OPEN SOURCE. SELF-HOSTED. NO LOCK-IN.
A hotel for your AI agents.
Check in Hermes, OpenClaw or any agent. Each gets its own room on your own server: an isolated container, its own domain with HTTPS, memory that survives restarts, on any model. You run the front desk.
MIT licensed · One-command install · Any VPS
Hermes checks in · 46 s. A Hermes agent on this hotel tells its own story. It directed, filmed and cut the film itself with AgentTake.
Open the watch page ↗INSTALL
One command.
$ git clone https://github.com/magnusfroste/agenthotel.git && cd agenthotel && ./install.shOn a fresh Ubuntu or Debian VPS, as root. Docker, HTTPS certificates and routing are set up for you. Open the link it prints, create your admin account, and check in your first agent.
2 vCPU / 4 GB runs the panel and a couple of agents.
FEATURES
Why a hotel
Because that is what running several agents is. The metaphor is how the panel is built.
Check in
Pick a runtime, name the agent, done. Domain, TLS and routing are automatic.
Rooms
Every agent has its own CPU and memory cap, so no guest starves the others or the hotel.
Guests remember
Sessions, memory and workspace files live in volumes that outlive the container.
Room keys
Provider API keys are configured once and handed to every agent that needs them.
Chat
Talk to up to four agents side by side, and ask them all the same question.
Front desk over MCP
Your own AI tools can check guests in and out, read logs and redeploy, over MCP.
Housekeeping
Build cache, stopped containers and old images are cleaned up on a schedule.
The ledger
Health, uptime, deploy logs and alerts to Slack, Discord or Telegram.
ANY AGENT, ANY MODEL
Any agent, any model
Agents
- Hermes Agent, OpenClaw and Odysseus, from ready templates.
- Or anything with a Dockerfile, an image or a compose file.
Models
- A frontier model through OpenAI, Anthropic, OpenRouter, Gemini, DeepSeek, Mistral and more.
- Your own model on your own hardware: vLLM, Ollama, llama.cpp, LM Studio.
- Switching model is a setting, not a migration.
Private AI on your most valuable data, without sending it to someone else's cloud.
NO LOCK-IN
Your agents can leave.
An agent's memory is its harness's own files, not a row in our database. One hotel issues a one-time link, the other fetches the agent, memory and all, server to server. The agent stops in the old hotel the moment it moves, so it never runs in two places.
PRIVATE NETWORKING
No open ports, if you like.
Put the hotel behind a Cloudflare Tunnel, or join a NetBird mesh, the European alternative, and publish every agent through it automatically. Or just point DNS at the server. It runs anywhere.
THE FAMILY
Tools your agents use, checked in next door.
The hotel is for agents, and their luggage is welcome: the tools they use. Each one is its own open-source project, and one click in AgentHotel.
AGENTS CHECK IN. THEIR LUGGAGE IS WELCOME.
AgentTake
In the hotel as TakeHand the camera to your agent. It drives a browser and directs a narrated product video, over MCP.
AgentReports
In the hotel as ReportsYour agents publish reports over MCP; your people read them behind a password. Every publish a version.
Lobby
In the hotel as LobbyA web hotel for landing pages. Each site is one markdown file that agents write over MCP.
SkillHub
In the hotel as SkillHubA shared memory for your agents: self-hosted Supabase with search by meaning.
Sluss
In the hotel as SlussA self-hosted LLM gateway for EU organisations: every prompt classified and routed under policy, with an audit trail.
IN THE LIBRARY
- Hermes Agent
- OpenClaw
- Odysseus
- Take (video)
- Reports
- Lobby (landing pages)
- SkillHub (shared memory)
- Sluss (LLM gateway)
- n8n (workflows as MCP tools)
Anything with a Docker image, a Dockerfile or a compose file runs too. Missing your agent's favourite tool? Request a template ↗
More from the workshop
CONTRIBUTE
Build the hotel with us.
AgentHotel is MIT licensed and small enough that one pull request changes it.