Getting started

How it works

Every Ohsee call runs the same core pipeline so you get clean, ranked, cited output instead of raw HTML.

The pipeline

Ohsee discovers relevant sources, reads and cleans each page, ranks the most useful passages against your query, and returns structured results with citations.

pipeline

Your request
      |
      v
Discover  ->  find relevant public pages
      |
      v
Read      ->  fetch and clean pages into text/markdown
      |
      v
Rank      ->  score passages against your query
      |
      v
Return    ->  cited, structured, LLM-ready JSON

Source discovery is resilient: if one provider is slow or fails, the others still return results, so a single upstream hiccup does not fail your request.

Search depths

Search lets you trade latency for depth. Deeper levels read more pages and rank more passages.

Depth	Latency	Cost	Behavior
`ultra-fast`	~0.5s	1 credit	Discovery snippets only, no scraping. Fastest path when you just need links.
`fast`	~1s	1 credit	Scrapes a few top results with one snippet each. Good default for chat agents.
`basic`	~1.5s	1 credit	Full extract and rerank pipeline. Balanced relevance and speed (default).
`advanced`	~2-3s	2 credits	Deeper fetch with multiple chunks per source. Highest signal.

Sync vs async

Fast operations respond immediately. Longer operations run as background jobs.

Synchronous: /search, /extract, /structure, /estimate, and /map when limit is 50 or fewer.
Asynchronous (return a job_id you poll): /research, /crawl, /domain, large /map, and monitor runs.

Async jobs reserve credits up front and settle on completion, refunding anything unused. See Jobs & async.

Introduction

Quickstart