Set Qwen 2.5 0.5B (GGUF IQ2_M) as default model

Add Qwen 2.5 1.5B (GGUF Q2_K) model (~676 MB) to catalog

Remove legacy static frontend prototype and set root API status endpoint

fix(llm): use Qwen 2.5 3B Q2_K/IQ2_M available in bartowski repo and sanitize fallback repo names

feat(llm): add Qwen 2.5 3B 1.6-bit IQ1_S and expose dynamic model catalog in GET /api/models

refactor(llm): modularize LLM with BaseLLMEngine ABC, engines provider, use_rag bypass, and remove legacy parser

fix(rag): clean markdown syntax, reframe prompt, and add repetition penalty to prevent SLM regurgitation

perf: set default to Qwen 0.5B Q2_K 2-bit quantization, increase n_ctx to 3072 and max tokens to 1024

fix(llm): map Ollama model names to HF models and isolate Pinecone session namespaces

Add session manager component with targeted session termination and lifecycle APIs

perf: add real-time SSE token streaming, stop tokens, and embedding singleton

perf: add OpenMP thread pinning, passive wait policy, and core utilization tuning

feat: enable 2-bit Q2_K/IQ2 GGUF quantization and CPU flash_attn batching

perf: add startup warmup, vectorstore caching, 512-token detailed prompt

feat: add dynamic set-engine endpoint and llama.cpp vs transformers toggle

build: install pre-compiled CPU wheel for llama-cpp-python

added cpp inference

added docs api correction

added diff models

Changed Embedding model

rename to pdf in controller and changed embedding

modularized chuncker and pipeline to service folder

Modularized code and detached FAISS

Added optional include

Always query Pinecone regardless of session_id; filter only when session scoped

Switch LLM to Qwen2.5-0.5B local; add retrieval debug logging

Running Qwen0.5B in HF space

Add Pinecone PDF ingestion pipeline

Deploy update

add browser-side PDF text extraction and parsing support

update frontend UI theme and threshold-based prompt retrieval fallback

feat: support client-side context to bypass FAISS retrieval

feat: design clean compact single-line chat input box

feat: deploy premium dark mode default and glassmorphic text bar updates

feat: add /api/ping endpoint for frontend cold-start detection and queueing handshake

fix max_tokens=0 Bad Request error in ChatHuggingFace

Improve heartbeat, error handling, and general answering logic

Optimize for mobile and improve WCAG accessibility

Downgrade numpy to <2 to fix FAISS compatibility

Update RAG system with new chatbot UI and unified server

Switch to Mistral-7B for serverless inference reliability

Fix HuggingFaceEndpoint validation error - remove repo_id when endpoint_url is used

Re-enable FAISS allow_dangerous_deserialization for new LangChain version

Upgrade LangChain to v0.2 to support new HuggingFace Router URL

Fix HuggingFaceHub router URL using environment variable

Update HuggingFace API Endpoint to router.huggingface.co

Fix HuggingFaceEndpoint validation error

Fix FAISS deserialization error

Deploying RAG backend

initial commit