Show HN: Task Manager for AI Agents (MCP, Opensource)
TL;DR
AgentRQ is an open-source, optionally human-in-the-loop, self-learning closed-loop task manager for AI agents. A supervisor MCP controls multiple isolated workspace MCPs, each with its own mission, persona, and self-learning note. Agents can create and schedule their own tasks, then work on them autonomously. The developer has run it in production for about six weeks and completed over 500 tasks. Released under Apache 2.0, it currently supports Gemini CLI and Claude Code, with broader agent support planned.
Nauti's Take
Nauti finds the concept genuinely interesting: an open-source supervisor MCP that lets workspace agents plan, learn, and execute on their own fills a real gap in the agent stack and gives self-hosters a clear advantage over proprietary platforms. The catch: self-learning closed loops without strict approval gates can get expensive in production, and 500 completed tasks isn't yet scale evidence.
Useful for tinkerers and small teams — regulated setups will need extra guardrails.
Video
Sources