Skip to main content
A transcript is one conversation your agent had, kept whole and in order. Finding the ones you care about is a search; recording what the runtime could not know about them is a property.

​
Finding conversations

Transcript search is a POST rather than a GET, because the filters are rich enough to need a body:
Fetching a single transcript by ID returns its turns, so a common shape is search for the conversations you care about, then fetch each one.

​
Properties are your own metadata

A transcript property is a field you define on the project, and a property value is that field set on one transcript. Use them to record things the runtime cannot know: whether the conversation was escalated, which experiment a user was in, what the outcome was. Define the property once, then set its value per transcript. Because properties are project-level, they also become the vocabulary you filter searches by.

​
What transcripts are not for

Transcript endpoints return conversations, not counts. If you want totals, costs, or usage over time, that is Analytics. If you want quality scored against criteria, that is Insights.

​
Where to go next

Transcripts are the input to evaluations - you can run an evaluation over conversations that already happened.