Sessions
The Sessions tab lists every conversation an agent has had. Each one is a session: a call for a Frontline agent, or a run for a Background agent. Use it to see what callers actually asked, check how the agent answered, and find the conversations that went badly so you can fix the agent.
Before You Start
- Role: the Admin or Organisation Admin role.
- A published agent. The Sessions tab is on the agent page, which opens once an agent has been published.
Opening Sessions
- Open Agents.
- Select the agent, then select the Sessions tab.
You can also choose Sessions from the agent's actions menu on the Agents page. The number on the tab is the agent's session count for the last 30 days.
The tab has two parts. The list of sessions is on the left, and the session you select opens on the right. The first session in the list is selected when the tab opens.
Filtering by Type
The switch above the list filters it by session type. See Live and evaluate sessions.
| Filter | Shows |
|---|---|
| All | Every session. |
| Live | Real traffic: callers on the agent's phone number, and systems or integrations using the agent. |
| Evaluate | Test conversations from the playground. In the list they show the IQ mark, a Playground badge and the word "Evaluate". |
Use Live when you want to judge how the agent is doing with real customers, so playground tests do not skew what you see.
Reading the List
Each row shows:
- The caller: the phone number the session came from, or "Unknown" when there is none, as with most runs and playground sessions
- The score: five pips that fill in with the score. A row with empty, faded pips has not been scored yet
- A summary line: the start time, the number of messages, the type (Live or Evaluate) and the score, or "not scored"
A session that is still going shows an in-progress marker instead of pips, and its line ends with "in progress". The list and the open transcript keep updating while a session is in progress, so you can follow a live call as it happens.
Reading a Session
Selecting a session opens it beside the list:
- Header: the caller, the status (In Progress or Complete), the date, the duration and the number of tokens used. For a phone call, the header also shows the Call SID, which is Twilio's own ID for the call. Quote it when you look the call up in your Twilio console or raise it with Twilio support.
- Summary: a short account of what happened in the conversation, with its score out of 10. The summary and score appear once the session has ended.
- Transcript: the full conversation, with the agent's lines under its name. Tool calls appear in the order they happened, each with the tool's name, a Success or Error badge, and the Input the agent sent and the Output it got back. Some rows record events during the session, and Show Data reveals the detail behind them.
Scores
A session is scored out of 10 when it is summarised. The colour gives a quick reading:
| Score | Colour |
|---|---|
| 8 and above | Green |
| 5 to 7.9 | Amber |
| Below 5 | Red |
Not scored means the session has no score yet. Scoring happens a little after a session ends, so a session that has finished in the last few moments usually shows Not scored for a short time.
When a session has no summary, the panel says "No summary available for this session." For an Evaluate session it says the run is only summarised when the session is ended. A text playground run ends when you select New run or leave the Playground tab. See Playground.
Full View
Select Full View to open the session on its own page, with more room for the transcript and more detail about the session.
- Session Summary shows the summary, with a Graded badge once the session has been scored.
- The Session panel lists the Type, Status, Duration, number of Messages, Tokens and, for a phone call, the Call SID.
- The Tool Calls panel counts how many times each tool was called, split into successful and failed calls. A tool with failed calls is the first place to look when a session went wrong.
- The Agent at the Time panel shows the agent with the phone number and model it used for this session. The name is the agent's current name, but the other details are as they were when the session happened, so you can tell which setup produced it even after the agent has changed.
Use Previous Session and Next Session to step through the list you opened the session from, and Copy Transcript to copy the whole conversation as text. Select the back arrow to return to the Sessions tab.
If a session cannot be found, for example when you follow an old link, the page shows Session Not Found with a Back To Sessions button.
Empty States
| What you see | Why |
|---|---|
| No Sessions Yet | Nobody has called, run or tested the agent yet. Try it in the playground. |
| No Live Sessions | No real traffic has reached the agent yet, so any sessions it has are playground tests. A Frontline agent is reached on its phone number or by your own telephony using its agent ID. See Phone Numbers and Deployment. |
| No Evaluate Sessions | Nobody has tested the agent in the playground yet. |
| Unable to load agent sessions | The sessions could not be loaded. Select Retry. |
What Happens Next
- Look for patterns in low scores and failed tool calls, then fix the agent in Quick Config or The Agent Editor.
- Reproduce a problem conversation in the playground against your draft before publishing the fix.
- Check whether scores are improving over time in Metrics.