Skip to main content

For AI agents: docs.octen.ai/capabilities/pulse-search.md

Agents stop thinking to search. Each search pauses the model for hundreds of milliseconds, and the more an agent searches, the longer it waits. Pulse Search turns that pause into a pulse. A generative retrieval model generates the most likely URLs for a query in a single inference. In lite mode, a search takes about 10ms, close to the time it takes to generate one token, so an agent can search mid-thought instead of waiting. For the full list of parameters, see the Pulse Search API reference.
  • Faster. Server-side P50 latency of 10ms in lite mode and 23ms in standard mode.
  • Cheaper. From $0.1 per 1,000 calls.
  • Lighter. The model is the index: one inference replaces query understanding, recall, and ranking.
  • Same format as Web Search. Authentication, response structure, and error codes all match.
Built for everyday web knowledge: docs, official sites, help centers, and popular destinations. For news, fresh pages, or filtered queries, use Web Search instead.

​
How It Works

  1. Send a query.
  2. The model generates up to 10 of the most likely URLs, in ranked order.
  3. Lite mode returns each URL with its title. Standard mode adds highlights, full content, and images.

​
Scenarios

​
Docs for coding agents

Pull the official docs or API reference a coding agent needs without breaking its flow.

​
Answers for voice assistants

Find the help page behind a spoken question fast enough to keep the conversation going.

​
Destinations for browsers

Suggest the destination page as the user types. Lite mode returns URLs and titles only.

​
Get the original text

When a model needs the page itself, turn on full content in standard mode.
The full content of the first result per search is free; only additional results are billed.

​
Next Steps

Try in Console

Run Pulse Search live in the Octen console.

Pulse Search API Reference

Full request/response schema.

Last modified on October 27, 2026