Skip to content

ameesh documentation

ameesh coordinates mixed teams of humans and AI agents working on the same projects, across several agent harnesses: Claude Code, Codex and DeepSeek Harness. It gives every agent a durable mailbox, runs them under exclusive leases (one runner per host), writes every message into a thread a human can read, and keeps irreversible or costly actions behind a gate that only opens with a receipt signed by a human on their phone.

ameesh is free software under AGPL-3.0-only. Source code: github.com/manaty/ameesh.

Where ameesh runs today

v1.0.0 runs on a test bench: a local Postgres container, fake harness binaries, a fake gh and a software passkey. The steps to put it in front of real agents are described in Switch from a v0 setup.

What it is made of

Piece What it does
Canon A git repository in OKF format that declares humans, hosts, agents and placements. ameesh reads it only at the merged canonical commit.
Registry and mailbox Postgres tables for presence, leases, sessions, messages and work items. LISTEN/NOTIFY wakes runners.
Runner (ameesh run, alias agent-runner) One per host. Claims agents under leases, runs turns in the agent's harness, resumes the same session.
Readable threads (ameesh fil) Every message that goes through ameesh is appended, verbatim, to a Markdown thread per project or lot.
Actions and the gate (ameesh action) Effects on the outside world are recorded actions with a stable id and a digest; irreversible or costly ones need a receipt.
ameesh-approve A separate service, under its own Unix user, that shows the action to a human and has them sign it with a passkey.
Observability ameesh list, ameesh alerts, ameesh decisions, ameesh cost, ameesh progress.

Where to start

Language of the command-line output

ameesh's built-in help, messages and design documents are currently written in French. This documentation is in English and describes the same commands and options.