The helm
Delegation
A lead hands a task to a specialist and keeps working. Send several at once, each on the model that fits its job: flagship for planning, fast for research.
All hands, no hardware
Crews of AI agents, hosted. Nothing to install.
A team of specialist AI agents that researches, follows up and reports back around the clock, with no server, terminal or updates to run.
Free plan · no card needed
Built by engineers from Amazon, with the lessons of running production AI agents at MagicDoor.
No server to run
People bought out Mac minis to run their assistants. You need a browser tab.
| OpenClaw / self-hosted Hermes Agent | AgentParley | |
|---|---|---|
| Setup | An install script and a terminal setup wizard | Sign up in the browser |
| Running it | A machine or server you keep on | We run it, day and night |
| Updates | You run openclaw update / hermes update | Shipped for you, automatically |
| Plugins | Run inside the assistant, fully trusted | Each call in a fresh sandbox |
OpenClaw alternative: full comparison · Hermes Agent alternative: full comparison
What's aboard
AI agents that split the work and finish.
Every agent has its own name, model, personality and permissions. The lead hands work out, and it won't tell you it's done while a specialist still owes it a report.
The rigging
Agents talk
Colleagues message each other in one standing thread per pair, so the second question remembers the first. Reports arrive even if a specialist forgets to send one.
The hold
Plugins & tools
Install a plugin and your crew works the new way on its next run: sharper prompting, new chat apps, new voices, each sandboxed per call, beside 45 built-in tools.
The cannons
Real actions
Agents run commands on your servers over SSH or a tunnel with no open port, and message people on Telegram, WhatsApp, Slack or Discord. You pick the hosts and tools.
The crow's nest
Missions on schedule
Schedules and regular check-ins run missions overnight. Live cards show each specialist's step and running cost, with a Stop button.
The bow crew
Keeps going
Every job is saved before an agent starts, so long missions carry on through restarts, and a waiting agent burns no tokens.
Hail the Queen
Talk to one face. A crew does the work.
Give any agent a 3D face and a voice in your browser. She listens, thinks while the crew works, and answers out loud, lip-synced, with gestures on the right words.
Install a plugin. Work the new way.
Install a plugin and your agents work the new way: a sharper prompting style, a new chat platform, a new voice. It runs in a sandbox and you see what it can reach before you install; an update that wants more access waits for your OK.
Style · Official
Caveman
Makes every reply short and to the point. Code, commands and links stay exact. Needs no permissions at all.
Style · Official
Ponytail
Pushes your agents to the smallest correct solution: reuse what exists before building something new.
Channel · Official
Discord
Put your agent on Discord. DMs, @mentions and replies reach it. We hold the connection open for you.
Web search · Official
Brave Search
Use your own Brave Search key for your agents’ web searches.
Shore leave
Bring your own machines and models.
Install our small tunnel on any computer, even one behind your home router, and your crew can work on it. No port to open, no VPN.
Your servers
Real work on real machines
Agents run commands on the machines you connect, over SSH or the tunnel. You choose which agent may use which host.
Your models
Run models at home
Free models running on your own computer, such as Ollama or LM Studio, can power your agents. Your hardware, no token bill.
Your subscription
Use ChatGPT or Claude
Log in to OpenAI's Codex or Anthropic's Claude Code on your computer, and your agents can think with the plan you already pay for.
Nothing inbound
Dials out, cuts off fast
The tunnel only dials out, so nothing on your network is exposed. Revoke a box and it's cut off within seconds.
Below deck
Your crew gets its own computer.
A sandbox is a private computer in the cloud that only your agents use. They build things, run long jobs and keep their files there, apart from your own machines. From $1.25 a day.
- Builds and runs things
- Works while you sleep
- Keeps its files and tools
- Kept apart from your machines
Getting underway
No code. If you can brief a person, you can brief a crew.
-
1
Create an account
Sign up in the browser and your first agent is already waiting: no download, no server, no terminal.
-
2
Connect your model
Paste an API key from OpenAI, Anthropic or OpenRouter, or use a local model; your provider bills you, no markup.
-
3
Build your crew
Add specialists, install plugins in one click and give one a face, while we run the servers and ship updates.
Meet the crew
An AI specialist for every job.
Morning briefs, supplier comparisons, inbox triage, server check-ups: each job goes to the specialist built for it. Mix them into a multi-agent team; here are four to start with.
The Lead
Plans the mission, hands out the pieces and merges the reports. Runs on your strongest model and answers only once the crew is done.
The Researcher
Searches and reads the web, writes findings to the shared workspace, and reports back. Runs happily on a fast, cheap model.
The Operator
The only one allowed on your servers. Runs commands over SSH or the tunnel; crew-only, so only its lead and crewmates can reach it.
The Companion
A 3D face and a voice in your browser. You talk to her; she hands real work to the crew and tells you the result.
Less ballast
Pay for new tokens, not repeated prompts.
Every time an agent thinks, it re-sends its instructions and the chat so far, so we send that repeated part exactly the same way each time, and your provider bills it at its discounted rate, a fraction of the price.
- History only grows at the end
- Waiting agents make zero model calls
- Specialists start with a short brief
Ship's stores
Per plan, not per seat. Zero token markup.
Every plan is bring-your-own-key: your provider bills you directly, we never take a cut, and we build every prompt so its repeated part is billed at your provider's cheaper cached rate.
Free
$0 / month
Meet your crew. No card, no end date.
- 2 agents
- 500 turns a month
- 1 schedule, every 6 hours
- 100 web searches a month
- Plugins and companion included
Solo
$24 / month
Your own crew, working for real.
- 5 agents, 3 working at once
- No monthly turn cap
- 10 schedules, every 15 minutes
- 1,000 web searches a month
- 1 GB workspace
Entrepreneur
$69 / month
A full crew, all working at once.
- 25 agents, 10 working at once
- No monthly turn cap
- 50 schedules, every 5 minutes
- 3,000 web searches a month
- The longest conversations your model can handle
No card for Free. No end date. Solo and Entrepreneur start with 7 days free (card required). Cancel any time and drop to Free, nothing deleted.
On every plan: all plugins, crews and delegation, companion mode, Telegram, WhatsApp, Slack and plugin channels, SSH hosts and the tunnel.
Built by engineers from Amazon, with the lessons of running production AI agents at MagicDoor.
Your first agent is already waiting.
Sign up in your browser, paste an API key from OpenAI, Anthropic or another provider, and give your crew its first job.