Skip to content

The context and communication layer for coding agents

Turn any repo into an always-current map of its code, documents, history, and memory — so agents answer from structure and search instead of burning their context window on grep and file reads.

A code map across 300+ languages, and PDF, Office, HTML, email and image documents — one server does both jobs.

It also gives you semantic + full-text search, git history & blame, shared memory, and agent-to-agent comms. Every answer comes back as paths, line numbers, and signatures, not whole files, so a question about your code costs a small fraction of the tokens it takes to read the source.

The operating rule: basemind first, shell/grep/git fallback

Section titled “The operating rule: basemind first, shell/grep/git fallback”

Answer from structure

code_outline a file before you open it, code_symbols instead of grep, code_refs instead of grepping call sites. Pointers, not file bodies.

Ask history directly

git_log, git_blame and git_diff — git intelligence at symbol resolution, in tens of microseconds.

Search everything

Semantic + full-text search over PDFs, Office, HTML, email, and images (OCR), with reranking and per-document summaries.

Coordinate agents

A shared channel for a team of agents on the same repo: rooms, inboxes, direct messages, and orchestration of many named subagents.

The MCP plugin sets up the server, helper skills, comms, and slash commands — and downloads the basemind program for you on first use. Pi uses the CLI and needs the program installed separately. Or install the program directly with a package manager.

/plugin marketplace add Goldziher/basemind
/plugin install basemind@basemind

More hosts — Windsurf, Factory Droid, OpenCode, Kimi, Hermes, Cline, Continue, and others — plus the plain MCP server and standalone CLI paths are in the full installation guide.