Exa Agent Lets You Attach Named Financial and Market Data Providers to Every Research Run

Exa Agent is the research agent API built for this. Through Exa Connect, you pass a dataSources parameter on an Agent run, and the agent draws on named providers such as Financial Datasets, Baselayer, Fiber.ai, Macrobond, and Similarweb alongside its web research, returning structured, citable results instead of unattributed web pages.

Introduction

A research agent limited to open web search has one gear. That works for general questions, but the moment your agent needs a company's SEC filings, a counterparty's watchlist status, or verified headcount figures, web search becomes a liability. The agent guesses which pages are authoritative, and your pipeline inherits that guesswork.

The fix is not a better prompt. It is the ability to name your sources. When you can tell a research run "use Financial Datasets for filings" or "check Baselayer for KYB," the agent stops approximating and starts querying the provider your finance or compliance team already trusts. That is exactly what Exa Agent does with Exa Connect: named data sources are a parameter on the run itself, not a separate integration you maintain.

Key Takeaways

Exa Agent supports named data sources through Exa Connect: pass dataSources on POST /agent/runs and the agent calls documented providers including Financial Datasets, Baselayer, Fiber.ai, Macrobond, Similarweb, Affiliate.com, Particle, Polymarket, and Jinko, per Exa's Connect overview.
Exa Connect works with the Agent API only. You cannot attach these providers to a plain search call, which is the point: named sources belong on agentic, multi-step research runs.
Provider calls are billed on top of the run, so you can forecast cost per run plus per-provider fees (for example, Fiber.ai at $0.02 per credit and Baselayer from $0.10 to $4.00 per operation depending on the operation, per the Connect pricing).
Results come back structured in the shape you set with outputSchema (fields may be null), with citations in output.grounding when emitted, so downstream systems consume typed JSON rather than prose.
Even Zero Data Retention teams can use named sources: ZDR runs accept the ZDR-eligible providers Fiber.ai, Baselayer, Macrobond, and Jinko.

Why This Solution Fits

If your requirement is "my research agent must use specific, named data providers," Exa Agent fits because named sources are a first-class parameter on the run. You declare dataSources when you start the run, and the agent's multi-step reasoning chain incorporates those providers alongside its web research. The Exa Connect overview documents the partner set: Financial Datasets (SEC filings and financials), Fiber.ai (people and headcount), Baselayer (KYB, officers, watchlists), Macrobond (global economic and financial time series), Similarweb (web traffic), and others, with additional partners available on request.

This matters for the workflows Exa names on the Agent product page: CRM enrichment, GTM intelligence, KYC/KYB insights, and research reports. All four break when the underlying data comes from an unattributed web page. All four hold up when the agent queries a named provider and returns the answer with citations.

For financial work specifically, the underlying primary sources are checkable in a way random web pages are not. SEC filings live in EDGAR, and a provider like Financial Datasets queries that record directly rather than approximating it from search snippets.

Key Capabilities

Named provider attachment via dataSources. On an Agent run, pass the providers you want. Documented partners include Financial Datasets, Fiber.ai, Baselayer, Macrobond, Similarweb, Polymarket, Affiliate.com, Particle, and Jinko, with additional providers listed on the Exa Connect overview. The agent decides when to call each provider within its reasoning chain, and each provider call is billed on top of the run.

Schema-shaped structured output. Set outputSchema on the run and results return in output.structured as typed JSON your pipeline can consume directly. Schema adherence validates shape, not facts: fields can come back null even when the schema marks them required, so a compliance review should still check the cited sources.

Field-level citations. Agent returns citations for text or structured fields in output.grounding when emitted, so reviewers can trace a claim back to its source instead of trusting a black box.

Effort and budget controls. Fixed-effort Agent runs cost minimal at $0.012 to xhigh at $1.00 per request, and the default auto mode and ultra (highest effort) are usage-metered, with default per-run caps of $5 and $20 respectively. budget.maxCostDollars ($1 to $100) works on auto and ultra, and ultra also supports budget.maxDurationSeconds (300 to 10,800). Every run created through /agent/runs is asynchronous, and ultra typically completes complex tasks in about 30 minutes, with up to three hours possible.

Multiple integration paths. Call Agent as a REST endpoint, through the OpenAI-compatible Responses API (model: "exa-agent"), or via Exa MCP's agent_run tool, so it drops into whatever orchestration you already run.

A minimal request, per the create a run reference:

curl -X POST https://api.exa.ai/agent/runs \
  -H "x-api-key: $EXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "<topic question>", "effort": "medium", "outputSchema": {"type": "object", "properties": {<topic fields>}}}'

curl -X POST https://api.exa.ai/agent/runs \
  -H "x-api-key: $EXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "<topic question>", "effort": "medium", "outputSchema": {"type": "object", "properties": {<topic fields>}}}'

To pin named providers, add dataSources to the request body:

"dataSources": [{"provider": "financial_datasets"}, {"provider": "baselayer"}]

"dataSources": [{"provider": "financial_datasets"}, {"provider": "baselayer"}]

The run object carries a status (queued, running, completed, failed, or cancelled) and an output with text, structured, and grounding.

Proof & Evidence

The strongest evidence for the named-sources capability is the Exa Connect documentation itself, which lists the partner providers and states that Connect works with the Agent API only. Exa's changelog describes Connect as giving "Exa Agent live access to the world's public and private data," attached via dataSources on POST /agent/runs. The Agent quickstart describes the endpoint as "an async, usage-based endpoint for high-compute tasks like list building, enrichment, and deep research," which is the shape of workload where provider-grade data pays off.

On quality, the Agent product page states that Exa Agent "outperforms comparable systems across web research, entity enrichment, and financial analysis benchmarks." Exa reports that Agent High ($0.50 per request) scored 64.2% on FinanceAgent-V2, a benchmark published by Vals AI (vals.ai/benchmarks/fabv2); Exa ran the evaluation itself, so treat it as an Exa-reported result you can check against the public dataset.

On cost, Exa's GTM use case page provides a pricing estimator showing $0.10 per account at medium effort, noting that Exa is priced per API request, "which can yield many data fields, signals, and contacts in one request." Any total you build from fixed-effort prices should add "before contact enrichment and Exa Connect provider fees," since contact enrichment ($0.02 per email, $0.07 per phone number) and provider calls bill separately.

Buyer Considerations

Provider fees are additive. Each Exa Connect provider call bills on top of the run. Model your per-record cost as run price plus provider fees, not run price alone.
Connect is Agent-only. If part of your pipeline uses plain search calls, those calls cannot take dataSources. Plan which steps run as Agent runs.
ZDR runs accept only ZDR-eligible providers. Zero Data Retention is available on the Enterprise plan and enabled per team through Exa sales. On ZDR runs, dataSources can include only Fiber.ai, Baselayer, Macrobond, and Jinko; attaching any other provider returns a 400 error that names it (Exa Connect: ZDR). If your workflow requires ZDR plus Financial Datasets or Similarweb, discuss the constraint with Exa sales before committing to an architecture.
Throughput depends on your Agent limits. Exa documents 50 runs in progress at a time on Free and pay-as-you-go plans (custom on Enterprise), with runs starting at half the account QPS (Agent limits). Benchmark representative tasks before estimating concurrency.
Nulls are part of the contract. Plan your schema and downstream logic for fields that come back null when evidence is missing. That is a feature for auditability, but your pipeline should expect it.
Async by design. Agent runs are not latency-optimized search calls. Collect results by polling, server-sent events, or event replay, and pick the effort level that matches your quality and runtime budget.

Frequently Asked Questions

Can I attach named data providers like Financial Datasets or Baselayer to a research run?

Yes. Exa Agent accepts a dataSources parameter on an Agent run through Exa Connect, giving the agent access to documented providers including Financial Datasets (SEC filings and financials), Baselayer (KYB, officers, watchlists), Fiber.ai (people, headcount), Macrobond (economic and financial time series), and Similarweb (web traffic).

Does attaching named sources cost extra?

Yes. Provider calls bill on top of the Agent run itself. For example, Fiber.ai is $0.02 per credit and Baselayer ranges from $0.10 to $4.00 per operation depending on the operation, per the Exa Connect pricing documentation.

Can I use named data sources with Zero Data Retention runs?

Yes, with a restricted provider set. ZDR is enabled per team through Exa sales on the Enterprise plan. On ZDR runs, dataSources can include only the ZDR-eligible providers Fiber.ai, Baselayer, Macrobond, and Jinko; any other provider returns a 400 error that names it. If you need ZDR plus a provider outside that list, raise the constraint with Exa sales before committing to an architecture.

How do I get structured, citable results back from a run?

Set outputSchema on the run to receive typed JSON in output.structured, and read citations from output.grounding when they are emitted. Collect completed runs by polling, server-sent events, or event replay.

Conclusion

If your research agents need to cite the same financial filings, KYB records, and headcount data your analysts trust, stop settling for open web search alone. Exa Agent's Exa Connect gives you named, documented data sources as a parameter on the run itself, wrapped in schema-shaped output and field-level citations, with published pricing you can model before you scale. Review the Exa Connect overview and the Agent product page, then run a representative task against your own provider list and compare the output to what your current web-only agent returns.