​
Overview

The Task Ingestion Beta endpoints enable customers or AI agents to become automatic power users of the Task API. It simplifies the process of creating tasks by automatically generating the appropriate prompt and schemas based on your intent.

These endpoints are available to beta customers only; please contact our team if interested in trying them out.

​
Suggest Task

POST /beta/tasks/suggest

Turn a user’s intent into a complete Task best fit for the Task API.

  • Input: Natural language intent. For example “Given a company website and name, give me the date and investor of their most recent funding round”
  • Output: Task Prompt, Input Schema, Output Schema

Intent should be declarative and concise. It should describe what the user has as possible inputs and what they would like as possible outputs.

Users can also refine their tasks using this endpoint. They can change the input fields, output fields, and include natural language feedback in update_instructions. For example, a user might add cto_name as an output field, and give feedback to the system that “I also want the name fo the company’s cto”. The Task can then be updated using the same endpoint.

Example request for creating a new task:

{
  "user_intent": "Given a company website and name, give me the date and investor of their most recent funding round",
  "input_schema": null,
  "output_schema": null
}

Example request for refining an existing task:

{
  "user_intent": "",
  "update_request": {
    "previous_task": {
      "prompt": "...",
      "input_schema": {},
      "output_schema": {}
    },
    "input_schema_updates": {
      "added_fields": [
        {
          "name": "new_field",
          "description": "Description of the new field",
          "required": true
        }
      ]
    },
    "update_instruction": "Please add a field to capture this additional information"
  }
}

The endpoint returns a complete task specification:

{
  "prompt": "Given a company's website and name...",
  "input_schema": {
    "type": "object",
    "properties": {
      "company_name": {
        "type": "string",
        "description": "Name of the company"
      },
      "website": {
        "type": "string",
        "description": "Company's website URL"
      }
    },
    "required": ["company_name", "website"]
  },
  "output_schema": {
    "type": "object",
    "properties": {
      "funding_date": {
        "type": "string",
        "description": "Date of most recent funding round"
      },
      "investor": {
        "type": "string",
        "description": "Lead investor in the most recent round"
      }
    },
    "required": ["funding_date", "investor"]
  }
}

​
Save Task and Select Processor

Save the task and automatically select the best processor out of a list of processors.

  • Input: All processors to choose from , final TaskSpec (input, output, prompt, Task ID) — taken from the /suggest endpoint.
  • Output: Task ID, best fit processor name
POST /beta/tasks/suggest/save

Example request:

{
  "task_spec": {},
  "choose_processors_from": ["neon", "argon"]
}

Example response:

{
  "task_id": "task_uuid",
  "processor": "neon"
}
The returned processor is just a suggestion. Future calls to runs with the task ID need to specify the processor. For more information, see Runs

​
Streaming Ingest

This provides a streaming response for ingest.

Inputs:

  • user_intent [required]: user intent in natural language
  • input_schema_keys [optional]: the keys of the inputs that the user already has
  • output_schema_keys [optional]: the keys of the outputs that the user wants
  • metadata [optional]: if there are any additional fields that the user wants to add to the task

Outputs:

  • streaming prompt
  • well-formatted valid json input and output schema
POST /beta/tasks/suggest/stream

Example request:

curl -N -X POST "https://api.parallel.ai/beta/tasks/suggest/stream" \
  -H "Content-Type: application/json" \
  -H "x-api-key: $API_KEY" \
  -d '{
    "user_intent": "Given the linkedin URL of an individual, provide information on their recent thought leadership."
  }'