From the source
Baseten is introducing Baseten Hosted Tools, an ecosystem for capabilities like web search, code execution in sandboxes and browser use, with Web Search as the first category.
- The company says developers building apps and agents with open-source models previously had to assemble one tool at a time, handling the ReAct loop, task coordination, model steering and eager or concurrent dispatch, work that becomes part of the product's critical path.
- Baseten Grounded Inference, the first hosted tool, gives models hosted on Baseten access to information as recent as the current day that is absent from their training data.
- Web search runs inside the Baseten Inference Stack and is enabled by adding the name of a hosted search tool as config to a standard Messages, Chat Completions or Responses request, in streaming and buffered mode.
- The model decides whether a prompt needs search based on its native tool use capability and may run multiple iterations, with a configurable budget ranging from one-shot to deep research agent.
- Baseten is launching with Exa, Keenable, Parallel and You.com; Keenable says its latency is sub-250 ms p95 in US East at $4 per 1,000 requests.





