An Interactive Guide

Agent
Pathfinder

Discover how to deploy every agent your teams need on one governed foundation.

Every way your developers use agents, on infrastructure you control.

Three Paths. One Coder Deployment.

Three ways your developers run agents, all on one Coder deployment.

Agent Relay

Agent Relay connects Cursor Cloud Agents and Claude Code cloud sessions to self-hosted Coder workspaces. The vendor runs the agent and the model. The work, like reading code, running commands, and reaching internal systems, happens in your workspace.

Agent CLIs in workspaces

Claude Code, Codex, or custom agents run inside Coder workspaces, with model traffic routed through AI Gateway. Developers work hands-on in a terminal or IDE. Ready-made agent modules are in the Coder Registry.

Coder Agents

Coder's native agent harness runs on the Coder control plane, so workspaces need no API keys or agent software. It works with the model providers you configure, and runs air-gapped with a self-hosted model. Hand off work through the chat UI or API. Built into Coder.

Three Questions

Answer for your whole organization. You can change answers at any time.

Your Agent Paths

Compare the Paths

Where each part of the work runs, and what that means for credentials, offline teams, and keeping a record of AI use.

How Agent Relay, agent CLIs in workspaces, and Coder Agents compare
	Agent Relay	Agent CLIs	Coder Agents
Where the agent runs	In the vendor's cloud	Inside a Coder workspace	In the Coder control plane
Where the model runs	In the vendor's cloud	Whatever the agent is set up to use	Any provider you configure, including self-hosted models
Where the agent executes	In a Coder workspace	In a Coder workspace	In a Coder workspace
Works fully offline	No	Depends on the agent	Yes, with a self-hosted model
AI provider credentials in dev environments	None for model calls, which happen in the vendor's cloud. A vendor worker in the workspace connects to the session.	The developer's Coder session token through AI Gateway, or a provider key without it	None
Every AI request recorded	No. The vendor handles the AI requests.	Yes, once the agent's traffic goes through AI Gateway	Yes, through AI Gateway automatically
How it's triggered	The Cursor or Claude Code app	A developer in a terminal or IDE	Chat UI or API

One deployment. Every agent.

Run every agent your teams need on infrastructure you control, with one set of controls.