Agent-3-7/minions
Mission Control for your Hermes agents
Minions is a centralized management interface for Hermes Agent that replaces fragmented terminal sessions with a unified dashboard. It utilizes a Kanban-style board to track persistent agent sessions, allowing users to monitor reasoning, tool calls, and task progress in real time. The system enforces a human-in-the-loop workflow where agents propose completions that require manual verification before being marked as done. Users can also manage scheduled recurring jobs and browse agent-created files through a local-first SQLite architecture.
- Kanban board for tracking autonomous task status and review queues
- Live streaming of agent reasoning and tool execution
- Human-in-the-loop verification for all agent-proposed task completions
full readme from github
Minions
Mission Control for Hermes Agent
Hermes Agent is powerful, but running real work on it means juggling terminal sessions, losing track of which job finished, and manually checking on long-running tasks. The more you delegate, the harder it gets to manage.
Minions gives you one screen to create, supervise, and review autonomous Hermes Agent work.
Hosted access option on Agent37.
Screenshots
Quick Start
Prerequisites: Node.js 18+ and Hermes Agent
npx minionsai
Open http://localhost:6969.
Local sqllite db is created on first run and state lives in ~/.minions/
Check the installed version:
minions --version
npm view minionsai version
The Settings page also shows the version of the running Minions server.
Features
- Kanban board: see every task at a glance: in progress, in review, done
- Autonomous execution: describe what you want in chat, walk away; the agent decides how to get it done
- Automatic review queue: successful agent runs move cards to ready for review
- Live streaming: watch tool calls, reasoning, and responses in real time
- Human-in-the-loop: agents propose completion; you verify and close. Nothing moves to done without your sign-off
- Per-task model control: override model and reasoning effort on any task
- Scheduled Tasks: create and manage recurring Hermes jobs, history, and output
- File browser: see files agents have created in the workspace directory
- Local-first option: self-host with SQLite, no account, and no cloud dependency. Your local data stays on your machine
How It Works
Each task is a persistent Hermes root session. You talk to it, it works, and the board reflects where everything stands. Chat transcripts live in Hermes's session database; Minions stores task metadata, status, and per-task settings in a local SQLite database.
Who It's For
- Hermes power users juggling multiple sessions across projects
- Indie founders delegating research, ops, writing, and coding to their agent
- Anyone running long-lived Hermes work who needs to know what finished, what's stuck, and what needs attention
Roadmap
- Scheduled task supervision: automatically monitor, recover, and report on scheduled agent jobs
- Notifications: get alerted via Telegram, WhatsApp, or webhook when a task needs review
- Skills library: pluggable skill templates for common workflows (lead gen, web research, content pipelines, data collection, competitive monitoring, outbound sequences)
- OpenClaw adapter: run Minions against OpenClaw-hosted agents
FAQ
Can I use this with other agents? Not yet. The adapter interface exists, but launch is Hermes-only. OpenClaw is next.
Contributing
Contributions are welcome. Please open an issue first with the feature or change you have in mind and why it should be added. Once the approach is approved, create a PR. See CLAUDE.md for architecture and development details.