Building and sharing a voice agent only tells you it can talk. Conversations is where you find out what it actually said: every call it has handled, in one list, so you can check whether it’s answering well, catch a call that went sideways, and see what people are actually asking it.
Why look at this
An agent’s instructions, questions and knowledge base are all decisions you made before anyone ever talked to it. Conversations is the other half: real calls, with real callers, showing whether those decisions held up. It’s also where you go first when someone reports a bad interaction, since every call is logged here whether it went well or not.
The conversations list
Open Conversations in the sidebar to see every call across all of your agents, newest first. Each row shows the agent, the call’s type, who called (when that’s known), how many messages were exchanged, how long the call ran, its status, when it happened, and its cost. A short line of the call’s AI-generated summary appears under the caller when one is available, so you can often tell what a call was about without opening it.
Use the search box to filter by agent name, or narrow the list further with:
- Agent: pick one or more agents, useful once you’re running several.
- Type: Web for calls started from an embedded widget, or Phone for calls that came in over a phone line.
- A date range selector, set to all time by default.
Clear filters resets all of the above at once. If you open Conversations from inside a specific agent instead of the main sidebar entry, the list is already scoped to that agent, without the Agent filter.
What status tells you
Every call is in one of four states, shown as a badge on its row:
- In progress: the call is happening right now.
- Completed: the call finished on its own.
- Ended: the call stopped before it reached a natural finish.
- Transferred: the agent handed the call off partway through.
A call that hasn’t finished processing yet won’t have a status you recognize as final, and its row sits in a muted state until it’s ready to open.
Open a call to read it
Select any row that has finished processing and it opens as a regular Speak media record, the same place a recording or transcription would live in your library. That’s not a separate “conversation page”: once a call completes, its recording and transcript are filed as one of your media items, so it shows up in your normal media library too, and everything you’d expect there is available: the full turn-by-turn transcript, an AI-generated summary, and the rest of the standard insight tabs.
On top of those, a call’s media record has one tab a regular recording doesn’t: Call details. It covers:
- Overview: the call’s type, status, why it ended, how many turns it had, and its duration.
- Usage & latency: which speech, language-model and text-to-speech services handled the call, and how quickly each one responded.
- A note on any handoff, when the agent passed the call to a person partway through.
If you attached Questions to the agent, whatever it collected during the call is saved onto that same media record’s fields, right alongside the transcript.
A row that never got a completed call stays muted and doesn’t open. The transcript and Call details tab both depend on that post-call processing finishing first.
For scripts and other agents
Everything on this page is also available as data: list calls, fetch one by id, and pull its turn-by-turn event log (every transcript turn, tool call, and status change the call went through) over the API. See Read Speak AI voice agent call history for the exact fields and example requests.
Related pages
- Read Speak AI voice agent call history: the API reference for listing calls, fetching one, and reading its event log.
- Questions: what gets asked mid-call and where each answer ends up.
- Analytics: the same call history rolled up into trends and an engagement funnel.