pdf-brain

Use it when

• Your research papers, books, and Obsidian notes are scattered across folders and you want one semantically searchable knowledge base without paying for embedding APIs
• Documents are sensitive so indexing, embedding, and enrichment must all run locally, while your coding agent still gets search access over MCP

What it solves

Not the fit when

• Only .pdf and .md inputs; scanned PDFs without a text layer need OCR first. Vector index storage is heavy (HNSW roughly 100KB per chunk row). Requires a running Ollama instance for embeddings; enrichment quality depends on the local model unless a cloud gateway key is configured.
• OCR of scanned image PDFs
• garbled PDF table extraction
• hosted multi-user knowledge base

Install

curl -fsSL https://raw.githubusercontent.com/joelhooks/pdf-brain/main/scripts/install.sh | bash (or npm install -g pdf-brain), then ollama pull mxbai-embed-large and ollama serve

Invoke

pdf-brain init, then pdf-brain add paper.pdf --enrich or pdf-brain ingest ~/papers --enrich, then pdf-brain search "query"; add {"command":"npx","args":["pdf-brain","mcp"]} to expose it to AI assistants

Alternatives

No reviewed alternatives recorded yet.