Introducing Parallel Decisions API: the first decision engine with world knowledge built in. Learn more.[Learn more.]
Baseten logo

## What you can do

  • - Ground open-source model responses in current web data without client-side orchestration.
  • - Let a hosted model decide when to search and iterate until it is confident in the answer.

## How it works

Baseten Hosted Tools runs web search server-side inside the Baseten Inference Stack, with Parallel as a selectable provider for Baseten Grounded Inference.

Add the Parallel web-search tool to a standard Messages, Chat Completions, or Responses request and Baseten interleaves model inference with Parallel search in a server-side loop, streaming tool-use progress until the model produces a grounded response.

## Set up in three steps

  1. 1. Enable Hosted Tools on a Baseten Model APIs request with the server-tools header.
  2. 2. Add the Parallel web-search tool to the request's tools array and optionally tune tool settings such as the iteration budget.
  3. 3. Send a test request and verify the streamed tool-use blocks and the grounded final response.
Use Parallel on Baseten[Use Parallel on Baseten]

## Related integrations