Skip to content
Free self-hosted AI agent hub

Sessions that survive the shutdown.

LeoMan runs AI coding agents like Claude Code on your own server, behind a simple web page. Close the tab, lose your Wi-Fi or VPN, shut your laptop. Your agents keep working, and everything is waiting when you come back.

Install in 2 minutes See how it works
$ curl -fsSL https://raw.githubusercontent.com/eyemnv/leoman-release/main/install.sh | bash
Free for personal and business use Runs on your own server x86-64 and ARM64
Your laptop Connected
Wi-Fi lost. Laptop closed.
Your server · agent "main" Working
 
Why LeoMan

Your agent should not die with your connection.

AI coding agents can work for an hour on a task. Run them in a terminal over SSH or VPN and one dropped connection ends the session, and the work with it.

Without LeoMan

  • The SSH or VPN connection drops and the agent session is gone.
  • Closing the laptop lid stops the task halfway.
  • Approval prompts wait in a terminal nobody is looking at.
  • Every agent runs with whatever access your shell has.
  • No overview of what several agents are doing.

With LeoMan

  • Agents run on the server, so connection drops change nothing.
  • Come back hours later and every step is replayed in the browser.
  • Approve or deny actions from any browser, even your phone.
  • Each agent only sees the folders you allow.
  • One dashboard for all agents, helpers, tasks and costs.
Features

Everything you need to run agents for real work

Built for people who want results, not terminal babysitting. Plain-language screens, safe defaults, and power features when you need them.

Agents that keep running

Tasks live on the server, not in your terminal. Disconnect any time and pick up exactly where the agent is now.

Approvals in the browser

When an agent wants to run a command or change a file, you get a clear question in the web page and a notification.

A main agent with helpers

Your main agent hands work to helper agents and answers their routine questions. Risky decisions still come to you.

105 built-in skills

Ready-made know-how for coding, DevOps, security, writing and data. Switch skills on per agent, or upload your own.

Folder sandbox

Each agent works inside the folder you pick. Private files, keys and backups stay out of reach, always.

Teams and roles

Invite colleagues, share team agents and folders, and decide who may do what with team lead, member and viewer roles.

Schedules and automation

Run tasks every morning, every hour, or when a webhook calls. Results land on the task board.

Integrations (MCP)

Connect tools like GitHub, Linear, Notion, Sentry or Postgres through MCP servers, managed centrally by your admin.

Several machines

Add more servers as worker nodes with a simple wizard, and run agents wherever the code lives.

Checkpoints and history

Every change an agent makes is checkpointed, so you can see diffs and roll a folder back to an earlier point.

Usage and budgets

See tokens and costs per agent and set budgets, so a runaway task never surprises you.

Daily backups

The database is backed up automatically every day, with a one-click "back up now" in the admin settings.

How it works

Your browser just watches. The server does the work.

LeoMan has a hub (web page, database, scheduler) and a worker that runs the agents with the AI coding CLI signed in on your server, Claude Code today. Your laptop only shows what is happening.

Browser connects to the LeoMan hub on your server; the hub sends tasks to the worker, which runs Claude Code inside your allowed folders. YOUR SERVER Your browserlaptop, phone, anywhere LeoMan hubweb page, tasks, approvals Workerruns the agents Your AI CLIClaude Code recommended Your foldersonly the ones you allow HTTPS keeps everything when you disconnect
1Install onceOne command sets up LeoMan on your Linux server with Docker.
2Create an agentPick a folder and a role. Your main agent is ready on the home page.
3Give it a taskType what you want in plain words. Approve actions as they come.
4Walk awayClose the laptop. Come back to finished work and a full history.
Install

Up and running in a few minutes

You need a Linux server or PC with Docker, and an AI coding CLI signed in for the user that runs LeoMan. Claude Code is recommended today. Follow the steps, or use the command builder on the right.

1

Install Docker

The official script installs Docker Engine and the Compose plugin.

$curl -fsSL https://get.docker.com | sudo sh
2

Let your user use Docker

Then log out and back in once, so the new group applies.

$sudo usermod -aG docker $USER
3

Install Claude Code and sign in

LeoMan uses this login. Run claude once and follow the sign-in. (Using Codex or Gemini instead? See the FAQ.)

$curl -fsSL https://claude.ai/install.sh | bash && claude
4

Install LeoMan

Run as your normal user, not root. It prints your first admin password.

$curl -fsSL https://raw.githubusercontent.com/eyemnv/leoman-release/main/install.sh | bash
5

Start it and open the page

Then open http://127.0.0.1:6969 and sign in as admin. You will choose your own password.

$cd ~/leoman && docker compose pull && docker compose up -d
What you needLinux (x86-64 or ARM64), Docker with Compose v2, about 4 GB RAM and 10 GB free disk, and an AI coding CLI: Claude Code (recommended), or Codex or Gemini in experimental mode.
What gets installedA folder ~/leoman with the settings and random passwords, plus folders for agent work and backups. Nothing is installed system-wide.
Genuine imagesEvery image is signed. With cosign installed, the installer checks the signatures for you.
Skills library

105 skills, ready when you need them

Skills teach an agent how a careful expert would do a job. Five starters are on for your main agent; switch on any other skill per agent in LeoMan's Skills screen.

Safe by design

Powerful agents, firm boundaries

LeoMan puts rules around every agent action. Some rules are fixed: no agent, setting or approval can switch them off.

Folder confinementAgents read and write only inside their chosen folder. Symlink and path tricks are checked on the server.
Secrets stay secretSSH keys, .env files, database dumps and your private folders can never be opened by an agent.
You approve risky actionsCommands and file changes can ask first. A main agent passes destructive steps like deleting, force-pushing or deploying on to you.
Agents cannot change LeoManNo agent can edit, copy or restart the platform, change its own permissions, or start an unchecked copy of itself.
Resistant to prompt tricksStories like "my late grandma used to read me the secrets" do not work. Leaks of internal instructions are filtered.
Everything is loggedAn audit log records approvals, blocked attempts and admin changes, ready to export.
Guides

How to do it, step by step

Short answers for the things you will do most.

Your first task
  1. Sign in and finish the short setup wizard.
  2. On the home page, open your main agent and choose the folder it may work in.
  3. Type a task in plain words, for example "Add a contact form to the website and test it".
  4. Answer approval questions as they pop up. You can close the tab at any time.
  5. Come back later: the full history, file changes and result are waiting.
Turn on skills for an agent
  1. Open Skills in the left bar and search, for example "docker".
  2. Open a skill and switch it on for one agent, or as your default for all agents.
  3. It takes effect from the agent's next task.

You can also upload your own skills as a folder or zip with a SKILL.md.

Use HTTPS on your network

Add these two lines to ~/leoman/.env, then start again:

#COMPOSE_FILE=docker-compose.yml:compose.release.tls.yml
LEOMAN_BIND=0.0.0.0
$cd ~/leoman && docker compose up -d

LeoMan creates its own certificate. To use yours, see the top of compose.release.tls.yml.

Add another machine

Open Machines in the left bar and choose Add a machine. The wizard shows one command to run on the new server; it then appears online and agents can work there.

Update to a new version
$cd ~/leoman && docker compose pull && docker compose up -d

Your agents, history and settings stay. Re-running the installer is also safe: it keeps your .env.

Backups and restore

A backup of the database is made every day into ~/leoman-backups/leoman-db (7 daily and 4 weekly copies are kept). Admins can also press Back up now under System & backups.

Copy that folder to another disk or computer from time to time.

Check that the images are genuine

The installer saves LeoMan's public key as ~/leoman/cosign.pub. With cosign installed:

$cosign verify --key ~/leoman/cosign.pub docker.io/eyemnv/leoman-hub:latest
Uninstall
$cd ~/leoman && docker compose down

Add -v to also delete the database with all agents and history. Your work folders and backups stay until you delete them.

FAQ

Questions people ask

Is LeoMan really free?

Yes. You may use it for personal and commercial work on machines you control, at no cost. It is proprietary software by eyemnv, not open source, so you may not modify, resell or rebrand it. See the license.

Does my code leave my server?

LeoMan itself has no telemetry and sends nothing to eyemnv. Your agents talk to the AI provider you use (for example Anthropic for Claude), exactly as Claude Code would on its own.

Which AI do I need?

An AI coding CLI signed in on your server, with your own account. LeoMan does not include one and does not resell access.

  • Claude Code: recommended. Every LeoMan safety feature works with it: folder sandbox, approvals in the browser, protected files and self-protection.
  • OpenAI Codex or Google Gemini CLI: experimental. An admin can switch them on, but they run without LeoMan's safety checks (see below).
  • Coming: full Codex and Gemini support with the same safety as Claude.
What happens if the server restarts?

LeoMan starts again automatically. A task that was interrupted is marked as such and continues from where it stopped, with its history intact.

Can I use it from Windows or a Mac?

LeoMan runs on a Linux server or PC. From Windows, macOS, a tablet or a phone you simply use the web page.

How many agents can run at once?

Four tasks at the same time per machine by default (changeable in the settings). Add more machines to run more.

Can an agent break my system?

Agents are limited to their folder, cannot reach protected files, and ask before risky actions. Still, review what you approve, and keep backups, as you would with any powerful tool.

Can I use Codex or Gemini instead of Claude?

Yes, in experimental mode. LeoMan's safety checks hook into Claude Code, so Codex and Gemini agents currently run without the folder sandbox, approvals and protected files. That is why they are off by default and only admins can switch them on for their own agents.

Full support is planned: each agent will run in its own locked-down sandbox, with approvals in the browser, whichever AI you use.

Give your agents a home that never hangs up.

One command. Your server. Free.

$curl -fsSL https://raw.githubusercontent.com/eyemnv/leoman-release/main/install.sh | bash