Open Multi Agent

Open Multi Agent is an open source, self hosted TypeScript runtime for coordinating teams of AI agents. A developer hands it a goal or an explicit task graph, and a coordinator decomposes the work, runs agents in parallel and returns a traced run that can be resumed. The project emphasizes control and auditability. Consequential actions such as a plan, a ready task or a tool call can be suspended until a reviewer decides, with the request persisted alongside a checkpoint. When the run journal is on, a finished run can be checked against its own recorded events offline. Production controls include token and cost budgets, per agent turn and time limits, retries with backoff, loop detection, context strategies and credential redaction in traces. It works with cloud model providers, local models or fully offline setups. The package installs from npm as the core package and is released under the MIT license, with development in public on GitHub. The site lists a price of zero and publishes no paid plans. Documentation includes guides such as a production checklist, and the site is available in English and Simplified Chinese.