Skip to main content

Overview

A post-conversation agent runs after a conversation completes and acts on what was said. Give it instructions like “If the caller mentions a bug, create a record in the Bugs object with a short title and the quote” and it does that on every matching call, email thread or chat, with no one in the loop. It’s a different thing from a chat agent, which is a persona you talk to in the Stella chat.

How it works

  • It runs when a conversation completes: someone marks it done, or Custral finishes analysing it and nothing is left to act on (no suggested actions waiting, no required scorecard review).
  • Once per conversation. Each agent runs at most once on a conversation, however many times it completes.
  • Only on real content. A conversation with nothing said in it (a silent call, a chat holding only a system notice) is skipped.
  • It reads the whole conversation. The transcript or thread is its context, along with your instructions.
  • It acts without asking, within a fixed set of tools. Nobody is there to approve a step, so its changes go through without the approval card Stella shows in chat. It can look things up in the workspace, create and update records, create, update and complete tasks, and update notes. It can’t send email or texts, enroll anyone in a sequence, delete or archive anything, change settings or the schema, or browse the web. Write instructions that say exactly what it may create or change, and when to do nothing.

Create one

1

Open the tab

Go to Settings → AI and automations → Agents → Post-conversation Agents and click New agent. Creating and changing agents needs the Admin role.
2

Name it and write the instructions

Fill in Name, and optionally Description and Avatar emoji. In Instructions, say what to do and when. Generate with AI drafts them from a sentence, or from the name and description if you leave it blank.
3

Choose when it runs

Leave Run after conversation completes on. Restrict to channel types limits it to Calls, Email threads, SMS, Web chat or Slack; leave it empty to run on every conversation type.
4

Pick a model and save

Model defaults to Auto (follow workspace default). Leave Enabled on and click Create agent.
The list shows each agent’s Triggers (On completed, Manual), Channels and Status. Click a row to edit it. Delete agent is in the row’s menu.

Run one by hand

Open a conversation, click its ⋯ menu, and pick the agent under Run agent. Every enabled agent is listed; a Manual only agent carries a Manual only tag. Running one needs edit access to the workspace. A toast says when it’s done: what it did, that it found nothing to do, or why it couldn’t finish. A run can take up to a minute. If it takes longer, the toast says it’s still running, and its result appears on the agent’s Runs tab.

See what it did

An agent’s Runs tab lists its runs, newest first: when it ran, whether by hand or after the conversation ended, the conversation’s name, a one-line outcome, and a status. Click a run to open its conversation. A conversation you can’t open, or one that was deleted, reads “Conversation unavailable”.

Troubleshooting

When reporting an issue, include the request id shown in the error toast. It links the in-app error to the server logs.