agent-machines

OpenRouter for agents and containers.

Agent Machines is the product layer above sandboxes — a control plane that deploys a persistent agent worker in one unit: runtime, skills, MCP, integrations, cron, observation, and fleet management — on any substrate.

Live site: https://www.agent-machines.dev
Source: https://github.com/Kevin-Liu-01/agent-machines

What it is

OpenRouter for agents and containers. One account to route which agent runtime (Hermes, OpenClaw, Claude Code, Codex) and which substrate (E2B, Sprites.dev, Dedalus Machines) — then get a full persistent worker, not a bare sandbox.

Vercel on AWS for substrates: we are the product layer; providers are interchangeable infrastructure underneath. People don't want an empty box — they want a worker that audits code, runs on a schedule, and accumulates skills. Dedalus currently benchmarks best on boot latency and sleep/wake among our providers (strong default, not the product). Building sandboxes is hard; we route instead of rebuilding infra.

Specialist fleet

Design agent, news agent, code agent — spin each up from opinionated presets (Hermes, OpenClaw, etc.). Named vendor SKUs (e.g. Anthropic design modes) are the same recipe: UI + skills + MCPs + system prompts. Here you provision each specialist in one click and supervise the fleet from one dashboard: activity, chat, cron, logs, cost.

Where the project is right now

• Three live substrates. E2B, Sprites.dev, Dedalus Machines — each implements MachineProvider. Dedalus leads our harness on boot (~250ms) and sleep/wake; all three are first-class.
• Four agent runtimes. Hermes, OpenClaw, Claude Code, Codex CLI.
• Browser provisioning. /dashboard/setup — credentials, agent, provider, spec, model, provision.
• 161 skills + MCP catalog. Registry-driven loadout; syncs to ~/.agent-machines/ on deploy/reload.
• Optional Cursor bridge. Playwright MCP on every Hermes machine by default.

Architecture

you
  | browser / CLI / API
  v
Next.js control plane ---------------- npm run deploy / chat / reload
  | Clerk UserConfig                                      |
  v                                                       v
MachineProvider ---------------------- E2B | Sprites | Dedalus
  | provision / wake / sleep / exec
  v
/home/machine  (persistent volume)
  |
  |-- :8642 agent gateway
  |-- ~/.agent-machines/         skills, mcps, chats, crons, sessions
  |-- /home/machine/agent-machines/   git checkout for reload
  v
OpenAI-compatible model endpoint (configurable)

This repo is the control plane, not the Hermes agent package.

Quick start

git clone https://github.com/Kevin-Liu-01/agent-machines
cd agent-machines
cp .env.example .env
npm install
npm run deploy   # CLI path; or use /dashboard/setup for any provider

CLI

npm run deploy             # provision + bootstrap Hermes
npm run deploy:openclaw
npm run chat -- "msg"
npm run status / logs / wake / sleep / destroy -- --yes
npm run reload / doctor

Web app

cd web && cp .env.local.example .env.local && npm install && npm run dev

Open http://localhost:3210.

Repository layout

agent-machines/
  src/           CLI + bootstrap
  knowledge/     skills, mcps, VISION.md, AGENTS.md
  mcp/           cursor-bridge
  web/           Next.js site + dashboard

License

MIT.