} +RULE_SET_CHOICES = list(RULE_SET_PRESETS.keys()) + + +def _provider_meta(provider: str) -> Dict[str, Any]: + return PROVIDER_PRESETS.get(provider, {}) + + +def _provider_summary_html(provider: str) -> str: + meta = _provider_meta(provider) + desc = html.escape(str(meta.get("description") or "OpenAI-compatible endpoint")) + signup = str(meta.get("signup_url") or "") + docs = str(meta.get("docs_url") or "") + base_url = str((SUPPORTED_PROVIDERS.get(provider) or {}).get("base_url") or "Custom / user supplied") + links = [] + if signup: + links.append(f'Signup') + if docs: + links.append(f'Docs') + return ( + '
' + f'{html.escape(str(meta.get("display_name") or provider.title()))}
{desc}
' + f'Base URL: {html.escape(base_url)}' + + (f'
{" · ".join(links)}' if links else '') + + '
' + ) + + +def _popular_choices(provider: str) -> List[str]: + return [str(item.get("display_name")) for item in _provider_meta(provider).get("popular_models", []) if item.get("display_name")] + + +def _find_popular_model(provider: str, label: str) -> Optional[dict]: + for item in _provider_meta(provider).get("popular_models", []): + if str(item.get("display_name")) == str(label): + return item + return None + + +def _cost_hint_html(model: Optional[dict], provider: Optional[str] = None) -> str: + if not model: + return '
Select a popular model to auto-fill model_id, endpoint and a rough cost hint.
' + cost = dict(model.get("approx_cost_per_1k_tokens") or {}) + input_cost = float(cost.get("input", 0) or 0) + output_cost = float(cost.get("output", 0) or 0) + typical = (input_cost * 2.0) + (output_cost * 0.5) + ctx = model.get("context_window") + best_for = ", ".join(model.get("best_for") or ["any"]) + notes = model.get("notes") + parts = [f'≈ ${typical:.2f} per typical bug fix (2K in / 500 out tokens)', f'Best for: {html.escape(best_for)}'] + if ctx: + parts.append(f'Context: {int(ctx):,} tokens') + if notes: + parts.append(html.escape(str(notes))) + return '
' + '
'.join(parts) + '
' + + +def _rule_set_defaults(rule_name: str, current_model_id: str = "", current_label: str = ""): + preset = RULE_SET_PRESETS.get(rule_name) or {} + rules = list(preset.get("rules") or []) + selected = rules[0] if rules else {} + needle_id = str(current_model_id or "").lower() + needle_label = str(current_label or "").lower() + for rule in rules: + target = str(rule.get("model") or "").lower() + if needle_id and needle_id in target: + selected = rule + break + if needle_label and needle_label.split(' (')[0] in target: + selected = rule + break + fallback = "local_model" if rule_name == "Privacy-hybrid" else "ask_new_key" + return list(selected.get("task_types") or ["any"]), int(selected.get("min_complexity", 0) or 0), bool(selected.get("always_use")), fallback + + CUSTOM_MODELS_CSS = """