Local RAG without the cloud or the bloat

What it does RLAMA is a Go CLI that creates local Retrieval-Augmented Generation systems. It reads documents from disk or crawled websites, chunks them using fixed, semantic, hierarchical, or hybrid strategies, sends the text to Ollama for embeddings, and stores the resulting vectors as JSON in ~/.rlama. When you ask a question, it compares your query embedding against the stored documents using a custom cosine-similarity implementation, retrieves the relevant chunks, and asks Ollama to generate a context-aware answer. It also exposes an HTTP API and can watch folders or websites for new content.

The interesting bit Instead of relying on a dedicated vector database, the project handles retrieval with a custom cosine-similarity implementation over local JSON files, keeping the entire stack to a single binary plus Ollama. It also bundles operational conveniences—directory watchers, website monitors, a Hugging Face GGUF browser, and an interactive setup wizard—that are usually scattered across separate tools.

Key highlights

• Runs fully offline after Ollama is available; all data stays in local JSON under ~/.rlama
• Supports multiple chunking strategies: fixed, semantic, hierarchical, and hybrid
• Includes an API server and can watch directories or monitor websites for automatic re-ingestion
• Offers integration with Hugging Face Hub for browsing and running GGUF models
• Distributed as a cross-platform single binary

Caveats

• The project is explicitly on pause due to the maintainer’s work and university commitments, so active development and support are currently stalled.

Verdict Worth a look if you want a quick, private document-Q&A setup without cloud APIs or container orchestration, provided you are comfortable using a project that is currently on hiatus. Those needing active feature development or enterprise support should probably wait for the pause to lift or look elsewhere.