Cloudflare AI Gateway web search in 2026

SpecStream on the same gateway, in two steps

The /ai/websearch/ endpoint takes three providers: ceramic, exa and linkup. AI Gateway also takes any HTTPS API as a custom provider, with its logs, caching and rate limits. Add SpecStream once, under **Compute & AI** > **AI Gateway** > **Custom Providers** > **Add Custom Provider** in the dashboard, or by API:

curl -X POST "https://api.cloudflare.com/client/v4/accounts/$ACCOUNT_ID/ai-gateway/custom-providers" \
  -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"name": "SpecStream", "slug": "specstream", "base_url": "https://api.specstream.ai", "enable": true}'

Then search through the gateway. Everything after custom-specstream/ is added to the base URL, so this call reaches https://api.specstream.ai/v1/web-search and shows in your gateway logs beside your model calls:

curl "https://gateway.ai.cloudflare.com/v1/$ACCOUNT_ID/$GATEWAY_ID/custom-specstream/v1/web-search" \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" \
  -H "cf-aig-authorization: Bearer $CF_AIG_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"q": "Cloudflare Workers changes in November 2026", "speed": "fast", "num": 10}'

The reply is {"data": {...}} with the results in data.sources, and data.price_usd says what the call cost: $0.001 for up to 10 results on the fast search.

The most accurate web search API in the world

On SimpleQA Verified the SpecStream Search API scores 97.9 F-score on all 1,000 official questions, 978 of them correct. On WideSearch it scores 64.0 item F1 and 43.4 row F1 on all 200 official tasks. The Extract API scores 0.908 F1 on WCXB. Every score and its method is on the benchmarks page.

Up to 100 results, and the text of the pages

The gateway's providers stop at 10 results a call. SpecStream returns up to 100 for one search with num: 50 results are $0.0058 and 100 are $0.0118 with speed=fast. Add passages=true and each of the first sources carries an excerpt, the text of its page closest to the question, up to about 3,500 characters, so the model reads the page's own words.

Search, pages and rows on one key

An agent that searches reads pages next. On the same key the Extract API reads up to 20 pages or documents a call at $0.80 per 1,000 pages, and Deep Search turns a request in plain words into structured rows that meet every condition you wrote, with a source URL and read time on every value. The most a run can cost is quoted before it starts, and a saved request reruns hourly to yearly.

You decide what is kept, on each key

Create a key with zero_retention and a search made with it keeps no request text and no answer after the reply. The charge keeps the account, the key, the request ID, the price and the time. spend_cap_cents is the most the key can spend, and eu_egress_only sends its searches out only from our server in Madrid.

curl -X POST https://api.specstream.ai/v1/account/keys \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "ai-gateway", "scopes": ["account:read", "finds:spend"], "spend_cap_cents": 50000, "zero_retention": true}'

SpecStream

The live web data engine for teams and agents, and the most accurate web search API in the world. $1 per 1,000 searches with speed=fast or $5 on the default search, up to 100 results a call, through AI Gateway as a custom provider, over the API or over MCP at https://api.specstream.ai/mcp. New accounts get $10 of credit, no card.

Ceramic.ai

The gateway's default provider, at $0.25 per 1,000 requests, from its own index of more than 40 billion pages, with a description of each page up to 8,000 characters. A call returns 10 results at most. SpecStream reads the live web at that second, returns up to 100 results and answers from the pages.

Linkup

$5.00 per 1,000 requests through the gateway, on fast search depth with raw results and no generated answer. SpecStream answers the same call at $1 per 1,000 with speed=fast, and Linkup code moves with one address: see Linkup alternatives.

Exa

$7.00 per 1,000 requests through the gateway, on Exa's auto search type, marked No for Zero Data Retention. SpecStream costs $1 per 1,000 with speed=fast, and a zero_retention key keeps no request text. See Exa alternatives.