Workspace local
The default store lives at `.metabrain/store.leveldb`, so tools can discover it without setup ceremony.
Open source. Local first. CLI ready.
metaBrain
Local document memory for AI tools, coding agents, and the humans who work with them.
Give every agent one durable, searchable place for notes, source snippets, task context, metadata, tags, links, and version history.
Start small
Install `mb`, then give agents a shared memory.
The quickest path is the `mb` CLI: initialize a local store, write a note, search it, and keep moving. The optional server is there later when several tools need concurrent access.
Homebrew
Installs both `mb` and `mbd`.
brew tap OpenCow42/tap && brew install mbQuick CLI tour
Try the CLI in a clean workspace.
Create a store, write a note, browse the tree, search it, and read the body back. That is enough to feel the shape of the tool.
Quickstart
mb init
mb put /notes/today "Remember the lexical store." --tag planning --meta source=agent
mb list /notes --recursive --dates
mb tree --max-depth 2
mb search "lexical store" --tag planning
mb get /notes/todayMade for AI and agents alike
A memory layer agents can actually use.
Agents do not need a private file convention for every task. metaBrain gives them a stable local store with predictable paths, searchable content, tags, metadata, references, and retained versions.
01
Write durable facts
Store summaries, decisions, and task state where the next agent can find them.
mb put /tasks/release-checklist \
"Prepare first public release." \
--tag release --meta status=active
02
Search workspace memory
Retrieve context by words, paths, tags, or metadata instead of hunting through scratch files.
mb search "public release" --tag release
03
Patch without rewriting
Apply focused unified diffs to stored documents and keep the old versions available.
mb patch /tasks/release-checklist --patch-file change.diffWhy try it
Small CLI. Useful memory model.
metaBrain keeps the interface plain enough for shell scripts and rich enough for long-running agent workflows.
Searchable
Current document chunks are indexed for lexical search with tag, metadata, and path filters.
Structured
Documents carry metadata, tags, references, retained versions, and filesystem-like paths.
Versioned
Updates keep snapshots, and `patch` can apply unified diffs to stored document bodies.
Local server
Set up `mbd`, then regular `mb` commands get concurrent access for free: agents and humans share one endpoint instead of competing for the store lock.
Embeddable core
`MetaBrainCore` keeps shared behavior in a Swift library for future tools and interfaces.
Optional local server
Add `mbd` when many tools share memory.
You can use metaBrain directly through `mb` without running a server. When agents and humans need concurrent access, `mbd` owns the LevelDB lock, routes requests to the selected store, and keeps the same command flow available through one local endpoint.
Multi-store
One daemon socket or loopback port can serve separate `.metabrain/store.leveldb` paths.
Auto-probed
`mb` checks `http://127.0.0.1:6374` and falls back to direct LevelDB access when no daemon is running.
Local only
Use Unix sockets on macOS and Linux, or loopback HTTP for integrations and Windows-friendly service setup.
Optional server mode
Default loopback endpoint: `127.0.0.1:6374`.
mbd serve --store .metabrain/store.leveldb --host 127.0.0.1
mb version
mb search "release checklist" --tag release
mb --no-server search "direct LevelDB access"Open source
Built in the open for local-first agent work.
The project is BSD 3-Clause licensed, hosted on GitHub, and currently supports macOS, Linux, and Windows.