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

# Errors and retries

> Interpret API errors and avoid duplicate actions when a request fails.

Check the HTTP status and response body together. Most errors include a `detail` field; it can be a string, a list of field errors, or a structured validation result.

```json theme={null}
{
  "detail": "Invalid Labs credential"
}
```

## HTTP status codes

| Status                     | Next action                                                         |
| -------------------------- | ------------------------------------------------------------------- |
| `400`                      | Check the request values and required configuration                 |
| `401`                      | Check the Labs credential and its expiry                            |
| `402`                      | Review the workspace's available usage or quota                     |
| `403`                      | Check workspace membership, permissions, and any required Labs role |
| `404`                      | Check the identifier, active status, and published version          |
| `409`                      | Resolve the reported state or identifier conflict                   |
| `422`                      | Correct the fields or workflow errors in the response               |
| `429`                      | Reduce request frequency and honor `Retry-After` if present         |
| `500`, `502`, `503`, `504` | Check whether the operation completed before retrying               |

## Workflow validation

The validation endpoint returns `is_valid` and `errors`. Publishing an invalid definition can return the same result inside `detail`.

```json theme={null}
{
  "detail": {
    "is_valid": false,
    "errors": [
      {
        "kind": "node",
        "id": "start-1",
        "field": "data.prompt",
        "message": "Prompt cannot be empty"
      }
    ]
  }
}
```

This illustrates the response shape; error messages depend on the failed check. Use `kind`, `id`, and `field` to locate the item to correct.

## Retry carefully

Retry read-only requests with bounded exponential backoff after a transient error. Do not blindly retry a request that starts a call, starts a campaign, creates a resource, or invokes an action: it might have succeeded before the response was lost. Check recent runs or the resource state first.

Record the status, time, target agent, and any returned run ID in your own application logs. Keep keys and customer content out of diagnostic logs.