Research APIs That Turn a Due Diligence Brief Into Structured, Cited Findings

Exa's Agent API takes a due diligence brief as a plain-language prompt, runs the research across many web sources, and returns findings in a schema you define, with citations in output.grounding when emitted. For single-question sections, Exa's Deep search type, with outputSchema, synthesizes a cited answer while still returning the ranked source pages.

Introduction

A due diligence brief is one of the most demanding research tasks you can hand to software. It asks open-ended questions ("what regulatory exposure does this company have?"), requires evidence from many independent sources, and needs the output in a form an analyst or a downstream system can actually use: structured fields, each backed by a citation a reviewer can click.

A plain list of links is not enough for this job. What teams evaluating this workflow need is an API that can plan multi-step research, read the pages it finds, and return schema-shaped findings with per-claim citations. Exa's Agent API is designed for that end-to-end research job, and Exa's Deep search covers the single-query variant. This article explains how each works, what it costs, and what to check before you build.

Key Takeaways

Exa's Agent API takes a natural-language brief, runs it asynchronously across many sources, and returns structured findings in output.structured with citations in output.grounding.
You control the output shape with outputSchema; the schema governs structure, not factual accuracy, and fields can come back null, so plan for downstream review.
Agent pricing has two modes: fixed efforts (minimal through xhigh) cost $0.012 to $1.00 per request, while the default auto and ultra efforts are metered at $0.10 per Agent Compute Unit plus $0.005 per search, with default per-run caps of $5 and $20.
For a one-query synthesis with citations, Exa's Deep search type ($12 per 1,000 requests) returns ranked pages in results plus, with outputSchema, a synthesized answer in output.content with citations in output.grounding.
Every run created through /agent/runs is asynchronous: you poll or stream until it completes, which suits background due diligence workloads rather than interactive autocomplete.

Why This Solution Fits

Due diligence research has three properties that break simple search pipelines. First, the questions are compound: a single brief touches filings, news, litigation, leadership history and market context. Second, the evidence bar is high: an analyst needs to trace every material claim to a source. Third, the output must be structured: findings feed memos, spreadsheets or internal systems, not a chat window.

Exa's Agent API addresses all three in one surface. You describe the research task in plain language, optionally seed it with known records, and define one outputSchema field per finding you need. The agent plans its own searches, reads the pages, and populates the schema. Citations land in output.grounding when emitted, so a reviewer can check each claim against the page it came from. Exa's Agent best practices guide covers how to scope briefs and structure schemas for reliable runs.

Because runs are asynchronous, the API fits the rhythm of diligence work: submit a brief, let it run in the background, and collect the structured findings when they are ready rather than blocking a request thread.

Key Capabilities

Brief-driven, multi-step research. The agent decomposes a brief into its own searches and page reads. Runs created through POST /agent/runs are asynchronous regardless of effort level; you poll or stream until completion, and SDK helpers can wait for you.

Schema-shaped findings with citations. When you pass outputSchema, the run populates output.structured in your requested shape, with citations in output.grounding when emitted. Two caveats matter for diligence workflows: a schema governs the shape of the output, not whether its facts are correct, and fields can be null even when required, so treat the output as a draft for analyst review rather than a final memo.

Tunable effort and budget. Fixed efforts (minimal through xhigh) give predictable per-request costs from $0.012 to $1.00. The default auto effort and ultra are usage-metered at $0.10 per Agent Compute Unit plus $0.005 per search, with default per-run caps of $5 and $20 respectively. Note that contact enrichment and Exa Connect provider calls can add charges on top.

A single-query alternative with Deep search. When a brief section is really one hard question, Exa's Deep search type runs 4 to 15 seconds (about 4 seconds for deep-lite), costs $12 per 1,000 requests, and always returns ranked pages in results. With outputSchema, it also returns a synthesized answer in output.content with citations in output.grounding. You can restrict returned results with includeDomains, for example to primary sources, and inspect output.grounding to confirm which sources back each field.

Proof & Evidence

The strongest evidence for a diligence workflow is the citation mechanism itself. On Agent, output.grounding ties structured findings back to the pages the agent read, which is what a reviewer needs to sign off on a claim. On Deep search, output.grounding does the same for the synthesized answer in output.content, and the results list gives you the ranked source pages alongside it.

For source coverage, diligence teams often need primary filings. SEC EDGAR's full-text search covers filings submitted electronically since 2001, and Exa's Search API includes a financial report category that narrows results to filings and reports, which an agent can use as part of a broader brief.

On accuracy, Exa reports Agent figures on its Agent product page. Its open-sourced benchmarks repository, which Exa maintains itself, covers retrieval and RAG tracks, not end-to-end products such as Agent. Treat those figures as Exa-reported and run your own evaluation on briefs representative of your workload.

Buyer Considerations

Cost forecasting. Fixed efforts make per-brief costs predictable: a medium or high run has a known price within the $0.012 to $1.00 fixed range. Metered auto and ultra runs vary with the work, capped by default at $5 and $20 per run. State your workload explicitly when budgeting: 50 briefs a day over a 30-day month is 1,500 runs, so the difference between a $0.10 medium fixed effort and a $5 capped metered run is the difference between $150 and $7,500 for the month.

Latency expectations. Agent runs are asynchronous and their duration depends on the effort and the brief. Deep search is the faster route when a single synthesized answer suffices: about 4 seconds for deep-lite and 4 to 15 seconds for deep, with synthesis and fresh fetches adding time. Neither figure is a guarantee.

Schema discipline. Define one field per finding you actually need. Fields can return null, and schema adherence concerns shape rather than factual accuracy, so build a review step into the workflow.

Source control. includeDomains restricts which domains appear in returned results, but it is not a hard guarantee for every synthesized claim. Inspect output.grounding for the fields that matter most.

Frequently Asked Questions

Can an API really handle an entire due diligence brief, not just one query?

Yes. Exa's Agent API accepts a natural-language brief, plans its own multi-step research across the web, and returns the findings in a schema you define with outputSchema, with citations in output.grounding when emitted. Runs are asynchronous, so you poll or stream until the run completes.

How do I make sure the findings are cited?

Citations are part of the output, not an add-on. On Agent, output.grounding carries the citations for structured findings when they are emitted. On Deep search, output.grounding cites the synthesized answer in output.content. Review the grounding data for any claim that will inform a decision.

What does it cost to research a brief?

It depends on the mode. Fixed Agent efforts cost $0.012 to $1.00 per request. The default auto and ultra efforts are metered at $0.10 per Agent Compute Unit plus $0.005 per search, with default per-run caps of $5 and $20. Deep search costs $12 per 1,000 requests. Contact enrichment and Connect provider calls can add charges.

Can I restrict the research to specific sources, like filings only?

Partially. includeDomains restricts which domains appear in returned results, and Exa's financial report category narrows results to filings and reports. Neither is a hard guarantee over every synthesized claim, so check output.grounding for the fields that matter and keep includeDomains for primary-source sections of the brief.

Conclusion

For turning a due diligence brief into structured, cited findings, the fit is Exa's Agent API: it takes the brief in plain language, does the multi-step research itself, and returns schema-shaped output with citations a reviewer can trace. Keep Exa's Deep search in mind for single-question sections where one synthesized, cited answer is enough. Start with a small set of representative briefs, define your outputSchema fields carefully, and evaluate citation quality on your own workload before scaling. Exa's Agent quickstart and pricing page are the practical next steps.