Launched this week
Minions
Open source mission control for Hermes agent
698 followers
Open source mission control for Hermes agent
698 followers
Your Hermes Agent works great for one task. Try managing 20 in parallel? It's chaos. Cron jobs fail silently, tasks are blocked and you're spending more time fixing your agent than getting results. Minions gives you a single task board to view it all. Every running task gets periodic check-ins, retry if stuck, and escalate only when it's genuinely exhausted alternatives Works with Hermes Agent today, more runtimes coming.
Agent 37
Hey Product Hunt, I’m Vishnu, maker of Minions.
I built this after watching Hermes/OpenClaw power users hit the same wall: one agent task works great, but 10–50 long-running tasks become operations work.
The core idea is heartbeat supervision: every in-progress task gets periodic check-ins where the agent is asked to make progress, retry with a different approach if stuck, and only escalate to you when it has genuinely exhausted alternatives. Blocked tasks surface automatically. Completed work moves to your review queue.
Minions is an open-source mission control layer for agentic harnesses starting with Hermes Agent.
What it does today:
- Create and manage Hermes tasks from one board
- Track work across in progress, needs help, ready for review, and done
- Run heartbeat check-ins so stuck work surfaces automatically
- Stream agent work live while it runs
- Keep humans in the loop before anything is marked done
- See scheduled job history and output
- Run locally with SQLite, no account required
This is the local/open-source version of the agent management layer we’re building at Agent37. Hosted access opensMay 10, but Minions is usable locally today.
I’d especially love feedback from people running Hermes or OpenClaw agents. Where does your agent work break once you have more than one task running?
Product Hunt
Agent 37
@curiouskitty The moment you're delegating more than 2-3 tasks it's immediately useful to have a central view to see it. I've often hated having to just use whatsapp and there's no easy way to start separate threads. So for people facing this, they'd find it a no-brainer.
Minions does track the sessionIds of it's chat in it's own DB, so a seamless way to migrate would be to start new long running tasks you want to mange in Minions instead of telegram or whatsapp.
@curiouskitty @vishnukool
So, all the tasks I already have won't be read by Minions, right? I need to create them in Minions for everything to work.
Agent 37
@curiouskitty @fabio_borges2 That's right, it only tracks the tasks created from within Minions
Love the idea of a single board for all parallel agent tasks. How's observability handled – logs/traces per task, or just status?
Agent 37
@ella_bye No logs as of yet, since I don't find it useful for the purpose of supervising. But the supervisor heartbeat could intervene if a task is errored out for instance and check the logs autonomously to try and heal it. It's activity is separated out from the actual Task chat, into an acitvity window so you can see what the supervisor did to progress on the task or heal etc.
@vishnukool I was looking for something like this, 2 questions here -
1. Does it handle the long standing tasks? Is it sitting on top of the coding copilot/
2. Whats the concurrency for the task execution can it support?
Agent 37
@achuth_noob Glad to hear it,
1. We'll be adding /goal support soon. Right now it supports Hermes as an adapter but will add more adapter later.
2. Any number supported by your CPU/RAM. There's no explicit limit.
Feel free to create github issues for anything you're looking support for.
Open SaaS
Nice idea! Could it be used for OpenClaw, too?
Agent 37
@matijash Yes, soon. We're starting with Hermes agent and will add OpenClaw support soon.
The "single task board for parallel agents" framing keeps showing up — fifth product in two weeks targeting the same observation that engineering teams have moved from one agent in Cursor to many agents running at once. When my org grew past 60 engineers, the single biggest productivity tax wasn't writing code, it was the meta-work of knowing which task was queued where and what state it was in. With agents in the mix that overhead becomes the gating factor, not a side cost. Curious whether the open-source angle is mostly licensing/transparency or whether teams are actually running their own Hermes deployments — what does the modal user setup look like so far?
The "single task board for parallel agents" framing keeps showing up — fifth product in two weeks targeting the same observation that engineering teams have moved from one agent in Cursor to many agents running at once. When my org grew past 60 engineers, the single biggest productivity tax wasn't writing code, it was the meta-work of knowing which task was queued where and what state it was in. With agents in the mix that overhead becomes the gating factor, not a side cost. Curious whether the open-source angle is mostly licensing/transparency or whether teams are actually running their own Hermes deployments — what does the modal user setup look like so far?