What are LLM Tools?
LLM tools run structured extraction against the current step’s conversation. The assistant reads what the caller has said and returns clean, typed data you can store in variables and route on. There are two LLM tools:| Tool | Use when you need |
|---|---|
| LLM Data | One typed value — a name, email, number, date, appointment time, reference number, or category |
| LLM Custom | Several flat fields from the same answer, extracted in one call |
When extraction passes
All LLM extraction only passes when the caller is directly answering the assistant’s current-step question, or affirming the assistant’s immediately preceding confirmation question. Counter-questions, subject changes, deflections, unrelated answers, and vague answers returnfalse.
Output
Both tools expose a singleresult boolean.
| Value | Meaning |
|---|---|
true | A valid value was extracted, and any configured writes succeeded |
false | Required inputs were missing, schema validation failed, no conversation was available, or extraction failed |
result boolean for tool-step conditions, and a destination/response map to store the extracted values.
Choosing between them
- Need just one field? Use LLM Data — it offers typed extraction modes (text, number, date, date & time, enum).
- Need a small group of related fields, like contact or claim details? Use LLM Custom with a flat schema.
Next Steps
- LLM Data — Extract a single typed value
- LLM Custom — Extract several flat fields in one call
- Tool Steps — Learn how to configure tool steps
- Variables — See all available variables