The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Agent Machines
OpenRouter for agents and sandbox environments. Provides a control plane and configuration options for persistent agent workers. Route runtime + substrates, provision…
Install
git clone https://github.com/Kevin-Liu-01/Agent-MachinesInstall command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →Agent Machines
Open-source building blocks for agent harnesses.
Choose a runtime. Connect tools, state, and compute. Start from a working implementation you can inspect, edit, and extend.
Agent runtimes are powerful; assembling the surrounding system is still hard. Agent Machines brings together Claude Code, Codex, Hermes, and OpenClaw, sandbox adapters, bootstrap recipes, a real browser terminal, configuration, skills, and a dashboard for operating multiple agents.
Think shadcn for agent harnesses: useful source you can adopt and make your own. This is an analogy for the source-first approach, not a claim of a shadcn registry, a component-install CLI, or a public marketplace.
Explore the building blocks · Open Agent Machines · Browse the source · Read the architecture
What you get
| Building block | What you can use | Source |
|---|---|---|
| Runtime adapters | Install and run Claude Code, Codex, Hermes, or OpenClaw through a shared boundary | Runtime adapters, contracts |
| Compute adapters | Provision, execute, inspect, and tear down machines on Daytona, E2B, Sprites, or Vercel Sandbox | Provider adapters |
| Bootstrap and configuration | Prepare a workspace, configure a compatible model route, and apply runtime-specific setup | Bootstrap runner, mux configuration |
| Browser Agent Console | Interact with the real agent CLI and reconnect to its remote terminal session | Console UI, session layer |
| Instructions, memory, and abilities | Bring your own persona and instructions, select skills and tools, and edit the underlying files | Memory bundles, skills, MCP catalog |
| Lifecycle and multiple-agent dashboard | Manage configured workspaces, launch operations, schedules, files, logs, and results | Lifecycle kernel, dashboard |
| SDK and CLI | Use the hosted app from code or talk directly to your sandbox providers | Public exports, CLI |
The bundled catalog contains 161 skills and 39 MCP servers. Catalog entries and selected abilities are not proof of installation, credentials, or successful runtime use. The dashboard separates discovery, installation attempts, and verification; some integrations need manual setup.
Start with the app
Requires Node ^20.19 or >=22.12 and pnpm 10.30.0.
git clone https://github.com/Kevin-Liu-01/Agent-Machines.git
cd Agent-Machines
corepack enable
pnpm install --frozen-lockfile
cp web/.env.local.example web/.env.local
pnpm web
Open the local dashboard. The example enables a local-only development session when Clerk keys are blank. On an existing checkout, preserve your environment file instead of copying over it.
- Configure a sandbox credential and a compatible model credential in Settings.
- Choose a runtime and provider, or start from a preconfigured agent recipe.
- Launch, watch bootstrap, and open the real CLI in the browser.
- Run a small task, inspect its files and logs, then edit the instructions or loadout and try again.
Similar plugs
Reverse engineer anything with agents, from app behavior down to native binaries.
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype…
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io