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

# Prepare a run

> Send a raw transcript and get back a parked run, its speakers, and a guess at your agent.

```http theme={null}
POST https://api.humalike.com/v1/social-observability/actions/audit_prepare
```

Send a conversation as raw text. The API parses it into messages, stores it as
a **run**, and returns the speakers it found plus a guess at which one is your
agent. Nothing is audited yet — the run is parked until
[`audit_launch`](/api-reference/audit/launch) confirms the agent.

Send the text exactly as you have it. Timestamps, usernames, join/leave notices,
and platform decoration are all handled; you do not need to strip or reformat
anything.

## Authorization

<ParamField header="Authorization" type="string" required>
  Your bearer token: `Bearer <token>`. See [Authentication](/authentication).
</ParamField>

## Request body

<ParamField body="raw_text" type="string" required>
  The conversation, as text. 1–300,000 characters. Longer than that is rejected
  with `422` before any charge.
</ParamField>

## Request

```bash theme={null}
curl https://api.humalike.com/v1/social-observability/actions/audit_prepare \
  -H "Authorization: Bearer $HUMALIKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "raw_text": "[10:02] casey: the export is broken again\n[10:03] support_bot: Have you tried clearing your cache?\n[10:04] casey: nevermind, i will do it by hand\n"
  }'
```

## Response

```json theme={null}
{
  "run_id": "9f8b1c04-6d2a-4f31-9c77-2b1d0e5a8c43",
  "messages": 3,
  "participants": ["casey", "support_bot"],
  "agent_guess": "support_bot"
}
```

<ResponseField name="run_id" type="string">
  The run's id. Pass it to [`audit_launch`](/api-reference/audit/launch) and to
  [`audit-run`](/api-reference/audit/run).
</ResponseField>

<ResponseField name="messages" type="integer">
  How many messages were parsed out of your text. The whole transcript is
  stored — nothing is dropped.
</ResponseField>

<ResponseField name="participants" type="string[]">
  Every speaker found, in the order they first speak. These are exactly the
  values `audit_launch` accepts as `agent_name`.
</ResponseField>

<ResponseField name="agent_guess" type="string | null">
  Which speaker looks like an AI agent. Use it to pre-select an option; do not
  treat it as the answer. `null` when the transcript gives no clear signal,
  which is a normal result and not an error.
</ResponseField>

## Errors

| Status | Code               | Cause                                                                                                           |
| ------ | ------------------ | --------------------------------------------------------------------------------------------------------------- |
| `400`  | `VALIDATION_ERROR` | The transcript has more than 250 messages, the text is too large to read, or no messages could be read from it. |
| `401`  | `UNAUTHORIZED`     | Missing or invalid bearer token.                                                                                |
| `402`  | `PAYMENT_REQUIRED` | Your credit balance can't cover the call.                                                                       |
| `422`  | `VALIDATION_ERROR` | `raw_text` is empty or over 300,000 characters.                                                                 |

The over-cap message names both numbers so you can act on it:

```json theme={null}
{
  "error": {
    "code": "VALIDATION_ERROR",
    "message": "This transcript has 5,319 messages; the audit accepts at most 250."
  }
}
```

Send the most recent 250 messages and call again. See
[Size limits](/api-reference/audit/overview#size-limits).

Next: [Launch a run](/api-reference/audit/launch).