emperorclaw/emperorclaw
Introduce AI to your company workspace.
What it solves
When you run more than one AI agent, it quickly gets messy: you can't see what each agent is doing, replies get lost in crashes, you repeat the same instructions every chat, deliverable files end up scattered, and a runaway loop can burn through tokens overnight. EmperorClaw is a control center that fixes these problems by giving each agent a home: a shared office with an org chart, task board, shared memory, and budget controls. Everything is recorded durably, so no request disappears, and every action can be traced and approved.
How it works
You self-host EmperorClaw on your own server (Docker installer included). You describe your company and paste a model key; a lead agent writes your "Company Brain" — a wiki of rules, procedures, and know-how that agents read before acting. Then you pick from ready-made worker agents (team lead, SEO, QA, content, accounting, and more) and they get installed and connected. You assign tasks (or just chat), watch each agent's live status — current step, tool calls, running timer — and approve anything important. Agents run fast via a persistent gateway (about 0.2 s overhead per message) and can be managed locally or on remote machines. EmperorClaw also offers automations on schedules, cost caps with warnings and pauses, and self-healing: if an agent goes silent, work is retried and then escalated to a human.
It supports bringing your own agents: Hermes, OpenClaw, or any MCP client including Claude, via an MCP API. Data is stored in PostgreSQL, with scoped permissions and per-client/project separation. The mobile-friendly dashboard lets you manage everything from your phone.
Who it's for
Companies or teams that already run more than one AI agent and want to stop asking "who is doing what?" It's for those who need 24/7 agent work with human approvals, need to keep each client's data separate, want role-based access for the whole team, and prefer self-hosting on their own infrastructure without per-seat pricing. It's also for anyone who's been burned by runaway token costs and wants a budget that actually pauses an agent.
Highlights
- Live status rows show each agent's current step, running timer, task, and cost — no more wondering if an agent is thinking or stuck.
- Durable inbox: every message and task has an owner and state; retries and escalations ensure nothing is lost.
- Company Brain: scoped knowledge (per company, client, project, or agent) that agents read before acting.
- Approval gates: tasks can require your sign-off before closing, with notifications via in-app, email, Slack, or Discord.
- Cost control: per-run tokens and cost, monthly budgets with 80% warnings and pause at limit.
- Automations: scheduled recurring work with approval gates between steps and a run record.
- Self-healing: heartbeats and task leases detect silent agents; duplicate agents go to standby.
- Bring your own agents: works with Hermes, OpenClaw, or any MCP client (including Claude).
- Mobile-friendly: approve, chat, and monitor from a phone.
- Self-hosted under FSL-1.1-Apache-2.0; every release becomes Apache 2.0 after two years.
Related
- Project
- Project
- Project
- Project
- Project