basemind 0.12.2

Full AI context layer over MCP — tree-sitter code-map, document RAG (PDF/Office/HTML/email + OCR + reranker), shared agent memory, on-demand web crawl, git history + blame + per-symbol diff. 300+ languages, 10+ coding-agent harnesses, content-addressed Fjall + LanceDB.

basemind

The context and communication layer for coding agents — a shared code-map, document RAG, memory, web crawl, git history, and agent-to-agent comms so multiple agents coordinate while they work. 300+ languages, one MCP server.

Install

pip install basemind

On first invocation, the pre-compiled Rust binary for your platform (macOS, Linux, Windows; x86_64 + arm64) is downloaded from GitHub Releases and cached under ~/.cache/basemind/<version>/.

Override the binary location with BASEMIND_BINARY=/path/to/basemind.

Quickstart

cd /path/to/your/repo
basemind scan        # index the working tree
basemind serve       # run the MCP stdio server

Wire basemind serve into Claude Code or any MCP client.

Full documentation

See the main README for complete docs, architecture, MCP tool reference, and per-harness setup instructions.

License

MIT.