Moadim

Moadim is an open-source, MIT-licensed loop engine for AI agents, built in Rust and distributed as a self-hosted local daemon. It lets developers define "routines" — a prompt, a schedule, and an agent — and fires them automatically against a repository on every cron tick, without manual prompting. The project is maintained by the moadim-io organization on GitHub and is free to use, modify, and self-host with no account required.

What It Is

Moadim introduces the concept of "loop engineering": instead of manually prompting AI coding agents, you design a loop that prompts them on a schedule. Each routine pairs a cron expression, a prompt, and a named agent (Claude, Codex, Hermes, NanoClaw, or Pi). The daemon fires the agent inside a fresh, isolated workbench on every tick, kills hung runs via a watchdog, and reaps the session when it finishes. The result is a persistent, automated agent workflow that runs while you sleep — no cloud queue, no sign-up, no hidden infrastructure.

Architecture and Interfaces

The daemon runs on http://localhost:5784 and exposes three interfaces simultaneously over a single port:

• Web UI — a browser dashboard for managing routines, viewing logs, and controlling the daemon
• REST API — a fully documented HTTP API with Swagger UI at /docs and an auto-generated OpenAPI spec
• MCP endpoint — a Model Context Protocol interface at /mcp, compatible with Claude Code and any MCP client

Routines are stored as TOML files under ~/.config/moadim/routines/, making them git-trackable and diff-friendly. The in-process scheduler (backed by tokio-cron-scheduler) handles all firing without relying on the host's cron daemon.

Supported Agents and Isolation

Built-in agents include claude, codex, hermes, nanoclaw, and pi. Each agent is configured via a TOML file at ~/.config/moadim/agents/<name>.toml. The daemon seeds default configs on first run but never overwrites user edits. Every routine run executes in a throwaway workbench directory under ~/.moadim/workbenches/, reaped on a periodic five-minute sweep. A watchdog enforces per-routine max_runtime_secs limits, and an overlap policy (configurable per routine) prevents concurrent fires of the same routine from interfering with each other.

Deployment and Platform Support

Moadim runs on macOS and Linux. It requires tmux on the PATH (used to launch each agent session) and optionally python3 for the built-in Claude agent's unattended trust pre-seeding. Installation options include:

• cargo install --locked moadim (from source via Cargo)
• cargo binstall moadim (prebuilt binary, no Rust toolchain compile)
• npm install -g moadim (thin npm wrapper over per-platform prebuilt binaries)
• Homebrew tap from the repository

Running moadim install registers a launchd (macOS) or systemd (Linux) user service so the daemon survives reboots. The daemon binds to loopback by default and enforces security guards against non-loopback exposure unless an API token is set.

Update: v3.2.8

The latest release is v3.2.8, published on 2026-09-03, with the repository last updated on 2026-09-10. The project was created in June 2026 and has accumulated 51 GitHub stars and 35 open issues, indicating active early development. The changelog follows the Keep a Changelog format and is tracked in CHANGELOG.md in the repository.