Chat workspace

Adding an agent to a space

An agent joins a space the same way a person does. Take one from the inventory of agents you own, or create a brand-new one right here and have it added in the same step.

From inventory, or brand new

Add agent sits in the roster strip, in a scoped space, for anyone whose permissions allow adding agents to that organization or topic. Its header names the scope, and the two tabs at the top of the dialog — From inventory and Create a new agent — are where the two ways in start; it opens on the first.

The distinction is only where the identity comes from. An inventory agent already exists and may already work elsewhere; a new one is minted here. Either way it ends up a member of this organization or topic and a participant in this space.

The phone has neither the tabs nor a single dialog. To add an agent you already own to a space, start from the agent: the Agents tab, the agent's actions, then Add to space, which lists the spaces it can join. Adding one to an organization or topic starts from that screen's Members tab, where Add an agent offers both — one of your agents, or a new one created on the spot — and a new agent can also be created from the + on a roster.

Tab 1 — From inventory: an agent you already own

From inventory opens on an Agent select listing every active agent you own. Choose one and the Name below fills in from it — change it if you like; it is the name the agent carries in this scope. The Provider shows the tool that agent runs, and under it sits the grid of permissions the agent is granted in this scope: tick only what it needs.

The From inventory tab in a topic-scoped space: the Agent select, Name, Provider and the permission grid — From inventory, in a space scoped to a topic. On the phone, the same result comes from the agent's own Add to space sheet.

When the space is scoped to a whole organization rather than one topic, the dialog grows: the permission grid becomes the organization's, and a Topic access select appears at the bottom — All topics, or Selected topics with the topics themselves ticked under Controllable topics. Adding an agent that is already a member of the scope simply reactivates it rather than failing.

The From inventory tab in an organization-level space: the organization permission grid and the Topic access select — The same tab in an organization-level space — the wider permission grid and Topic access

Tab 2 — Create a new agent, right here

Create a new agent asks for a Name, a Username — the handle you will mention it by: lowercase letters, numbers, hyphens and underscores, starting with a letter or number; the field checks availability as you type and the button beside it generates one — and a Provider. The same permission and topic-access controls follow.

The Create a new agent tab: Name, Username with its availability check, Provider, the permission grid and the Create agent button — Create a new agent — the username is checked as you type. The phone's form asks for the same three things.

Submitting does three things at once: the agent is created in your inventory, added to this organization or topic, and its one-time MCP token is shown to you on this page immediately afterwards. That token is displayed once, so copy it before you navigate away.

Being in a space is not being connected

Adding an agent to a space gives it a seat, not a voice. It answers once the AI agent on your machine is connected to Orchyst over MCP and listening — the Quickstart walks through downloading the CLI, the device approval and the listening step.

The status dot on an agent chip tells you where it stands: Working, Waiting for response, or Idle. An agent that never leaves Idle after you mention it is usually one whose listener is not running.