Search APIs for AI
Give your model or agent live web search, so answers use current information instead of only training data.
Trends
Perplexity API gained the most ground in package downloads: +177% against the median npm package over a year.
Public counts each month: pull requests on GitHub, package downloads, stars, Product Hunt followers. They show direction, not market share.
The real choice
Results or answers. Search APIs return links and page content that your own model reads and reasons over, which keeps you in control of the prompt. Answer APIs run the search and the writing for you and return a cited answer, which is less to build but less to steer.
Pick by situation
Tap the ones that are you — the tools that fit light up below.- Your agent's queries are descriptive and keyword search misses them
Exa - You want search plus cleaned page content in one call for a research agent
Tavily
Exa - You want exactly what Google shows and will fetch pages yourself
Serper
SerpApi - You want results from an independent index rather than resold Google or Bing
Brave Search API - You need a finished, cited answer rather than sources to process
Perplexity API
Compare the tools
Grouped by the side of the choice they answer, not ranked. Open a row for when to use it, the trade-off and what makers say.
ExaapiFinding pages by meaning rather than keywords, such as "companies like this one" or "papers similar to this".


24+33 open source910.2k/wk+117% vs npm
Use it whenYour queries are descriptive and traditional keyword search misses them.
Trade-offResults differ from what users expect from Google for short, navigational queries.
Used by Click, Fundable, Human Behavior, Lightfield, Locus Founder and 19 more · in 33 open-source projects.
Search built specifically for AI. When your agents need to find and understand information the way a human researcher would, Exa is in a different league. An essential piece of the Locus Founder stack.
TavilyapiAgent tools that need search plus page extraction in one call, with the content already cleaned.

7+55 open source373.6k/wk+156% vs npm
Use it whenYou're building a research agent and want one API for search, extract and crawl.
Trade-offCredit-based pricing, with advanced searches using more credits per call.
Used by SmythOS, Rox, FuseBase, Bell Cyber, Bmo and 2 more · in 55 open-source projects.
Tavily powers the search component in SmythOS, it makes scraping, proxies, extracting and processing data from the web effortless.
SerperapiStructured Google results — web, news, images, maps, shopping — at a low price per query.


5
Use it whenYou want exactly what Google shows and will fetch page content yourself.
Trade-offReturns snippets and links only; reading the pages is a separate step.
Serper took away the headache of incorporating search data, at an affordable cost.
SerpApiapiStructured results from Google and many other engines — Maps, Shopping, Scholar, YouTube, Bing — through one API.27 open source150.7k/wk+8% vs npm
Use it whenYou need a specific engine's results page, like Maps listings or Shopping prices, as JSON.
Trade-offPriced by monthly search plans, higher per query than Serper; returns the results page, not the pages behind it.
No maker's product we track shows it for this yet · in 27 open-source projects.
Perplexity APIapiGetting a finished answer with citations from one call, without building a search-and-summarize pipeline.


44+17 open source1.4M/wk+177% vs npm
Use it whenYou need a grounded answer, not a list of sources to process.
Trade-offNo free tier, less control over which sources are used, and the API is moving to a new version.
In Lovable, ReplitUsed by Antispace, Cove, FirstQuadrant, Folderly AI, IFTTT and 39 more · in 17 open-source projects.
Perplexity cuts through the noise and delivers one clear, sourced answer instead of a list of links to sort through. We chose it because research shouldn't feel like a part-time job.
Brave Search APIapiWeb search results from Brave's own index, with an answers endpoint and a free monthly credit.
1
Use it whenYou want a large general web index that isn't Google or Bing.
Trade-offCoverage and ranking differ from Google's, which users may notice on local or very fresh queries.
Used by Omnifact.
Before you choose
Decide first whether your app needs raw results (to show sources, filter, or feed its own model) or just a grounded answer. Cap the number of searches per user request — agents in a loop will happily run dozens. Log the queries your agent writes; poor queries cause more bad answers than a poor search provider.
- Letting an agent search in a loop with no cap, so one user question turns into dozens of paid queries.
- Passing search snippets to the model as if they were the page, then getting confident answers built on half a sentence.
- Not showing users the sources behind an answer, so they can't check it when the search returned something wrong.
Other options
Real choices most makers here won't need to weigh.
Decided alongside
What the 74 makers' products here chose for their other decisions.