open-source tools
for agentic composition.
Local-first building blocks for agent-driven workflows. Memory across sessions, signed code review, whole-codebase audits, tickets agents can claim, and browser views for any CLI — pieces small enough to compose, opinionated enough to ship.
// the suite
think
@openthink/thinkLocal-first memory for AI agents — resident daemon, vector recall, an L1 outbox, git-shareable cortex. Bundles think serve.
stamp
@openthink/stampAuthor-agent opens a diff. Reviewer-agents you control return signed verdicts. The gate stays closed until they approve, and the attestation verifies downstream — at a self-hosted gate, on a GitHub PR check, or local-only.
ui-leaf
@openthink/ui-leafCustomizable browser views, on demand, for any CLI in any language. Mount a template once with a single command; ship the data — get a rich rendered view.
audit
@openthink/audit · oauditReads a repo, a folder, or an agent skill and tells you what it actually does, before you install or run it.
pm
@openthink/pm · pmLocal-first tickets on an append-only op log. --json everywhere, stable exit codes, optional sync.
// the shape of it
Each tool stands alone. Each tool composes with the others. stamp reviews the diff before merge; think remembers what was decided so the next session doesn't repeat the work; audit sweeps the whole tree on demand; pm holds the tickets agents claim and close; ui-leaf renders any of it in a browser when a terminal isn't enough.
All local-first. All MIT. No mandatory cloud. No vendor lock-in. The pieces are small on purpose — pick the ones you need, leave the rest.