Self-Hosted AI Agents.
Own your files. Own your keys. Own your future.

Run a team of AI agents on your own Ubuntu VPS. No SaaS subscription, no vendor lock-in, no telemetry — just a folder of plain-text files you can read, audit, and version-control. Powered by OpenClaw, the open-source AI gateway we maintain.

Zero lock-in · Works with 🧱OpenClaw CClaude GChatGPT Gemini

Why self-host your AI agents?

🔒

Your keys, your data

Your Anthropic / OpenAI / OpenRouter API keys live on your server. Conversation history, memory, and agent files never touch our infrastructure after download. There is no "cloud dashboard" storing your prompts.

📄

Plain-text, auditable

An agent is seven markdown files describing its identity, soul, behavior, tools, and memory. Diff them in git. Copy them between servers. Ask Claude to review them. No black-box container, no proprietary format.

No vendor lock-in

OpenClaw is open source. The agents you generate here work on any Ubuntu box — ours, yours, or a client's. You can walk away from OpenAgent.Mom at any point and keep every agent you've built.

From bare VPS to running agent in three hours

1
🖥️

Spin up an Ubuntu VPS

Any small provider works — Hetzner, DigitalOcean, Vultr, Linode. Around €5-10/month for a box that comfortably hosts multiple agents. SSH in, follow the guide.

2
🔧

Install OpenClaw from the guide

Our free 68-step setup guide is tested end-to-end. No assumptions, no "obvious" leaps. Copy-pasteable commands. You finish with a working gateway accepting agent workspaces.

3
📦

Drop in your agent bundle

Generate an agent with our wizard, download the ZIP, upload to ~/.openclaw/, reload. Done. Do it again for every agent you want.

What you get in every agent bundle

📝

7 workspace files

IDENTITY.md, SOUL.md, AGENTS.md, TOOLS.md, USER.md, HEARTBEAT.md, MEMORY.md. Everything the agent knows, wants, and remembers — in plain text.

⚙️

Config snippet

JSON block to merge into your openclaw.json. Tells the gateway where to find this agent, which model to use, which channels to connect.

📄

DEPLOY.md playbook

Step-by-step deployment instructions for the specific agent you built. Commands to run, files to edit, credentials to inject.

🎓

Birth certificate

Decorative HTML + PDF with a unique serial, the agent's archetype, and a QR code to verify authenticity. Frame it, ignore it, or use it to impress clients.

Questions from people who self-host everything

Do I need to run the DIY guide myself, or can you set it up for me?

Both. The DIY guide is free forever — use it to self-install on your own box. If you'd rather not spend an afternoon on sysadmin, the Custom Setup Service installs OpenClaw on your VPS for you, tested and handed over ready-to-run.

What if I want to run agents on Claude, ChatGPT, or Gemini instead of OpenClaw?

The wizard exports to all four. Same interview, different output format. Pick OpenClaw for full self-hosted workspaces, or pick Claude/ChatGPT/Gemini to get a drop-in Skill / Custom GPT / Gem instead. Start the wizard →

Is OpenClaw actually open source?

Yes. It's the AI gateway we maintain and run ourselves. The agents we generate plug into it as plain files — nothing in the generated bundle is tied to our SaaS. You can fork OpenClaw, patch it, or walk away entirely and everything you built still runs.

Can I run a team of agents that coordinate?

Yes — that's what the Team Builder is for. Visually design a team of agents, draw handoff connections, generate a full multi-agent deployment bundle. Free to use. Every agent gets its own workspace folder; they share a single OpenClaw instance.

Ready to own your AI stack?

Start with the free guide, generate your first agent, and deploy it on your own server this week.

Open the Free Guide Generate an Agent