Deployable-Knowledge

Version 0.9.0

Offline‑first retrieval‑augmented generation (RAG) stack for disconnected or bandwidth‑constrained environments.

Overview

Deployable‑Knowledge bundles a local vector store, prompt management and a lightweight web UI around a pluggable large‑language model. Documents are embedded locally and queried through FastAPI endpoints which power the JavaScript front end.

Features

• Document ingestion for PDF and plaintext sources
• ChromaDB vector store with sentence‑transformer embeddings
• Chat and search endpoints with optional streaming responses
• Configurable prompts and persona editing
• Authentication middleware with session and CSRF protection

Quick start

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
make run

Visit http://localhost:8000 once the server starts. ollama must be running locally and can be configured via environment variables such as OLLAMA_MODEL.

Architecture overview

The system is split into three layers:

core/  – retrieval, prompt rendering and LLM adapters
api/   – FastAPI routers translating HTTP ↔ core
app/   – static assets and UI routes

See docs/ARCHITECTURE.md for detailed diagrams and data‑flow breakdowns.

Documentation

Additional guides live in the docs/ folder:

• API reference
• UI overview
• Backend services
• Configuration guide
• Prompt & LLM integration

Contributing

1. Create a feature branch off main.
2. Add tests and run pytest before submitting a pull request.
3. Follow the existing coding style and keep docstrings concise.
4. Open a PR describing the change and link to any relevant issues.

---

Released under the MIT license.