Build your agent
Steps
The building blocks of a workflow: what each kind of step does and when to use it.
Each step has a type, its settings in parameters, and where to go next in next. Pick a step to see exactly what it does and every setting it takes.
| Step | type | Use it to |
|---|---|---|
| Say | out | Speaks a line you write: greetings, confirmations, goodbyes. Can also speak what an AI decision wrote. |
| Listen | input | Waits until the caller finishes talking, then saves their words for the next step. |
| Filler | random-speak-out | Plays a short “one moment…” so the caller isn’t left in silence while a slow step works. |
| AI reply | llm-streaming | The AI answers out loud, word by word, so the caller hears it sooner. Best for normal chat. |
| AI decision | llm | The AI thinks first and fills in fields (a reply, “end the call”…). Silent: follow it with Say or Branch. |
| Branch | conditional | Sends the call down a different path based on a value, like an AI decision’s yes or no. |
| Guide on website | llm_webnavigation | Talks the caller through your website while scrolling and clicking it for them. (Coming soon — not available on calls yet.) |
Choosing between the AI steps
| AI reply | AI decision | |
|---|---|---|
| Speaks | Yes, as it writes | No: follow it with Say |
| First word | Fastest | After the whole answer |
| Saves its fields for later steps | No | Yes |
| Can end the call | Yes, with hangup | Yes, via a Say that passes hangup |
| Emotion cues | No | Yes |
| Use for | Conversation | Decisions, data, branching |
Both share the settings on AI settings: instructions, models, memory and decided fields.
Next steps
- Say: the step that speaks.
- AI settings: models, memory and decided fields.