Skip to main content

Agents

The Agents page is where you configure every AI agent that can work on your project. You control what each one reads, which tools it is allowed to call, which model it runs on, and the shared context all of them draw from. It lives at Manage → Agents (/agents) and always acts on whichever project you have selected in the sidebar. There is no cross-project view.

Nothing here is a fixed list baked into the frontend. The catalog of agents, tools, and models comes live from the runtime. Any agent can be given any tool in the pool, and the pool itself grows as you connect integrations or MCP servers.

In this section​

  • Workspace context: the four shared layers every agent reads from (Project, Skills, Knowledge, Payloads), and the "What this agent sees" panel that shows the exact order context gets injected into a prompt.
  • Configuring agents: the Behavior / Tools / Settings tabs on a Specialist or Custom agent, the tool picker, and creating, editing, disabling, or deleting a custom agent.
  • MCP servers: connecting an external Model Context Protocol server to hand its tools to your agents, and how that differs from the MCP tools an integration already contributes.

What's on the page?​

A left rail lists everything you can select; the right pane shows its configuration.

  • Workspace: five entries: Project, Skills, Knowledge, Payloads (four shared doc layers), and MCP (connected external tool servers).
  • Specialists: the built-in agents that ship with the platform.
  • Custom: agents your team has added, plus a New agent button at the bottom of the rail.

Disabled custom or specialist agents show an off badge in the rail; the root orchestrator is always in the list and cannot be turned off.

Where to find it​

Left sidebar → Manage → Agents, or go straight to /agents. The old standalone Docs page has been folded into this one; a link to /docs now redirects here (the Workspace rail replaces it).

When does the page go read-only?​

If you can see the Agents page but cannot manage agents, or the project itself is not in a normal, active state (still provisioning, degraded, or archived), every field stays visible but disabled, and a banner at the top says so explicitly.

Who can do this: Anyone with project access can view everything on this page. Editing workspace files, changing an agent's configuration, and creating or deleting custom agents all require agent management access (agent:manage), and only while the project is active and mutable. See People and access.

How this connects to the rest of the platform​

  • Knowledge: the Knowledge workspace layer is the same reference library covered in the Knowledge guide, including codebase maps Discovery generates automatically.
  • Integrations: tools tied to a not-yet-connected provider (GitHub, Slack, and so on) show up as dormant: selectable, but inert until you connect that integration. See Integrations.
  • Secrets: anything an MCP server or a custom tool needs to authenticate is a reference into your project's Secrets vault, never a pasted value.
  • The Development board: agents you configure here are the same ones that pick up cards on the board; an agent you disable stops accepting new work, but anything already in progress finishes normally.
On this page