Agents
Creating an agent, writing its instructions, giving it skills, knowledge, tools and entities, and publishing it.
An agent is the AI persona that joins meetings and chats. Everything else in the Configure group exists to be attached to one.
The agents list shows every agent in the organization, filterable by name, each marked Live or Draft. Only live agents attached to the right entity can be picked for a session, which is what makes draft a safe place to work.
Creating one
The create form asks for:
- A name and a description.
- An avatar. If you do not upload one, Auron generates a distinct image for the agent so it is recognisable in a list.
- A model, either one of the platform models by provider, or your own model configuration from Secrets.
- A voice, used when the agent speaks in meetings.
- Wake words, which are the words that get its attention while it is listening.
- Labels, and any custom fields your organization has defined.
The agent's tabs
Once created, an agent has five tabs.
| Tab | What it holds |
|---|---|
| Activities | Every meeting, chat and call this agent has run, with a calendar view and a New button |
| Canvas | The agent's instructions |
| Configuration | Everything else about how it behaves and what it can reach |
| Webhooks | Delivery of this agent's events to your systems |
| Signals | The signal definitions this agent looks for |
Canvas
The canvas is where the agent's instructions are written, as a document rather than a form field. It is the single biggest influence on how the agent behaves, and it is worth writing at length.
A variable picker inserts your organization's variables into the text, so an instruction can reference a value that is maintained once in Secrets rather than pasted into every agent.
Users with the user role can read the canvas but not edit it.
Configuration
The configuration tab is grouped into sections:
General. Name, description, avatar and labels.
Model. Either a platform model, chosen by provider and model, or one of your own model configurations.
Voice. The voice the agent uses when it speaks in a meeting.
Wake words. What gets its attention when it is listening rather than talking.
Initial meeting mode. Which of the four modes a meeting starts in. Listen keeps the agent silent, conversation answers when asked, discovery probes actively, and recap synthesises. Whoever is in the meeting can change it mid-session.
Calling capabilities. Whether the agent takes inbound calls, makes outbound calls, or both.
Teams. Which teams can use this agent.
Skills. The packaged behaviours it has. See Skills.
Toolkits. The systems it can call out to. See Toolkits.
Entities. Which kinds of record it can be used on. This is the setting that decides where the agent appears: a session can only use an agent attached to that record's entity type.
Knowledge stores. What it grounds its answers in. See Knowledge stores.
Phone number. A number in international format, with its identifier and a friendly name, assigned to this agent so its calls arrive here.
Signals
The signals tab attaches signal definitions to the agent, which is what determines what it looks for in a conversation. An agent with no signals attached will still hold a conversation and produce a transcript; it just will not detect anything. See Signals.
Publishing
An agent stays a draft until it is published. Publishing is what makes it selectable for sessions, including on the phone. Unpublishing takes it out of circulation without deleting anything.
Deleting
Agents can be deleted from the list. Their past sessions are what carry the value, so consider unpublishing before deleting.