Skip to content

Generate media AI output

Generates one of a video's or audio file's AI outputs, or generates it again, landing as a draft; subtitles first, the rest from them.

Needs the file:write scope.

Call

  • Tool: generate_media_output
  • Scope: file:write
  • Writes, and may change or remove what is there.
  • Acts on a store.

Arguments

Name Type Required Description
store string No The store to act on.
asset object Yes
kind "summary" or "chapters" or "transcript" or "subtitles" Yes
{
  "type": "object",
  "properties": {
    "store": {
      "type": "string",
      "description": "The store to act on."
    },
    "asset": {
      "type": "object",
      "properties": {
        "id": {
          "type": "string",
          "minLength": 1,
          "maxLength": 512
        }
      },
      "required": [
        "id"
      ],
      "additionalProperties": false
    },
    "kind": {
      "type": "string",
      "enum": [
        "summary",
        "chapters",
        "transcript",
        "subtitles"
      ]
    }
  },
  "required": [
    "asset",
    "kind"
  ],
  "additionalProperties": false
}

Answer

Name Type Required Description
kind string Yes
asset_id string Yes
status string Yes
has_content boolean Yes Whether content landed, which a run still generating or one that failed may keep from before.
revision number Yes The file’s revision the content came from, or the running generation reads.
language string or null Yes
reason string or null Yes
bytes number or null Yes The output’s size in bytes; null when it is charged nothing.
deadline string or null Yes
updated_at string or null Yes
published_at string or null Yes
created_at string or null Yes

The answer’s full shape is on its API reference page.

Errors

A refused call is a tool error. Its structuredContent holds a code and its context, except for client-fault. This tool may answer:

  • invalid: an argument is malformed; context.path names it.
  • invalid at ["store"]: store is missing while the connection reaches more than one store, or names a store it doesn’t reach.
  • not_found: nothing matches a reference given.
  • conflict: the call clashes with the current state; context names the field and, where it helps, its current value, and for a from, live, expect or version that no longer matches, the argument (path) and what it takes (reason).
  • not_permitted: this connection may not make this call. When context.reason is limit, too many calls: retry_after says when to try again.
  • internal or unavailable: the call failed inside Sherpo, and the write may have landed. This write takes no idempotency key, so read the state back before trying again.
  • client-fault: the server failed around the call. core_answer ok means the write landed, so don’t send it again. When reached_core is false, nothing happened and the call may be made again. Otherwise, this write takes no idempotency key, so read the state back before trying again.
  • HTTP 403 with insufficient_scope: the connection lacks the scope this tool needs.
  • HTTP 401 with invalid_token: the access token expired or the connection ended. Refresh the token, and connect again if the refresh fails or the new token is refused too.

Any other code may come back too, such as a conflict when the state changed under the call. When the answer was already streaming, a 401 or 403 comes back as a tool error with only its text. See Errors for every code and its context.

Type to search…

↑↓ navigate↵ selectEsc close