Samna

Appearance

Generate an Action ​

POST /ai/action takes a natural language prompt and returns a complete action draft. The draft is ready to post to POST /action to create the live row.

Request ​

json
{
  "prompt": "When a booking is created on a meeting room, send a confirmation email to the booker with the room name, start time and end time."
}

prompt is required. input is optional: pass a pre-defined input schema when an upstream form already knows the input shape and you want the generated script to bind to those exact keys.

Response ​

json
{
  "data": {
    "request_id": "5fb1c0ad-8a02-4e2a-9aa5-2d51c4f8bba4",
    "action": {
      "name": "Send booking confirmation email",
      "description": "Sends an email to the user who created the booking with room and time details.",
      "summary": "Loads the booking, formats the start and end time, and emails the booker.",
      "trigger": "booking_created",
      "code": "/* generated tengo */",
      "input": {
        "email_from": { "type": "string", "description": "Sender address" }
      },
      "public": false,
      "continue_on_fail": false,
      "dedup_mode": "per_attachment"
    }
  },
  "request_id": "fa1de5d0-e9f0-46fb-9aa1-32b6b8b9a2a4",
  "count": 1
}

data.request_id points at the ai_request row that holds the prompt, model, and audit trail for the generation. The outer request_id is the standard response envelope id.

Saving the Draft ​

Send the action block directly to POST /action. The draft is discarded if not saved. The ai_request row stays for the audit trail regardless.

Following Progress ​

Generation runs inline, so the response body arrives only when the model finishes. To observe progress while a run is in flight, open a second connection to GET /ai/request/{id} or GET /ai/request/{id}/response with Accept: text/event-stream. The server streams status, progress, and messages until the run settles.

Refining the Code ​

If the structure of the draft is right but the script needs changes, hand the code and summary back through Generate Action Code along with a new instruction.