Product · Agents

Agent-native.

Works with Claude Code, Codex, DOSSBOT, and whatever harness you bring. Every remoco command is directly invokable from the agent. No adapter layer, no MCP dance. The workstation is the agent's runtime.

The prompt-to-shipping loop

You paste a prompt into your agent. It opens an SSH session to your workstation, runs remoco commands, sees the results, iterates. By the time you circle back, there's a live URL and a branch to review.

# paste this into Claude Code or Codex
set me up on a fresh remoco workstation for this repo.
install deps, start the dev server, expose it private,
and report the URL.

Why "native"

Most dev environments assume a human is typing. Agents assume:

• Commands produce structured output they can parse
• Long-running processes can be backgrounded without screen sessions
• URLs are stable and shareable (no "copy this ngrok token to your prompt")
• Secrets resolve automatically (Doppler injection, not .env edits)

Remoco is built to those expectations. tmux wraps every dev server. Every command takes --json. URLs are derivable from the shell.

Skills that ship with the workstation

On first boot, the workstation pulls down a skills pack that teaches the agent how to operate the stack. Each skill is a small markdown file with frontmatter triggers. They live at ~/.remoco/skills/.

• remoco — core operating guide (SSH, tmux, expose, worktree)
• remoco-artifact — publish HTML reports via the control plane
• remoco-ds — load a custom design system for artifact generation
• remoco-debugger — interpret the overlay dashboard via HTTP

Bring your own harness

The remoco CLI is a shell binary. Anything that can exec a command can drive it. Supported today:

← Live debugger overlay Back to product index →