Better results in. Better answers out.

Kagi is a meta-search API for AI agents. The same search quality 75,000+ people pay for, even though free search exists. Every query is proxied through Kagi, so upstream providers never see who’s asking.

Get an API key Try the playground

Ranked #1

of 7 search APIs at
finding the right source
35% fewer tokens

per correct
answer
12 sources

per
query

The Test

Highest recall of
any provider tested

What matters for agents is whether the page with the answer made it into the results.
Kagi ranked best among seven search engines tested.

How we tested

Same model, same prompt, only the engine is swapped. Kagi showed the highest share of queries with the answer that matters.

How Kagi Search API works

Each query fans out to 12 sources in parallel, proxied through Kagi. We merge the results and rank them in one pass so the best source rises to the top.

URL recall

Kagi 41%
Exa 38%
Bing 38%
Brave 33%
Parallel 32%
Yandex 29%
Serper 28%

The Problem

Cheap searches get expensive when they’re wrong

Better search means fewer retries, fewer tokens, and answers you can trust.

Every search engine has blind spots

We tested 7 search APIs on 6,789 questions. Each one found answers the others missed, 422 in total. If your API’s index misses a page, your agent never sees it and has to search again.

A search that misses isn’t cheap

Agents often search several times to find one answer. Kagi uses 35% fewer tokens per correct answer than the most accurate API we tested. Those savings add up with every search.

The web is filling up with low-quality AI slop

About 1 in 10 searches across seven APIs returned at least one known AI-slop page. On one engine, it was nearly 1 in 4. Your model reads it, cites it, and builds answers on garbage.

SlopStop

Quality control powered by over 75,000 real people.

SlopStop flags sites that are mostly AI-generated and downranks them across web, image, and video results. Flagged results still appear, just ranked below original human work.

Community reporting Auto-classification Consensus threshold

Any Kagi member can rely on Kagi's SlopStop feature to flag a page, image, or video using the shield icon next to a result.

Reports are verified against Kagi’s own detection signals. Confirmed AI posts, images, and videos are labeled and downranked, with an option to hide them entirely.

A domain is downranked only when a majority of its content is AI-generated. On mixed sites, AI pages get labeled but not demoted.

API Playground

Try the Search API

Pick a sample query and switch between Lenses, regions, dates, and file types to see how precise your results can get. When you’re ready to build, jump into the API Explorer or browse the docs.

Search Extract

curl https://kagi.com/api/v1/search \
  -H "Authorization: Bearer $KAGI_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
  "query": "steve jobs"
}'

Press Send to load the saved response.

Get your API keys and plug in to our hosted MCP server at mcp.kagi.com/mcp.
Official clients for Go, Python, TypeScript, and Rust, generated from a published OpenAPI spec.

Your agent is only as good as its search results...

Most APIs search a single index. Kagi searches 12 sources with less AI slop and the best results. It’s built for research, compliance, and anything you have to cite.

The Price