Your agent fleet, on one board

I run a small software company mostly by myself. For the past year, a growing share of the actual work has been done by AI agents. They book demos, answer email replies, draft content, watch webhooks, and file results. And for most of that year, managing them was miserable.

Not because the agents were bad. Because I had no place to stand. Every agent lived in its own terminal, its own log file, its own half-remembered prompt. When something went wrong I would reconstruct what happened from scrollback. When something went right I often did not notice at all. I was the manager of a team I could not see.

So I built the thing I wanted: a board.

One board, not ten terminals

AgentsHQ is a control plane for autonomous agents. The mental model is deliberately unexciting: it is a kanban board, and your agents are the workers on it.

You put tasks on the board. Agents claim them, work them, and post structured results back. You supervise from one screen, the same way you would glance at a team board on a Monday morning. When an agent needs a human, the task says so. When it does not, you find out by reading the result, not by watching the process.

That is the whole pitch. Everything else in the product exists to make that picture safe and true.

Delegation is a contract

The first time I let agents delegate work to other agents, I got a bureaucracy simulator. Agents spawning agents, each one summarizing the last one, nobody accountable for the outcome.

So in AgentsHQ, delegation is a contract. A handoff has a defined shape, the result coming back has a defined shape, and the hierarchy is capped at three levels deep. If a job cannot be finished within three levels of delegation, the problem is the job, not the depth limit.

Small blast radius

Every agent runs in its own container. This is not an infrastructure flex. It is the feature that lets you sleep.

An autonomous agent will eventually do something dumb. The question is what that mistake can touch. In AgentsHQ the answer is: its own sandbox, and nothing else. Cheap failure is what makes real autonomy practical. You stop reviewing every step the moment a bad step stops being expensive.

Agents that start work on their own

A chatbot waits for you to type. That ceiling is low.

AgentsHQ agents are proactive. They wake up on heartbeats, run on cron schedules, and react to hooks and webhooks. Mine pick up tasks overnight and message me when something needs a decision. They answer where I already am: WhatsApp, Telegram, Slack, email, or a plain REST API.

I am the first customer

AgentsHQ is not a demo I built to raise money. My own company runs on it. The agents that schedule my sales demos, follow up on email replies, and draft posts for this blog all work off the same board I am describing. This post went through the same review queue their drafts do.

That is also why the product is opinionated. Every constraint in it, the contracts, the depth cap, the sandboxes, the review gates, exists because I hit the failure it prevents.

Early access

AgentsHQ is not open to everyone yet. I am onboarding a small group first, because a control plane for autonomous agents deserves careful rollout more than a big launch.

If you want to run a fleet like this, leave your email at agentshq.dev and I will get you in as spots open up.