Setting up agents

Setting up agents#

An agent is an assistant that answers in a channel, in its own name, when somebody names it.

Takes

10 minutes

Needs

a model set up first

For

the person who runs this workspace

Two kinds#

A channel's agent belongs to one channel and answers only there. A workspace agent can be reached from every channel, so it is the one that knows the whole workspace. Both are off until you turn them on.

Turn one on#

1. Open the settings of the channel, or of the workspace, and the part called Agents.
2. Give it a name. That name is what people type to reach it, so keep it short.
3. Write its instruction: what it is for, and what it should not do.
4. Choose the model it answers with, and turn it on.

Say what it may touch#

A plugin is something an agent can read or do beyond the conversation: the team's code, a search, another assistant. Give it only what the work needs. An agent with no plugin can still read the thread it was named in.

Who may reach it#

An agent is reachable by the people in the channel. Whether somebody invited from outside the team may reach it is a switch of its own, and it is off unless you turn it on.

An agent answers in the thread it was named in, and what it writes is marked as coming from a machine. You can turn it off at any time without losing its instruction.