Build a Dated, Sourced Company Event Timeline From the Web With Exa Agent

Compiling a company's major events from the web, with dates and citations attached to each entry, is a deep research task, and Exa Agent is built for exactly that. Exa's Agent quickstart calls it "an async, usage-based endpoint for high-compute tasks like list building, enrichment, and deep research," which describes a corporate timeline precisely.

Introduction

Anyone who has tried to reconstruct a company's history by hand knows the workflow: search for funding rounds, open a dozen news articles, copy dates into a spreadsheet, then hunt down the original press release to confirm each one. Repeat that across a portfolio of companies and the manual effort becomes the bottleneck. Generic chat assistants make it worse in a specific way: they produce plausible narratives with no way to check where each date came from.

What buyers actually need is a research pipeline that returns a timeline as data: one row per event, an ISO date, an event type, a short description, and a citation to the page that supports it. That is a structured, verifiable research output, not a chat response. This article explains why Exa Agent is the right tool for the job, what it can do, and what to check before you buy.

Key Takeaways

Exa Agent is an async, usage-based API endpoint for high-compute research tasks such as deep research, list building, and enrichment, per the Agent quickstart.
You can define an outputSchema so each timeline event comes back as JSON shaped by your schema with fields like date, event type, and description (any field may be null).
Citations are returned in output.grounding when emitted, so each event can carry evidence rather than an unsupported claim.
Fixed-effort Agent runs cost $0.012 (minimal) to $1.00 (xhigh) per request, and the default auto mode and ultra are usage-metered with default per-run caps of $5 and $20, before contact enrichment and Exa Connect provider fees.
Exa reports that Exa Agent outperforms comparable systems across web research, entity enrichment, and financial analysis benchmarks.

Why This Solution Fits

A company event timeline is not a single lookup. It is a multi-step reasoning chain: identify the company, sweep news coverage and press releases, distinguish funding rounds from leadership changes from product launches, normalize conflicting dates, and attach a source to every entry. Exa's quickstart explicitly lists multi-hop tasks like "find companies, then find their decision makers" as core Agent work, and timeline reconstruction is the same class of problem.

It also fits because the output is consumed by software, not by a human reading prose. Agent supports outputSchema, so you can request a JSON array of events with typed fields. Fields can return null even when the schema marks them required; as Exa's Agent best practices put it, "Schema adherence validates shape, not facts." That honesty is a feature for timeline work: an event with a null date is a flag for review, not a fabricated date that silently corrupts your dataset.

Finally, the async architecture matters. A thorough timeline for one company can require dozens of searches and page reads. Exa Agent runs those as background jobs you collect by polling, server-sent events, or event replay, so a long research run does not hold an HTTP connection open.

Key Capabilities

Structured timeline output. Define an outputSchema describing your event model: event_date, event_type, headline, summary. Agent returns output.structured in that schema's shape (strict validation isn't guaranteed), ready to load straight into a database or visualization library. Do not add a source URL field to the schema: Exa's best practices say "Do not duplicate Exa's built-in citations or confidence in your schema," and sources come from output.grounding.

Field-level citations. When emitted, output.grounding returns citations for text or structured fields, so each timeline entry can be traced back to the page that supports it. This is what separates an auditable timeline from an AI-generated summary.

Deep research at scale. Agent is designed for multi-step research chains, not single queries. For a portfolio-wide project, such as timelines for 200 companies, you can launch runs per company and collect results programmatically. Concurrency is documented, not unbounded: 50 Agent runs can be in progress at a time on Free and pay-as-you-go plans (custom on Enterprise), and runs start at half the account QPS, per Exa's Agent limits. Benchmark a representative task first, as Exa's best practices recommend.

Tunable effort and cost. Fixed effort levels run from minimal ($0.012) through low ($0.025), medium ($0.10), high ($0.50), and xhigh ($1.00) per request. Metered modes include auto (the default, with a $5 default cap) and ultra (highest effort, $20 default cap), where budget.maxCostDollars acts as a hard ceiling. Ultra runs typically finish complex tasks in about 30 minutes and can take up to 3 hours, per Exa's Agent Ultra docs. A timeline sweep across a decade of news is a good candidate for high or ultra; a quick recap of a small startup is not. All of these figures are per Exa's pricing documentation, before contact enrichment and Exa Connect provider fees.

Finance-grade sources when you need them. Through Exa Connect dataSources, Agent can pull from Financial Datasets (SEC filings and financials), Fiber.ai, Baselayer (KYB, officers, watchlists), Similarweb, and others. SEC filings are often the most authoritative source for dates like IPOs, acquisitions, and officer changes; the dated primary documents are searchable on SEC EDGAR.

Flexible access paths. Call Agent through its native API, the Responses API (model: "exa-agent", with synchronous, streaming, or background execution; synchronous high, xhigh and ultra requests are rejected and need streaming or background mode), or Exa MCP's agent_run tool, per the Agent quickstart.

Proof & Evidence

Exa's Agent API product page states that Exa Agent "outperforms comparable systems across web research, entity enrichment, and financial analysis benchmarks," and lists research reports among its use cases. On public benchmarks, Exa reports that Agent High ($0.50 per request) scored 74% on BrowseComp, 0.77 on DeepSearchQA, 52% on WideSearch, and 64.2% on FinanceAgent-V2, a benchmark published by Vals AI (FinanceAgent-V2). These evaluations were run by Exa, so treat them as Exa's reported results.

For timeline work specifically, the relevant capabilities are the ones documented in Exa's own reference material: schema-shaped structured output, grounding citations returned when emitted, multi-step research as a first-class task, and finance data sources via Exa Connect. Each of those is verifiable on the Agent quickstart and best practices pages.

Buyer Considerations

Citations are conditional. Grounding is returned when emitted, not guaranteed on every field. Build your pipeline to handle events without citations, and spot-check a sample before publishing.
Nulls are part of the contract. A required field can still be null when evidence is missing. Design downstream code accordingly.
Cost scales with depth. A fixed-effort estimate like 500 companies at high ($0.50) comes to $250, before contact enrichment and Exa Connect provider fees. Metered auto and ultra runs bill per Agent Compute Unit ($0.10/ACU) and per search ($0.005), so set budget.maxCostDollars on heavy jobs.
Throughput is account-dependent. Do not assume unbounded parallelism; 50 runs can be in progress at a time on self-serve plans, so check Agent limits and benchmark your task shape first.
Verify before you publish. For anything public-facing, treat Agent output as a researched draft with evidence attached, and confirm load-bearing dates against the cited sources.

Frequently Asked Questions

Can Exa Agent return each timeline event with its source URL?

Yes, through grounding rather than a schema field. Each event field's citations arrive in output.grounding, so you do not need a source URL field in your outputSchema. Citations are returned when emitted, so handle the case where grounding is absent for a given field.

How much does a company timeline run cost?

Fixed effort levels range from $0.012 (minimal) to $1.00 (xhigh) per request, and metered auto and ultra bill per Agent Compute Unit and per search with configurable budget caps. Any total you estimate from fixed-effort pricing excludes contact enrichment and Exa Connect provider fees.

Can I get timelines for many companies at once?

Yes. Launch one Agent run per company and collect results by polling, server-sent events, or event replay. Concurrency is capped at 50 runs in progress on Free and pay-as-you-go plans (custom on Enterprise), so benchmark a representative run before scaling.

What if the agent cannot find a date for an event?

The field can return null even when your schema marks it required, because schema adherence validates shape, not facts. That is preferable to a guessed date: your pipeline can route null-date events to manual review.

Conclusion

A company event timeline is only as good as its dates and its sources, and that is exactly where manual research and generic chat tools fall short. Exa Agent approaches the problem as infrastructure: an async, usage-based endpoint that runs multi-step web research, returns JSON shaped by your schema, and attaches citations in output.grounding when emitted, with finance-grade data sources available through Exa Connect. If you need dated, sourced timelines you can audit and pipe into your own systems, start with the Agent quickstart, define your event schema, and run your first timeline today.