Writing · Feature
Web search that ends in documents, not links
A search API returns links. For a pipeline that ends in documents — RAG ingestion, research corpora, monitoring — links are the starting line, not the finish. So snakereach now searches the web, and every result feeds the same extraction pipeline as everything else: follow one hit into a scrape, or send all five to batch and compare them.
Try it in the playground — the query is prefilled, and it waits for your click. That last part is deliberate; more on why below.
What it returns
Ranked hits with titles and snippets, plus which provider answered and a trace with a single search stage. The shape is the same everywhere it appears:
snakereach find "underwater farming" -n 3
curl $SNAKEREACH_API_URL/v1/search \
-H "X-API-Key: $SNAKEREACH_API_KEY" \
-H "content-type: application/json" \
-d '{"query": "underwater farming", "count": 3}'Snippets are the provider's words, not extracted content. The honest version of that sentence: never trust a snippet past the shortlist. The document behind each hit is one scrape away, which is the whole point of putting search and extraction in one product.
Why the demo waits for a click
Search resells upstream quota — ours comes from Brave, whose free tier is about a thousand queries a month. Shared links for cheap verbs auto-run when opened; a search link only prefills, because every uncached run spends real budget and opening a link should never spend someone else's.
The endpoint is guarded accordingly: a strict per-IP hourly limit, a monthly budget ceiling, and an identical-query cache in front of the provider, so popular queries cost one upstream call no matter how often they are asked. Full numbers in rate limits, and the endpoint reference in POST /v1/search.
Search is discovery
This completes a shape the product has had for a while: discover a page's links, map a site's sitemaps, search the web, crawl from a seed — four ways to find URLs, one pipeline to turn them into documents. The query was the missing verb. It is not missing anymore.
snakereach turns any URL into the shape described here: markdown first, structure preserved, with per-stage timings. Try it on a page of your own, or read what comes back.