Using Quetz
Agents
Because an inbox is just an address, software can hold one too. An agent joins a conversation on the same terms a person does.
What an agent is#
An agent is a program with its own address and its own inbox. It is a member of the thread rather than an integration bolted onto the side, which means it reads what is addressed to it and nothing else, exactly like any other member.
Talking to one#
- Message its address the way you would message a person, or open the agent’s Quetz link if it has published one.
- Add it to a room when you create one, and it takes part in the conversation there.
- Remove it like any other member when you are done with it.
What people use them for#
| Agent | What it does |
|---|---|
| Price alerts | Watches a level and messages you when it breaks. |
| Trading desk | Answers quote requests inside an existing thread. |
| Ops runbook | Reports failures where the team already talks. |
| Room host | Answers questions inside a gated room. |
An agent is a member, so treat it like one
Whoever runs an agent can read everything it can read. Adding one to a room adds a participant, not a passive tool. Only add agents you actually trust, and remove them when they are no longer needed.
A gated room applies to agents too
In a room that requires a $QUETZ balance, an agent has to meet the same requirement as everyone else. Either fund its address or leave the room ungated.
Building your own#
Agents are built on XMTP rather than on Quetz, so anything you write works in every XMTP client, not only this one. The protocol documentation walks through it at docs.xmtp.org. Once your agent is running, share its link and people can start a conversation with it in one click.