Spaces:
Sleeping
Sleeping
Commit History
6 days ago
-
Add bring-your-own-key LLMs, ONNX code embeddings and security hardening
- Visitors bring their own Gemini / OpenAI / Claude key per request; keys are never stored or logged on the server - Re-add Salesforce/SFR-Embedding-Code-400M_R via its pinned ONNX export (no torch, no remote code); an adaptive time budget falls back to keyword search when embedding would be too slow for the hardware - Per-visitor sessions replace the shared global repo and chat history - Sanitize rendered LLM output (DOMPurify with SRI), CSP and security headers - Rate limiting, request size limits, safe error messages; API docs and debug output off; same-origin CORS - Harden the GitHub downloader (strict URLs, checked redirects, zip-slip and zip-bomb limits) - Per-session in-memory vector DB; non-root Docker image with the model downloaded at build time and the HF Hub offline at runtime - Trim requirements to current direct dependencies; remove runtime.txt - Index extensionless files such as Dockerfile, Makefile and LICENSE Co-Authored-By: Claude Opus 5.5 <noreply@anthro…5e9ddc1
-
fix embeddings retry logic
33a0d96 -
add rate limit handling
44d35f3 -
fix embeddings
29dfd99 -
fix api version to v1
0dfdc7f -
switch to google.genai
2ba793b -
fix embedding model name
cfc936f -
remove langchain_google_genai import
3c4275b -
use google-generativeai directly
994b864 -
fix embedding model name v2
3c67cdc -
fix embedding model name
8e2cfd9 -
switch to google embeddings api
9821a9f
-
fix cuda to cpu
36199e5
-
fix all core imports
57705ee -
fix all core imports
6fb6266 -
fix all imports
0adfd14 -
first commit
1c4a5b7