Host your OpenClaw agent.
Your own OpenClaw, always awake, at an address that belongs to you. It answers from the first minute on a free model we provide — bring your own key whenever you want to.
From $14/mo. Cancel anytime.
[01] One agent, your key
Deploy in a minute, keep your key
Sign up, name your agent, paste a model key. It is answering at your own address before you finish reading this page.
For you
Deploy from the dashboard
One screen: state, address, model key, and start, stop, restart. Your files survive restarts.
- OpenClaw chat at you.agentsrun.app
Your key, your provider
Bring your own model
Anthropic, OpenAI, MiMo or any OpenAI-compatible endpoint. The key lives in your sandbox and goes straight to the provider.
Paste it once in the dashboard. Change it whenever you like — your agent picks up the new one in about a minute.
We are not in the request path and take no markup.
[02] Living with it
It is yours, and it stays put
Not a session that expires, not a container that gets recycled overnight. One machine, running, with your name on it.
Awake when you are not
There is no scale-to-zero and no cold start. The agent process stays running, so the first thing you ask at 3am gets the same answer speed as the last thing you asked at noon.
Your files are still there
Everything your agent writes lands on a disk that belongs to your agent alone. Stop it for a month and start it again — the disk, and everything on it, is where you left it.
Stop means stopped
Press Stop and nothing starts it again behind your back — not a health check, not us. It stays off until you say otherwise, and it is not billed differently for it.
[03] Security
What your agent cannot reach
Your agent runs on its own kernel, not a slice of ours. It cannot see our network, the machine it sits on, or anybody else's agent — and you do not have to take our word for any of it.
Its own kernel
Not a container with rules
Ask your agent what kernel it is running on and it answers 4.19.0-gvisor. Ask the machine underneath and it answers 6.8.0-90-generic. Two different kernels, so a syscall that escapes your agent lands in a kernel that has nothing else in it.
We run tenants as root today, and say so. Non-root is planned.
See every reading →And what we can see
The honest list
We run the servers, so the first line is a yes. A page that claimed otherwise would be lying to you.
[04] Reach
A private link, or no public address at all
Default
you.agentsrun.app
A normal address with a login in front of it. Requests that are not yours are refused before they reach the agent. TLS terminates at our edge, not at a CDN.
Tailscale
Inside your own tailnet
The agent joins your Tailscale network and your device rules decide who reaches it. You approve it from your own account; we never hold a key to your network.
[05] Pricing
Simple pricing, no surprises
Flat monthly price. You pay your model provider directly, at their price.
For a personal agent.
- A free model included, metered
- Or bring your own API key
- Files kept across restarts
- 2 vCPU · 4 GB RAM · 8 GB storage
More CPU and memory for heavier agents.
- Everything in Starter
- 4 vCPU · 8 GB RAM · 20 GB storage
For heavier workloads, with priority support.
- Everything in Pro
- Priority support
- 8 vCPU · 16 GB RAM · 40 GB storage
[06] FAQ
Frequently asked questions
Do I need my own API key?
Not to start. Every agent runs on a free model we provide and meter — capped per day, reset at midnight UTC, and we are in the request path for it, which we spell out on the security page. Add your own key whenever you like: Anthropic, OpenAI, MiMo or any OpenAI-compatible endpoint. It is stored in your own sandbox, requests go straight to the provider, we take no markup, and we are out of the path. Remove it later and you come back to the free model.
What can you see?
We run the servers, so your agent's destinations and traffic volume are visible to us at the network layer — though nothing records them today, so we could not tell you what your agent talked to yesterday. We cannot see inside TLS sessions; no root certificate is installed. The one exception is the free model above: while your agent is on it we are the endpoint, so we can see those prompts and replies, and adding your own key ends that. We do not train on your data.
Can I connect Telegram?
Not yet, and nothing on this page should have said otherwise. OpenClaw can install a Telegram channel, but it ships switched off and agentsrun gives you no way to add a bot token from the console. Today your agent answers on the web, at your own address. Messaging channels are on the roadmap.
Where does it run?
In Germany. Your data stays in the EU.
Is there any lock-in?
No. Your files are on a disk you can download, your key is yours, and OpenClaw is open source. Cancel any time.