Remote Agents — Documentation skill for Claude Code

Remote Agents

Run **Cursor, Codex, and Claude Code** on your own machine. Access your agents from anywhere, follow their progress, manage tasks, and share reports through a mobile-first interface.

What you can do

• Run multiple agents, each with its own conversation, model, workspace, and task queue.
• Send instructions, stop a task, or remove queued work.
• Follow readable activity summaries or inspect the full execution log.
• Find automatically collected reports in Documents, then read, copy, or share links to them.
• Read Markdown on mobile with section navigation, a focused reader, and scrollable code and tables.
• Configure your fleet and check provider and connection health in one place.

How it works

A worker on your computer runs the agents and connects outbound to your Linux control server over WSS. You use the web app over HTTPS to send tasks, receive updates, and read documents. The control server relays commands; the agents execute on the worker machine.

**Agents have full access to files available to the worker user, with approval prompts disabled.** Give access only to trusted operators. OS permissions still apply. See [SECURITY.md](SECURITY.md).

Default fleet

**The default setup is customizable.** Choose your agents, providers, supported models, names, and working directories in a fleet JSON file. Set `AGENT_FLEET_FILE` in the worker’s private environment file and restart the worker. Start with [examples/fleet.defaults.json](examples/fleet.defaults.json); see [AGENTS.md](AGENTS.md) for the schema and persistence rules.

Without an override, the worker starts these eleven agents:

Provider	Agents
Cursor	Agent 1 and Agent 2 (Grok 4.6), Agent 3 and Research (Composer 2.5)
Codex	Astra 1 and Astra 2 (GPT-6 Astra, Medium), Sol (GPT-5.6 Sol, Medium)
Claude Code	Fable 5.1, Fable 5, Opus 5, Opus 4.8

Each provider requires its own authentication and model access. All eleven slots remain visible if a