[Go to site: main page, start]

Skip to main content
For AI agents: a documentation index is available at https://docs.parallel.ai/llms.txt. The full text of all docs is at https://docs.parallel.ai/llms-full.txt. You may also fetch any page as Markdown by appending .md to its URL or sending Accept: text/markdown.

API Overview

The Parallel Ingest API provides endpoints for creating intelligent task runs that can perform web research and data extraction. The API is built around a stateful architecture where task creation and result retrieval are separate operations.

Endpoints

Suggest Task

POST /v1beta/tasks/suggest Generate a task specification based on user intent. This endpoint helps you create properly structured tasks by analyzing your requirements and suggesting appropriate schemas.

Request Parameters

Response Schema

Returns a SuggestedTaskSpec object with the following fields: Warning Types:

Example Request

With previous task iteration:

Example Response

Suggest Processor

POST /v1beta/tasks/suggest-processor Enhance and optimize a task specification by suggesting the most appropriate processor and refining the schemas.

Suggest Processor Request Parameters

Valid values: base, base-fast, core, core-fast, core2x, core2x-fast, pro, pro-fast, ultra, ultra-fast, ultra2x, ultra2x-fast, ultra4x, ultra4x-fast, ultra8x, ultra8x-fast
The lite and lite-fast processors are available for task execution but will never be returned by this endpoint.
See Processors for details on each processor.

Suggest Processor Example Request

Suggest Processor Response Schema

Returns an enhanced task specification with additional fields and optimizations.

Suggest Processor Example Response

How Processor Suggestion Works

The /suggest-processor endpoint analyzes your task specification to recommend the most appropriate processor. The algorithm considers:
  1. Task Complexity - Number of output fields, depth of research required
  2. Research Pattern - Whether the task requires single-step lookups, multi-step reasoning, or parallel breadth-first research
  3. Data Sources - How many disparate sources need to be consulted
  4. Special Tools - Whether the task requires specialized capabilities like entity ranking
The recommendation balances task requirements against processor capabilities, selecting the lowest-cost processor that can reliably complete your task.
The first processor in recommended_processors is always the best recommendation. The API may return multiple processors if several could handle the task, but we recommend using the first one.

Examples

Select Input Columns from a Predefined Set

Sometimes you have a specific dataset with fixed columns and need to create a task that works exclusively with those columns. The previous_task parameter allows you to constrain the API to generate task specifications that match your exact data structure. When to use this approach:
  • You have a fixed dataset schema that cannot be modified
  • You want to ensure the task only uses your specific input columns
  • You need to provide examples that match your exact data format
  • You want to prevent the API from suggesting additional input fields
The workflow:
  1. Define Your Schema: Specify exactly which columns you want to use as inputs with their descriptions
  2. Provide Sample Data: Include examples that match your exact data format
  3. Generate a SuggestedTaskSpec: Use the helper function to create a properly formatted SuggestedTaskSpec object
  4. Refine with API: Pass this as previous_task to get a refined task spec that respects your column constraints
The API will use your predefined input schema as a foundation and refine the output schema while preserving your input columns. This guarantees the final task specification integrates seamlessly with your existing dataset.

End-to-End Ingest to Task Execution

The following Python script demonstrates the complete workflow of the Ingest API, from task suggestion to result retrieval:
Running the Example
This example demonstrates the complete workflow:
  1. Suggest Task: Generate a task specification from natural language intent
  2. Suggest Processor: Get processor recommendations and enhanced schemas
  3. Create Task Run: Submit the task for processing with proper schema formatting
  4. Get Results: Poll for completion and retrieve the final results
The script includes proper error handling, status polling, and demonstrates the correct format for task specifications required by the API.