One key, several AI searches
POST /v1/ai-search tries providers in cost order and falls over when one is down. Your agent keeps answering — you never wire a second vendor in a hurry.
POST /v1/ai-search returns an answer plus the sources behind it, so your agent can quote instead of guess. Not every provider writes an answer: Tavily, ChatGPT and DeepSeek (synthesized) do; Yuanbao and Exa return cited results only.
Microsoft shut the Bing Search API v7 on 11 August 2025 — HTTP 410, no migration path. Teams wired to one provider had to rebuild on the spot.
Provider auto picks the cheapest provider that can answer; if it fails, the next one is tried. Only a failed call switches provider: if a provider returns citations but no answer, that counts as a result and auto does not move on. Billing follows the same rules as search: cache_first is half price, cache hits, failures and empty results are free.
POST /v1/batch runs up to 5 searches in one request; POST /v1/fetch turns a public page into clean Markdown. One wallet for all of it.
Agent search →Example: answer with citations
curl https://api.serphubapi.com/v1/ai-search \
-H "X-API-Key: shapi_..." \
-H "Content-Type: application/json" \
-d '{"q": "What changed in the latest EU AI Act guidance?", "provider": "auto"}'The response carries the answer and its cited sources — where the provider writes one (see above). Slow providers answer asynchronously: POST returns a job id and GET polls until it is done.
Use with AI agents →Tell us your fallback needs
Source: Bing Search API v7 was shut down by Microsoft on 11 August 2025.
Make your first call in five minutes
1,000 free credits on signup. No credit card.