AI coding agents are increasingly useful for repetitive repository work—but running them reliably is a systems problem, not simply a prompting problem. Moadim, a new open-source project shown on Hacker News, is positioning itself as a local scheduler for that operational layer.
The project lets developers define a “loop” that combines a prompt, a schedule and an agent. On each scheduled tick, Moadim can launch supported agents—including Claude, Codex, Hermes and Pi—against a repository. It also supports queuing a NanoClaw task. The practical pitch is straightforward: replace manual, ad hoc agent sessions with repeatable jobs that can inspect, update or otherwise work on codebases at a defined cadence.
What Moadim changes
Moadim runs as a daemon on macOS and Linux, scheduling work internally rather than relying on a host cron service or a cloud queue. Each job is launched in a fresh, isolated workbench inside a `tmux` session. A watchdog is intended to terminate hung runs, and completed sessions are reaped.
That isolation matters for teams experimenting with unattended agent workflows. Long-running or repeated agent tasks can otherwise leave behind process state, stalled sessions or unclear execution history. Fresh environments do not eliminate the risks of agent-written code, but they establish a more predictable unit of execution: one prompt, one scheduled run, one contained workbench.
The tool is self-hosted and MIT-licensed, according to its project site. A local install starts a server on port 5784; users can register it with `launchd` or `systemd` to keep it running through logins and reboots. The stated requirements include a Unix-like operating system and `tmux` on the machine’s path.
Built for programmatic workflows
Moadim is not only a local web interface. It exposes routines through documented REST endpoints, an OpenAPI schema and Swagger UI. It also makes loops available as MCP tools through an `/mcp` endpoint, and includes an iCal feed.
Those interfaces could make the scheduler relevant beyond a single developer’s laptop. An internal operations tool, for example, could query configured routines over HTTP, while an MCP-compatible agent could inspect or invoke them as tools. The architecture suggests a distinction that will matter for engineering leaders: an agent’s reasoning environment may change, while the scheduling, execution and observability layer can remain comparatively stable.
Why operators should pay attention
The emerging practice often described as “loop engineering” shifts attention from crafting one ideal prompt to designing a recurring process: what task runs, how often, with which agent, under what constraints, and how failure is handled. Moadim packages several of those basics into a developer-run daemon.
That could be useful for bounded jobs such as recurring repository audits, backlog triage, documentation maintenance or test-failure investigation. But scheduled autonomy also raises familiar governance questions. Teams will need to define repository permissions, secrets access, branch and review policies, allowed network behavior, spending limits for model usage, and an approval path before changes reach production.
Local execution is a meaningful choice here. It can keep source code and job control on infrastructure a team already manages, but it also makes the team responsible for securing the host, maintaining agent credentials and monitoring the service.
What to watch next
Moadim’s value will depend less on its ability to launch an agent than on the reliability of its execution controls in real development environments. Builders evaluating it should test failure behavior, concurrent routines, logging and auditability, as well as how cleanly its isolated workbenches interact with existing Git and CI practices.
The broader signal is clear: as coding agents move from chat windows into recurring workflows, scheduling and containment are becoming product categories of their own. Moadim offers a lightweight, self-hosted entry point for teams that want to experiment with that model without adopting a hosted orchestration platform.
