Skip to main content
Every conversation your agent handles — test sessions, real inbound traffic, outbound campaigns — is stored and browsable. The Conversations drawer is where you go to see what’s happening on a live agent, find the cause of a conversation that went wrong, or audit historical activity.

​
Opening the drawer

In the rail on the left of the agent’s canvas, click Conversations. The drawer opens on the right.

​
The list

Each card is one conversation:
  • Title — a short summary written by the platform (e.g. “Inquiry regarding SSO support in Acme Pro vs. Enterprise”). Until there is one, the card shows who the conversation is with.
  • Participant — who’s on the other end (a name, an email, a phone number, or a stand-in such as Anonymous (a1b2c3)), under the title.
  • Channel — the icon on the left: widget, email, Slack, phone and so on.
  • Goal — what the person wanted, or else their last message.
  • Status — In progress, Ended, Failed, Cancelled or Queued.
  • Test — the conversation came from the Test drawer.
  • Outcome — when it’s over, a short note on how it ended, for example Visitor left, Caller hung up, No reply, timed out, Reply not delivered or Couldn’t start.
  • Duration and time — how long it ran (from the start to the last message) and when it last moved. Hover for the message count.
The list refreshes on its own every 15 seconds while it’s on screen, so new conversations and status changes show up without reloading. Three menus above the list narrow it down: Clear resets all three. Search conversations… matches the participant and the platform’s summary of the conversation. Some filters are applied to the most recent conversations first, and the list loads older ones as needed. If nothing matches yet, it says No matches in the latest N conversations.

​
The transcript

Click a card to open the conversation. Back returns to the list where you left it. The header shows who the conversation is with, its channel, status, Test tag, outcome and duration. If the conversation started with Initial Parameters, a Parameters block under the header lists each value and where it came from — Start (the Start Conversation form or a workflow), Embed code or Page URL (the website widget). Values from the embed code or the page URL are marked, because visitors can change them. Next comes the platform’s summary of the conversation: the person’s goal, Key facts and Open questions. What you see per turn:
  • The person’s message on the right, in the accent colour, as in the Test drawer.
  • The agent’s reply on the left, with the name of the actor that answered. Click the name to show that actor on the canvas: the drawer closes and the actor’s panel opens.
  • Why? under a reply, when the platform recorded a reason for it. See Why a reply happened.
  • Tool calls — a pill per call with the tool’s name and how long it took. A failed call is red. Click a pill to see the arguments, the result and any error.
  • Hook runs — centred rows; click one for its details.
  • Delivery status — for replies sent on a channel, a check when it was delivered, or Delivery failed with the reason.
While the conversation is in progress, the transcript updates on its own.

​
Why a reply happened

Click Why? under a reply to see what the platform wrote down about it:
  • The router’s reason, when it answered directly, asked a question, or handed the conversation to an actor.
  • An actor finishing its task, with its reason.
  • A reply blocked by the safety check.
  • A turn that failed, with the error.
A saved conversation doesn’t keep everything the Test drawer shows. Route conditions, hidden routes, slots saved and AI Condition results appear only in the Test drawer’s trace. To see them, replay the conversation in Test.

​
Actions

​
Common things you’ll do here

Filter to Real only and Failed, or search for the person. Open the conversation, read the turns, and open Why? under the replies that look wrong.Look for:
  • Handoffs that picked the wrong actor (fix activation prompts).
  • Tool calls with wrong arguments (open the pill; fix actor instructions).
  • An actor that finished too early (fix its Definition of Done).
Click the actor’s name to jump to it on the canvas and fix it there.
For email, Slack and Teams channels, each reply has a delivery status. If it says Delivery failed, the integration connection probably hit an error — check the connection’s status.
Some conversations linger — the user never replied, but it’s still In progress. Open it and click Stop.
Found a real conversation that exposed a routing bug? Click Replay in Test. Send the first message, compare, edit the prompts, and retest. Deploy when it’s fixed.

​
Privacy and retention

  • Conversations are stored per workspace and visible to workspace members.
  • Phone calls are stored as text transcripts only — no audio is retained.
  • Test sessions live alongside real ones with the Test tag. They’re never delivered to real channels.

​
Next

Test an agent

Replay a problem case in the sandbox.

Deploy an agent

Take edits live after debugging.