@teambrain/cli
v0.3.0
Published
tb — git-native, human-governed shared memory for AI coding agents (CLI, daemon, MCP server)
Downloads
602
Maintainers
Readme
@teambrain/cli
tb — git-native, cross-vendor shared memory for AI coding agents.
TeamBrain gives Claude Code, Cursor, and other MCP-capable agents one shared, human-governed memory: markdown files in your repo, served over MCP by a local daemon, with new memories proposed as pull requests — never written silently.
npm install -g @teambrain/cliQuick start
tb init # import CLAUDE.md / .cursorrules / ADRs → a PR branch
tb install claude-code # register the MCP server + capture wiring (also: cursor)
tb serve # run the local daemon (index + watcher + MCP backend)
tb doctor # verify daemon, index, and capture healthEveryday commands
tb audit --last-session # see exactly what was recorded, post-redaction
tb propose --class learning --title "..." --body "..." # queue a candidate
tb retire <id> "reason" # open a PR retiring a memory
tb reindex # rebuild the local index (recovery path)
tb lint .teambrain # validate memories (schema, size, injection heuristics)
tb distill --dry-run # (CI) cluster sessions → memory-proposals PR
tb digest --dry-run # (CI) people-free weekly digestRun tb --help or tb <command> --help for grouped help, exit codes, and
examples.
How it works
- Git is the source of truth. Memories are markdown + YAML front-matter in
.teambrain/; approval is a pull request, history isgit log. - A local daemon serves them over MCP (
memory_context,memory_search,memory_propose,memory_feedback) with hybrid lexical + vector retrieval. - Capture is metadata-only and redacted on-device — never raw prompts,
file contents, or diffs.
tb auditshows you exactly what was stored. - Nothing writes to the brain without a human merge. Automation proposes; people approve.
Part of TeamBrain
This is the command-line entrypoint. The implementation lives in
@teambrain/core,
@teambrain/index,
@teambrain/mcp,
@teambrain/hooks,
@teambrain/redact, and
@teambrain/distill.
Full docs, format spec, and threat model: github.com/donatienmigue/TeamBrain
Apache-2.0
