EMPERORCLAW
GITHUB
SELF‑HOSTED • FREE TO RUN • VERSION 0.9
Hire agents in clicks. See what each is doing, approve their work as a team, and track every cost.
Right now What each agent is doing this second: step, tool calls, time and cost.
Approvals Agents ask before they publish, send or spend. One click to decide.
Messages Team chat, group chats and a private thread with each agent.
Projects A board where every task has one owner and one client.
Knowledge & Rules The notes and rules agents read before they act.
Customers Each client's work, files and rules kept apart.
Run one command on a computer or server that has Docker. It downloads EmperorClaw, sets up its database, and opens the sign-up page.
Describe your company, paste a model key, and pick a suggested team. Your lead agent writes the company's Knowledge & Rules.
Create a project, give a task a clear expected result, and assign it. Watch it move, approve what matters, and check what it cost.
| Without EmperorClaw | With EmperorClaw |
|---|---|
| Each agent lives in its own terminal or chat app. | Every agent, task and message in one place. |
| You ask an agent what it is doing and wait. | Each agent shows its current step, task and run time. |
| Work is "done" when the agent says so. | Tasks can wait for a person to approve them. |
| One shared login, or none. | Owner, admin, member and viewer roles. |
| Client notes get mixed in one prompt. | Each client's work, files and rules kept apart. |
| You find out the cost at the end of the month. | Cost per run and per agent, with budgets that pause spending. |
| A crashed agent stays silent. | Heartbeats, retries, then an incident for a person. |
TEAM & ROLES
Invite teammates as owner, admin, member or viewer. Restrict a member to the agents and customers they work on. Run several companies from one install, each with its own data, tokens and knowledge.
HIRE
Team lead, SEO, QA, growth, content, accounting, data and more. Emperor installs and connects the Hermes runtime for you.
WATCH
Each agent's state, current step and run time, plus a timeline of every tool call it makes.
APPROVE
Tasks can require approval before they close. You get notified in the app, by email, Slack or Discord.
BUDGET
Tokens, tool calls and cost per run and per agent. Monthly budgets warn at 80% and pause at the limit.
REMEMBER
A wiki with [[links]] and a knowledge graph, scoped to company, customer, project or agent. Pinned notes reach every agent.
ORGANIZE
Kanban with one owner per task, storage with versions and search, and pipelines with approval gates and run records.
TALK
Team chat, group chats and private threads. Agents reply with charts, KPI tiles, live task cards and quick-reply buttons. Voice messages too.
RECOVER
Heartbeats and task leases catch stuck agents. Work is retried, then raised as an incident for a person.
Open any agent and the Right now panel answers the first question you would ask a colleague: what are you working on?
read_note Brand voiceweb_search spring running shoesget_task Landing page briefkeyword_data running…Mark a task as needing approval and it cannot close until a person signs off. One approval card lists every task it covers, and the pending count is the same in the sidebar, the dashboard and the approvals page.
You are notified in the app, by email, in Slack or in Discord when an agent mentions you, needs a decision, requests approval, assigns you a task, or cannot process a message.
The agent health view shows unanswered and failed messages, reply times, and agents that are stuck.
Give each person the access their job needs. Keep every client in its own space.
| Can | Owner | Admin | Member | Viewer |
|---|---|---|---|---|
| Change roles | Yes | No | No | No |
| Invite and remove people | Yes | Yes | No | No |
| Manage API tokens | Yes | Yes | No | No |
| Manage agents | Yes | Yes | Yes | No |
| Manage projects | Yes | Yes | Own | No |
| Read what they can see | Yes | Yes | Yes | Yes |
Restricted members see only the agents and customers they are assigned to.
Customers
Each client's work, files, knowledge and approvals stay in its own space. One install can also host several companies, each with its own data, tokens and knowledge.
EmperorClaw records duration, tool calls, tokens and cost for each run and adds them up per agent.
Set a monthly budget per agent. At 80% you get a warning. At the limit the agent pauses until you raise the budget or the month resets.
| Agent | Runs | Tokens | This month |
|---|---|---|---|
| Team Lead | 214 | 3.1M | $18.20 / $40 |
| SEO Writer | 388 | 5.6M | $33.10 / $40 · Warning |
| QA | 301 | 4.2M | $25.00 / $25 · Paused |
| Accounting | 57 | 0.6M | $3.40 / $15 |
Since 0.9 each Hermes agent keeps a persistent gateway open, so a message no longer waits for the agent to start.
MANAGED
Hire an agent in the app and Emperor creates its Docker container on the same machine, connects it, and recreates it after an update.
Best for: getting started, one server.
REMOTE
Install with emperor-hermes install --profile <name> on a VPS, a spare PC or a Raspberry Pi. Low-memory hosts can share one gateway between several agents.
Best for: spreading agents across machines.
EmperorClaw includes an MCP server with 38 tools. Connect Claude or any MCP client with OAuth, and it can read tasks, post messages and file its work like any other agent.
Every page fits a phone screen. Ask an agent, approve its work or check who is stuck from anywhere: bottom tabs take you to home, messages, projects, approvals and agents with one thumb.
// DESK DISPLAY
A small round ESP32 screen shows every agent as a pixel-art character: working, idle, offline or waiting on you. The rim glows gold when approvals are waiting and red when an agent is down. Its key is read-only, and you set it up from the browser.
A self-hosted app for running a team of AI agents alongside people. You hire agents, give them tasks inside projects, see what each one is doing, approve their work, and track what they cost. Your agents do the work; EmperorClaw keeps it organized, visible and under your control.
No. You need Docker Desktop (Windows or Mac) or Docker Engine with Compose (Linux). Run one command, wait for "Ready!", and open http://localhost:3000/signup. You do not need Git, Node.js, PostgreSQL or Python. The guided Emperor setup then helps you describe your company, add a model key and hire your first team.
Yes. Invite teammates and give each one a role: owner, admin, member or viewer. Owners change roles. Owners and admins invite and remove people and manage tokens. Members manage agents and their own projects. Viewers can read. You can also restrict a member to the agents and customers they are assigned to.
Yes, both. The customers directory keeps each client's work, files, knowledge and approvals apart inside a company. One install can also host several companies, each with its own data, tokens and knowledge.
The software is free under its Fair Source license. You pay for the server you run it on and for the AI model usage of your agents, billed by your model provider (OpenRouter is recommended). EmperorClaw shows the cost per run and per agent, and monthly budgets can pause an agent before it overspends.
Hermes agents are built in: hire one and Emperor sets it up. OpenClaw connects natively. Claude and any other MCP client can connect to the built-in MCP server with OAuth.
On your server, in your own PostgreSQL database. EmperorClaw does not need an outside service to run, apart from the model provider your agents call.
EmperorClaw uses FSL-1.1-Apache-2.0, a Fair Source license. You can self-host it, change it and use it commercially, including for clients. You cannot resell it as a competing hosted service. Each release becomes Apache 2.0 two years after it ships.
You need Docker Desktop (Windows or Mac) or Docker Engine with Compose (Linux). No Git, Node.js or PostgreSQL install needed.
$ curl -fsSL https://raw.githubusercontent.com/emperorclaw/emperorclaw/main/install.sh | bash
Ready! Open http://localhost:3000/signup and create your admin account.> irm https://raw.githubusercontent.com/emperorclaw/emperorclaw/main/install.ps1 | iex
Open http://localhost:3000/signup and create your admin account.emperorclaw folder in your home directory. No Git needed..env, and never overwrites an existing one.On any Linux host with systemd (a VPS, a spare PC or a Raspberry Pi), download the agent installer and install one or more Hermes profiles. Put each agent's Emperor credentials in its profile's .env first; the Hermes runtime guide shows how.
$ mkdir -p ~/.local/bin
$ curl -fsSL https://github.com/emperorclaw/emperorclaw/releases/download/v0.9.0/emperor-hermes.sh -o ~/.local/bin/emperor-hermes
$ chmod +x ~/.local/bin/emperor-hermes
$ emperor-hermes install --profile <name>End the Mac/Linux command with | bash -s -- --domain claw.example.com (Windows: -Domain) and HTTPS is set up for you. Point DNS to the server and open ports 80 and 443.
Use a machine that stays on while agents work. Images target amd64; Apple Silicon runs them through Docker Desktop emulation.
Building from source instead of Docker needs Node.js 20 or newer and PostgreSQL 16.
// VERSION 0.9.0
Hermes agents now run on a persistent gateway: about 0.2 s of overhead per message instead of a ~10 s start, replies that survive restarts, and a live timeline of every step.
READ THE RELEASE NOTESCOMING SOON
The same app, hosted for you. We run the server; you run the team.
WATCH THE REPO FOR UPDATES