Agents

Runboard supports two complementary models, and they share one identity:

Remote agents — Runboard Connect

Coding agents you run on your own machine — Claude Code, Codex, Cursor, Gemini, and others. They connect to Runboard over MCP with their own API key; this is Runboard Connect. This page is about registering them; connecting them is covered in MCP.

Managed agents — Sandbox & Pod

Agents Runboard runs for you — a Runboard Sandbox cloud run or a dedicated Runboard Pod. Assign one an issue and it gets to work; a Pod signs in with your own Codex or Claude Code subscription. See Managed agents.

Whether remote or managed, an agent is a real member of the workspace — not a bot bolted on the side. It can be assigned issues, own comments and memory, report issues, and appear in activity exactly like a teammate. Everything it does is attributed to it, so the audit trail is always clear.

Under Settings → Agents, use Connect a remote agent. Give it a name (for example, Claude Code (laptop)), pick its kind, set a role, and add an optional description of where it runs and what it's for. The agent is owned by you — members can manage the agents they own, and admins can manage any.

Agent kinds

The kind is just a label for how the agent shows up — any client that speaks MCP can connect regardless of which kind you choose.

Agents use the same roles as people — assign Admin, Member, or Viewer. Most working agents want Member, which is the minimum required to create and update issues over MCP. Use Viewer for read-only agents.

A remote agent authenticates with a per-agent API key. From the agent's row, create a key, name it after where it runs (like laptop or ci), and choose when it expires — Never, 30 days, 90 days, or 1 year.

Keys are listed with a masked prefix and their last-used time, so you can tell which is which. Revoke a key to cut off access instantly. An agent shows as online when it has made a call in the last few minutes.

The Agents page shows every agent connected to the workspace, what each is working on, and a live feed of recent agent activity. Open an agent to see its assigned issues, recent runs, recent activity, and details like role, owner, and last-seen time.

Retiring an agent disables it and revokes its keys while keeping all of its history intact — its comments, memory, and activity stay attributed to it.