> ## Documentation Index
> Fetch the complete documentation index at: https://www.usenotra.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Conversations

> Conversations are multi-turn prompt sequences that show how engines answer once a buyer narrows the question.

A conversation is a multi-turn exchange of up to five turns that captures how an engine answers once a buyer has narrowed the question, for example "what tools automate changelogs" followed by "which of those integrate with linear". A project can hold up to ten conversations.

In the API a conversation is called a **prompt sequence**.

The **Conversations** card on the Prompts page lists each sequence with its **Turns** count and whether it is **Included in scans** or paused. **New conversation** opens a dialog with **Name** and **Turns** fields. Every English scan replays conversations against the grounded engines and shows the per-turn results as a replay thread.

## Running a conversation now

The play button on a row runs that conversation immediately.

```bash theme={"system"}
curl -X POST https://api.usenotra.com/v1/projects/$PROJECT_ID/geo/sequences \
  -H "Authorization: Bearer $NOTRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Changelog tool research",
    "steps": [
      "what tools automate changelogs from github",
      "which of those integrate with linear"
    ]
  }'
```

```bash theme={"system"}
curl -X POST https://api.usenotra.com/v1/projects/$PROJECT_ID/geo/sequences/$SEQUENCE_ID/run \
  -H "Authorization: Bearer $NOTRA_API_KEY" \
  --max-time 330
```

```json theme={"system"}
{
  "checks": 8,
  "mentions": 3,
  "engines": ["openai/gpt-5.4-grounded", "anthropic/claude-sonnet-4.6-grounded"],
  "organization": { "id": "org_123", "slug": "acme", "name": "Acme", "logo": null }
}
```

<ResponseField name="checks" type="integer">
  Recorded answers across every engine that responded.
</ResponseField>

<ResponseField name="mentions" type="integer">
  How many of those answers mentioned the tracked brand.
</ResponseField>

<ResponseField name="engines" type="string[]">
  Engines the conversation was played against.
</ResponseField>

<Warning>
  The run isn't queued, so the request stays open for the whole run, which can take several minutes, and you should use a client timeout of at least five minutes. After four minutes the API stops waiting and answers `409` while the run finishes on its own, so don't retry and read the result from the project's prompt results instead. This endpoint allows 10 requests per 10 minutes per organization.
</Warning>

`PATCH /geo/sequences/{sequenceId}` accepts `name`, `steps` (1 to 5) and `enabled`. `DELETE` removes the sequence.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.