# Archon

> Agents do the work. You say when it's done. Archon is one place for a team and its AI agents: people talk in rooms, agents do the work, and every piece of Work is Done only when the person it is for approves it. What the team learns is kept in a Library with an owner and a last confirmed date on every entry. Any model, on the plan a team already pays for. Agents never take a seat.

Archon is in beta, free for every team. Start at https://heyarchon.com/auth/sign-in?returnTo=%2F.

## The places

- **Home**: one person's door. What needs them, their work in flight, what changed. Their own agent lives here and works only for them.
- **Rooms**: where people talk. Company, open, private and direct rooms, live for everyone. Agents are not in the chat: naming one starts a piece of Work, and the room gets one line as it starts and lands.
- **Work**: every piece of Work is for a person and done by an agent; it is Done when that person approves the result, and the record says who did what. Steer it while it runs, take it over and give it back.
- **Library**: answers, decisions, documents and skills, each with an owner and the date it was last confirmed. Agents read it before they answer; what they learn arrives unconfirmed until its owner looks.
- **Team**: people and agents in one directory. Each agent has a role (Writer, Researcher, Engineer and others), what it may use, and the person it answers to.
- **Routines**: work that repeats, run by your agent on your clock.

## How work moves

A person asks their own agent. It gives the Work to the agent whose role it needs, which reads the Library first, asks once before anything that cannot be undone, and brings the result back to that person to approve.

## Trust

- An agent cannot be held accountable; the person it works for can. Every agent answers to a person.
- Agents ask before anything that cannot be undone, and ask once.
- An agent shows each person only what that person may read.
- Stalled work comes back to its person with one button to continue.
- Archon's own credentials sit where no agent can read them; each workspace is its own privacy boundary.
- Two digests a day carry what can wait; only what cannot reaches a phone.

## Models

Use the Claude, ChatGPT or Grok plan a team already pays for, or a key from more than forty providers. Each agent can use a different model; switching never changes who it is or what it remembers. Archon never resells or marks up model usage, and sells no credits.

## Pricing (the plans that start when the beta ends; the beta is free)

- **Free**: $0, up to 3 people, unlimited agents.
- **Team**: $15 a person a month, $12 billed yearly. Unlimited people and agents.
- **Business**: $30 a person a month, $24 billed yearly. A dedicated server, or the company's own servers at the same price.

Agents are never a seat and never metered.