Show HN: Flamehaven FileSearch – Self-Hosted RAG Semantic Search (github.com)

Flamehaven FileSearch is an open‑source, self‑hosted RAG (retrieval‑augmented generation) tool that bundles a FastAPI server, CLI/SDK parity, and Google Gemini for generation. Out of the box you can pip install, set GEMINI_API_KEY, upload PDFs or other docs, and run semantic searches that return generated answers plus cited source snippets. The project is MIT‑licensed, uses local SQLite/file storage for auditable persistence, ships a Dockerfile and env‑based production defaults, and exposes an interactive Swagger UI. Key defaults: gemini-2.5-flash as the generator, 50 MB max file size, 1024 token max output, and up to 5 cited sources; example latencies reported are ~1s for store creation, ~5s to upload a 10 MB PDF, and ~2s per search (Gemini 2.5 Flash). This is significant for teams wanting a lightweight, extensible RAG stack they can run on-prem while retaining open governance over indexing and source tracing. The architecture supports drop‑in store adapters and response post‑processors, CI/test scaffolding, and clear environment overrides, making it practical for production or experimentation. Important tradeoffs: generation currently depends on the Gemini cloud API (so document content is sent to Google), even though storage is local — a privacy consideration for sensitive data. Roadmap items include caching/rate limiting, batch searches and WebSocket streaming, multi‑language support, and analytics, and the repo actively invites contributions.