Upload 3 files

Upload 3 files

Upload 3 files

Upload 2 files

Upload guardrails.py

Upload 5 files

Upload 2 files

Upload 4 files

Upload main.py

Upload 2 files

Upload llm_client.py

Upload 2 files

Upload 2 files

Upload 3 files

Upload 3 files

Upload llm_client.py

Upload main.py

Upload 3 files

Upload Dockerfile

Upload Dockerfile

Upload 2 files

Upload main.py

Upload main.py

Upload 4 files

Fix Ollama ReadTimeout: retries, longer timeout, context cap, fallback

RAG: qwen2.5:3b, distance-filter context, strict prompt, Ollama options

Tighten Ollama prompt: focus on question, no extra topics

Redeploy rag-api: RAG_TOP_K, retrieve_only, tenant-scoped search

RAG API: legacy schema migration for source_id (add_columns + rewrite fallback)

RAG API: source_id column, delete-before-reindex, delete-source/delete-project endpoints

Fix keep-alive: ping public URL every 5min to prevent HF sleep

Non-blocking bucket sync: upload returns immediately, sync runs in background thread

Add /debug-storage endpoint to diagnose bucket sync

Handle bucket I/O errors gracefully on startup

Add incremental bucket sync and keep-alive background task

Fix: handle LanceDB ListTablesResponse object properly

Fix: use list_tables() instead of deprecated table_names()

Fix: LanceDB on local disk, sync to/from bucket for persistence

Fix: robust table handling + /reset endpoint

Fix: handle stale LanceDB table with try/except + overwrite

Add multi-format upload (PDF, TXT, CSV, DOCX) + error logging

Fix: move HF_HOME to /tmp — bucket FS lacks symlink support for HF cache

Fix: run as root so mounted bucket at /data is writable

Fix: hardcode /data mount path, remove STORAGE_PATH env var

Switch to Ollama (qwen2.5:1.5b) on remote HF Space, remove llama-cpp

Perf: reduce n_ctx/max_tokens/context for 2 vCPU viability

Fix: correct GGUF repo to unsloth/gemma-3-1b-it-GGUF

Fix OOM: single-thread llama-cpp compilation

Fix: install build tools for llama-cpp-python compilation

Upload 3 files