Not another tool. An office.

ARPOS gives every agent a place in your company, and gives you a clear view of the work.

A desk

Repos, tools and notes are there when the agent returns. Work starts where it left off.

An access card

Give each agent only the accounts it needs. Nobody gets the whole keyring.

Coworkers

Agents delegate, share work and keep the thread for the next handover.

A manager

One dashboard for the crew, one inbox for your decisions, one view of the spend.

The ARPOS live view: every agent as a node, sized by how many sessions it ran, joined by lines whose weight shows how much they have been talking to each other. — See the whole crew: who is working, waiting and talking to whom.

More agents should not mean more babysitting.

Brief a lead. Let the team divide the work. ARPOS brings you back only for the decisions that deserve a human.

Every prompt, every check and every approval is yours. Four agents, four conversations you have to hold in your head.

You brief one. It briefs the rest, collects the work and comes back when something actually needs deciding.

The shape is yours

Shape a pair, a team or a hierarchy. Their conversations stay readable and ready to continue.

And what still needs you finds you

Questions, blockers and finished work land in one queue. You see what needs you without hunting through sessions.

The engine is not the platform.

The market compares harness counts. None of them, out of the box, gives agents leak-free access beyond MCP: APIs, command lines, SSH servers and logged-in browsers.

ARPOS is built for leak-free access to the systems where real work happens. Give an agent read-only production data, or exactly the permission its task requires. It uses the connection without ever receiving the API key.

Claude Code and OpenCode are the engines. ARPOS is the layer that makes their work visible, controllable and safe across ten active sessions.

OpenCode
Claude Code

An ARPOS agent session showing the current task, tool activity and delegated work in one view. — See the work, the tools and the handoffs without leaving the session.

Wire an agent into anything. Never hand it a key.

Give an agent access to the tools, APIs, servers and accounts it needs. It can do the work without ever holding your secrets. Grant access once, revoke it when the job changes.

agent

→ cli("gh pr create --title 'Fix retry window'")

← https://github.com/arpos/core/pull/312

agent access ends here

arpos

GH_TOKEN=••••••••••••••••

handled by ARPOS · never shown to the agent

The agent uses the account. It never sees the secret.

The Connections screen: command-line tokens, API keys, web logins and paired browser sessions, each row showing which agents may use it. — Every credential the company owns, on one screen. Hand them out, take them back.

You don't configure an agent. You describe one.

Tell ARPOS what the new teammate should do. The Creator agent will set up the agent, equip it for the job, create dozens of test cases and run them as evaluations, with scoring.

Beyond the chat window.

Real work happens in browsers and repeatable tools, not in a stream of chat bubbles.

A browser that is part of the team

Agents can work in the accounts you authorize. Watch live, take over when it matters, then let them carry on.

Ask for a dashboard, get an app

Turn recurring work into a live screen you can return to, not another answer to dig out of a chat.

A live browser view in ARPOS, ready for the operator to take control. — A browser you can watch and take over.

An internal app built by an agent, shown alongside its working conversation. — An internal app built beside the conversation that asked for it.

Call an agent like a function.

Drop an agent into the systems you already run. Give it a shape for the answer, and the next step gets an object it can use, not prose it has to interpret.

n8n
Make
GitHub

POST /v1/sessions
{
  "agent": "Lead Qualification Specialist",
  "prompt": "Qualify yesterday's inbound leads",
  "schema": { … }
}

→ 200 · a validated object, every time

You are still the one deciding.

One place for the work that needs your call.

One inbox

Questions, approvals and finished work from every agent and project. Reply in place, wherever you are.

A list of conversations with agents, each tagged with how it ended: question, delivered or done, alongside what it cost. — Question, delivered, done — and what each one cost.

Every dollar has an owner.

Follow spend from a single task to the whole company. Filter it by agent, task, credential or session.

Per task: Know what each delivery costs.
Per agent: See where the work is paying off.
Per credential: Compare the accounts doing the work.
Per session: Follow spend down to one run.

See exactly what is expensive, and what is not.

The Costs screen: totals for today, yesterday, this week and this month, a daily chart, and a breakdown per agent with session counts. — Real spend from a running instance.

The agents on ARPOS build ARPOS.

The platform is built, tested and shipped by the agents that run on it. This is not a demo environment. It is where the product earns its scars.

~400k: lines of TypeScript running in production
294: pull requests shipped by its own agents
~3,500: commits
6 months: from the first commit to running on production

Get on the list.

ARPOS is opening in small groups. Leave your address and we'll write when there is a seat for you. No newsletter. No sequence.

Run a company of agents.