Clean Markdown
Each result carries the readable content of the page as Markdown, headings and lists intact, with the navigation and cookie banners gone. Roughly a tenth of the tokens of raw HTML.
Search API · built for agents
One request returns ranked, deduplicated results from the live web, delivered as LLM-ready Markdown with citations. Built to be called by an agent, not read by a person.
"EU AI Act" · advanced · 2.47s
Results come from more than one index, merged and deduplicated by canonical URL, and ordered by agreement rather than by any single index's opinion. What comes back is a shortlist an agent can act on, with the content of each result already included.
Every response reports its own coverage. A result whose content could not be retrieved is marked rather than dropped, so an empty result array only ever means the web had nothing to say.
A consumer search engine is built to hand a person ten blue links. An agent needs the opposite: fewer results, already read, small enough to fit in a prompt, and ranked against the question rather than against a page.
That is the whole product. Ask once, get the part your model actually needs.
curl https://api.trawlia.co/v1/search \
-H "authorization: Bearer twl_..." \
-H "content-type: application/json" \
-d '{
"query": "what changed in the EU AI act",
"search_depth": "advanced",
"include_answer": true
}' Each result carries the readable content of the page as Markdown, headings and lists intact, with the navigation and cookie banners gone. Roughly a tenth of the tokens of raw HTML.
Results are fused by rank, so a page two sources agree on rises above any single source's favourite. Passages are then scored against the question rather than against the page.
Ask for an answer and you get one, with citations validated against the sources actually supplied. A hallucinated reference is dropped before it reaches you.
Free tier, no card. The request shape mirrors what your agent code probably already speaks, so trying it is a base URL change.