AI Search API for web context and cited answers

Choose sources, an answer or a platform AI result

01

Semantic web search

Use Exa retrieval to find pages by meaning, or explore similar-page operations. Supply relevant source material to your own agent and keep control of how it is used.

02

Answers with citations

Exa’s answer endpoint answers a question from web sources with citations. Use it when the task calls for an answer and source references in the same operation.

03

Search-engine AI results

Read supported AI-mode or summary results from search providers. These operations retrieve a platform’s AI output and serve a different task from choosing sources for your own answer.

Explore AI Search API endpoints

Open an endpoint to see its inputs, sample request and price. Choose the provider and operation that fit your task.

8 of 11 matching endpointsCurrent catalog prices · Inspect before running

SERP AI summaryAI-written summary of a finished Google organic SERP task, steered by an optional prompt, optionally with source links and fetched page content. Takes the task id of a Google organic SERP task Run of this Workspace (within 30 days); the SERP task is never repeated here.

DataForSEO$0.01per callView details →

SERP live advancedLive advanced via DataForSEO SERP. One task per call, priced by the vendor's own rate card; an empty result set is reported as no result.

DataForSEO$0.004per callView details →

Answer One Question From Web Sources, With Citations (Exa)Answer one natural-language question from Exa's page index: Exa retrieves the relevant pages first, then returns a direct written answer together with the source pages it cited, each with title, link, published date and author. One question per request, not a batch of questions, and the citations come back with the answer — it is grounded in retrieved pages rather than in a model's memory.

Exa$0.005per callView details →

Deep Neural Retrieval by MeaningMulti-step neural retrieval via Exa's deep mode: the request is expanded into several passes over the page index and the strongest pages are returned, each with title, link, published date and author. Up to five extra query variations, the same category, domain and date filters, and text or highlights per hit. One request per call, up to 10 pages; slower and dearer than the standard mode.

Exa$0.013201per callView details →

Deep Reasoning Retrieval by MeaningMulti-step neural retrieval with an added reasoning pass via Exa's deep-reasoning mode: the request is expanded, the page index is searched in several passes, and the pages are re-judged before the strongest are returned, each with title, link, published date and author. Up to five extra query variations, the same category, domain and date filters, and text or highlights per hit. One request per call, up to 10 pages; the slowest and dearest mode.

Exa$0.0165per callView details →

Find Web Pages Similar to One URL (Exa)Find the web pages that most resemble one page already in hand: Exa's neural index returns the closest pages by meaning and excludes the source page's own domain, each with title, link, published date and author. One URL per request, not a list of urls, and up to 25 pages. The input is a URL rather than a description — the resemblance is to the page behind that link.

Exa$0.007per callView details →

Search the Web for Pages by Meaning, From a Natural-Language Description (Exa)Search an index of web pages by the meaning of a natural-language description instead of by exact keywords: Exa's neural retrieval returns the pages whose content is closest in meaning to what was described, each with its title, link, published date and author. One description per request, up to 25 pages. Ranking is semantic, so a full sentence works better here than a two-word keyword phrase.

Exa$0.007per callView details →

Bing Copilot AnswersBing Copilot's written answer to one question via SerpApi: the answer with its cited sources. One request per call; SerpApi counts every completed request, and an empty result set is reported as no result.

SerpApi$0.015per callView details →

Connect your agent to AI Search API

Step 01
Set up Glasser
Connect a compatible agent with the setup prompt below, or use the HTTP API.
Step 02
Choose your data endpoint
Describe your task. Check the selected endpoint’s required inputs, coverage and price.
Step 03
Run your first request
Call the endpoint through your Glasser key. Use the returned source data in your workflow.

Example agent task

# First, set up your agent
set up https://glasser.ai/SKILL.md

# Then describe the data you need
Find an endpoint that answers a question from web sources with citations. My question is: what are the main approaches to battery recycling? Inspect the endpoint inputs and price, then retrieve the cited answer and preserve its source references.

Agent setup guide ↗API documentation ↗

AI Search API questions

What is the difference between semantic search and an answer API?

Semantic search retrieves relevant pages for your workflow. An answer endpoint generates an answer from sources. Choose retrieval when your agent should handle synthesis; choose an answer operation when you want its cited response.

Does every AI search endpoint return citations?

Citations are part of Exa’s documented answer operation. Other retrieval, summary and platform-AI operations have their own responses. Check the selected endpoint rather than assuming the same citation format.

Can I use this to host a private search index?

This page focuses on web retrieval and supported answers. Private-index hosting, Azure AI Search setup and building a private knowledge base are outside its scope.

Can I restrict sources, time or language?

Use the controls exposed by the selected endpoint’s input schema. Search filters and context controls differ between retrieval, answer and platform-AI operations.