Search, Page Contents, and Research Agents From One Web Data Platform

Exa covers search, page contents, and research agents on one platform, so you do not need three separate vendors. Its product menu lists a Search API, a Contents API, and an Agent API under one account and one integration: find pages, pull their full text, and hand open-ended research tasks to an agent.

Introduction

Many teams assembling web data for AI applications end up with an awkward trio: one vendor for search results, another for scraping page content, and a third for research workflows. Each vendor means its own API key, rate limits, billing curve, and failure modes to monitor. When something breaks, you are debugging across three dashboards instead of one.

Exa's homepage describes the product in its own words as "Low-latency to deep research in one API". That maps directly onto the three jobs in question: the Search API finds pages, the Contents API returns their full text, and the Agent API runs multi-step research tasks on top of both. This article walks through what each surface does, what it costs, and what to check before you consolidate.

Key Takeaways

• Exa covers all three needs in one platform: Search API for finding pages, Contents API for full page text, and Agent API for autonomous research, all under one account and one integration.
• Pricing is published per surface: Search starts at $4 per 1,000 requests (lite), Contents at $1 per 1,000 pages per content type, and Agent from $0.012 per request at minimal effort.
• Search offers four recommended latency and depth tiers, from lite at roughly 450 ms to deep at 4 to 15 seconds, so you can match the call to the job.
• All three surfaces share the same official SDKs (exa-py and exa-js), so consolidation does not mean learning three client libraries.

Why This Solution Fits

The question you are really asking is whether one vendor can credibly do three different jobs. Exa's answer is that the three jobs share one foundation: a web index built for AI applications. The Search API queries that index, the Contents API extracts from it (including live fetches when you need fresher pages), and the Agent API orchestrates searches, crawls, and reasoning into finished research output.

That shared foundation matters in practice. When your agent finds a page through Search and then needs the full document, you call Contents with the same SDK and API key. When a research task needs ten searches, a docs crawl, and a structured summary, the Agent API does the orchestration instead of your code gluing three vendors together. Exa's docs point long-running research and list-building work at the Agent API rather than at ever-deeper search modes, keeping each surface on the job it is priced for.

It also fits because the surfaces are independently usable. If your immediate need is just search with clean page content, you can start with Search and Contents and add Agent when a workflow genuinely needs autonomy. Consolidation does not have to mean adopting everything on day one.

Key Capabilities

Search API: finding pages, with four tiers of depth and speed. The recommended types are lite, auto (the default), deep-lite, and deep. Latency-sensitive paths can use lite at roughly 450 ms and $4 per 1,000 requests (including up to 10 results, cached content only), while auto returns in about 1 second typical at $7 per 1,000 requests. When the task needs synthesis on top of the ranked results, deep-lite runs in about 4 seconds and deep in 4 to 15 seconds, both at $12 per 1,000 requests; with outputSchema they add synthesized output while the ranked pages stay in results. Results above 10 cost $1 per 1,000 results, and AI page summaries $1 per 1,000 pages. Structured output via outputSchema is available on every Search type except lite, which rejects it, adding a synthesis step and returning the value in output.content with field-level citations and a confidence value in output.grounding.

Contents API: full page text, including subpage crawls. When your code already holds a URL, Contents returns the page's actual content at $1 per 1,000 pages per content type. It supports subpage crawling, so one request can follow a set number of linked subpages (0 to 100) from each starting URL, such as pages in a product section or docs tree; it may return fewer, and each returned URL and content type is billed.

Agent API: research tasks that run themselves. The Agent API is an async, usage-based endpoint for high-compute tasks like list building, enrichment, and deep research. Fixed effort levels are priced per request: minimal at $0.012, low at $0.025, medium at $0.10, high at $0.50, and xhigh at $1.00. The default auto mode and ultra are metered at $0.10 per Agent Compute Unit plus $0.005 per search, with default per-run caps of $5 and $20. Agent returns output.text, output.structured (shaped by outputSchema), and output.grounding with citations when emitted. Runs are collected by polling, server-sent events, or event replay.

A minimal Search request looks like this (REST bodies use camelCase; the Python SDK uses snake_case keyword arguments):

curl -X POST https://api.exa.ai/search \
  -H "Authorization: Bearer $EXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "<topic query>", "type": "auto", "numResults": 10, "contents": {"highlights": true}}'

An abbreviated excerpt of the response shape, with placeholder values:

{
  "results": [
    {
      "url": "https://example.com/page",
      "title": "<page title>",
      "publishedDate": "<YYYY-MM-DD>",
      "highlights": ["<passage that answers the query>"]
    }
  ]
}

The same client can then call exa.getContents([...]) for page text and exa.agent.runs.create({...}) for a research run.

Proof & Evidence

Exa publishes benchmark results for the Agent API on public datasets, with the evaluations run by Exa in June 2026 at high effort ($0.50 per request): 74% on BrowseComp, 0.77 on DeepSearchQA, 52% on WideSearch, and 64.2% on Finance Agent v2. BrowseComp is documented in its own arXiv paper, so the dataset behind the score is independently checkable even though the evaluation run is Exa's own.

On the search side, Exa's benchmark comparison page (updated October 19, 2026, from runs July 8 to 24, 2026) reports a ~450 ms p50 for lite on SealQA and about 1.5 s median for auto in its SealQA comparison benchmark; the Search quickstart lists auto at about 1 second typical. These are Exa's own measurements, so treat them as vendor-published figures and verify against your own workload.

On cost, the published price list makes forecasting straightforward: $4 to $7 per 1,000 search requests depending on search type, $12 per 1,000 for the deep types, $1 per 1,000 pages per content type for Contents, and fixed-effort Agent pricing per request (with the default auto mode and ultra usage-metered), per Exa's pricing documentation. You can test the whole stack before paying anything: Exa's free tier includes $10 in credits every month (up to 2,500 Lite searches), plus a one-time $10 onboarding bonus, with no payment method required (see Exa's pricing page).

Buyer Considerations

• The surfaces price differently, so model your mix. A workload that is 90% Search and 10% Agent costs very differently from one that is mostly Agent runs at high effort. Estimate per surface using the published rates, and remember that Agent contact enrichment and Exa Connect provider calls are billed separately.
• Agent is asynchronous. Runs are collected by polling, server-sent events, or event replay, and throughput depends on your account's Agent limits. Exa advises benchmarking before estimating concurrency, so if your workflow needs synchronous answers in seconds, a deep search type may fit better.
• Structured fields need validation. Agent's outputSchema governs shape, not facts; strict validation isn't guaranteed, and fields can be null even when you marked them required. Citations appear in output.grounding when emitted, not on every data point. Plan a light verification step for business-critical fields.
• One vendor means one bill, but check the line items. All three surfaces bill against the same account, which simplifies procurement, but you should still attribute spend per surface so a spike in Agent usage does not surprise the Search budget.
• Consolidation is still a migration. Moving off existing vendors means re-testing result quality against your own queries. The free credits make that comparison cheap to run before you cancel anything.

Frequently Asked Questions

Can I start with just Search and Contents and add the Agent API later?

Yes. All three surfaces share one account, one API key, and the same SDKs, so you can integrate Search and Contents first and add Agent when a workflow genuinely needs autonomous research. Nothing about the Search or Contents integration changes when you add Agent.

How does pricing work across the three surfaces?

Each surface bills separately against published rates: Search at $4 per 1,000 requests for lite and $7 for auto (the deep types at $12 per 1,000), Contents at $1 per 1,000 pages per content type, and Agent at $0.012 to $1.00 per request for fixed efforts, with the default auto mode and ultra usage-metered (default per-run caps of $5 and $20). One bill, but line items you can attribute to each surface.

Does the Agent API return citations for its research?

Agent returns citations in output.grounding for text or structured fields when they are emitted. It is not guaranteed that every field carries a citation, so treat citations as strong evidence when present rather than a universal guarantee.

Do I need separate SDKs or API keys for each surface?

No. The official exa-py (Python) and exa-js (JavaScript/TypeScript) SDKs cover the surfaces, and all three run against one account and API key. That is the practical difference between one platform and three vendors: one client library and one auth story, with rate limits set per your account plan.

Conclusion

You do not need three vendors to cover search, page contents, and research agents. Exa's Search API, Contents API, and Agent API cover those three jobs on one index, under one account, with published pricing for each surface and public benchmark datasets behind the Agent results. Start with the surfaces your workflow actually needs, use the free credits to benchmark result quality on your own queries, and consolidate the rest only when a workflow demands it. The fastest next step is to try the API for free and run your first real query through Search and Contents in an afternoon.

Related Articles

• The Default Web Search API for Agent Frameworks: Why Exa Is the Practical Choice
• What Web APIs Are Designed for Agents to Call as Tools, with Structured Outputs and Citations?
• The Cost of Building Financial Research Infrastructure With a Web-Data API