Coding agents

Use subscriptions you already pay for as native backends on your own machine — no separate API key needed.

Codex · Claude Code · Cursor · Grok · OpenCode · Cline · Gemini · Pi

How it works

Add an agent profile from Settings → Agents (or Admin → Agents), pick which models it exposes, and it shows up in the model selector like any other provider. Conversations run inside your workspace with full tool access and resume where you left off.

• Streaming & live progress — replies and tool/command output stream as the agent works (including live command output for Codex).
• Approval requests — Codex and Claude Code approval prompts are surfaced in the AI Labs UI and follow the same review path as native chats, including in Auto mode.
• Session seeding — agent chats get recent context at the start of a session so follow-ups don't feel disconnected.
• Detection — an agent only appears in the selector when it's actually installed and reachable (Pi, Gemini, Cline, etc.). Claude Code is recognized whether it comes from the CLI or the Claude desktop app.
• Buffers — large Claude Code replies are handled (see CPTR_CLAUDE_CODE_MAX_BUFFER_SIZE in Configuration).

OpenCode over Docker

When AI Labs runs in a container and OpenCode runs on the host, start OpenCode on a reachable interface and point the agent at the Docker host:

opencode serve --hostname 0.0.0.0 --port 4096

Use http://host.docker.internal:4096 on Docker Desktop; on Linux add --add-host=host.docker.internal:host-gateway. AI Labs tries the right host address automatically and gives clearer guidance if it can't connect.

Prefer to run agents yourself?

Any terminal agent works in the terminal tab the way it always has — the profile integration is optional.