Finance Search in the Perplexity Agent API is basically a purpose-built financial data nerve center for your AI agents, letting them pull live market data, fundamentals, and documents in one clean, consistent call instead of hacking together five different data vendors and a bunch of brittle scrapers. It takes the finance-heavy workflows that analysts already run in tools like Perplexity Computer for Professional Finance and exposes the same capabilities directly to developers through a single tool interface.

What Perplexity is doing here is formalizing something that was already happening in professional finance: people were using AI search to research companies, draft memos, and prep investment decks, but they still had to reconcile that with terminals, filings, and bespoke APIs. Finance Search collapses that gap by plugging your agent directly into licensed financial datasets, real-time market feeds, and cited web sources, then returning everything in a standard schema the model can reason over. Instead of asking a model to “go search the web” and hope it doesn’t hallucinate a number from an outdated PDF, you ask it to use Finance Search, and it fetches the current price, the latest 10-Q, the relevant line from the earnings call transcript, and the estimates spread it needs to answer the question with traceable evidence.

Under the hood, the Agent API treats Finance Search as one of its built-in tools, alongside things like web_search and fetch_url. When your agent gets a prompt that clearly depends on financial data – say “build a quick DCF for Nvidia based on the latest earnings and current price” – the model autonomously decides to invoke finance_search, hits Perplexity’s financial data layer, and gets back structured JSON with fields for prices, fundamentals, transcript excerpts, and metadata about the source. That JSON becomes the raw material for the rest of the workflow: the agent can plug numbers into a valuation template, generate charts, or write a memo without any extra glue logic or bespoke integrations.

For practical use, Perplexity calls out a handful of “bread and butter” finance workflows that become much easier with this tool. You can build a valuation lookup that doesn’t just show a quote but also slices segment performance and pulls a few key comments from the last earnings call so the numbers are contextualized instead of naked. You can run a change monitor over balance sheets, income statements, cash flows, and other disclosures, asking the agent to flag material moves in leverage, margins, or cash generation across a coverage universe. You can automate earnings recaps that line up reported numbers with what management actually said, or track estimate revisions over time in a way that looks a lot closer to a junior analyst’s notebook than a generic AI summary.

This matters because finance is extremely sensitive to freshness and precision: knowing “roughly” what a company earned last year is useless if the model doesn’t know the updated guidance from last week’s call or the intraday price action after results. Perplexity explicitly designed Finance Search to hit structured, time-sensitive sources instead of letting the model wander through generic web results, which reduces the surface area for hallucinations and outdated numbers. In practice, that means live stock and crypto prices, FX rates, fundamentals, and filings are coming from integrated vendors and official disclosures, not scraped web pages of unknown provenance.

Benchmarking is a big part of the story. Perplexity evaluated Finance Search on FinSearchComp T1, a benchmark focused on time-sensitive data fetching where queries look more like “latest close price” and “current dividend yield” than static textbook questions. In those tests, Perplexity’s finance retrieval configuration started with the highest accuracy for live financial data and stayed the most consistently accurate over time, while also delivering the lowest cost per correct answer in the cohort. Because Finance Search brings back only the relevant structured data rather than massive blobs of unstructured text, the token count stays low and the model spends its budget reasoning instead of parsing pages and pages of noise.

There is also a bigger ecosystem angle. Perplexity recently introduced Computer for Professional Finance, which bundles more than 40 live finance tools, 35 prebuilt analyst workflows, and integrations with providers like Morningstar, PitchBook, Daloopa, Carbon Arc, and others. Finance Search is essentially the developer surface for that same live financial retrieval inside the Agent API: where Computer gives finance teams a full interface for building tearsheets, monitors, and research memos, Finance Search gives engineers the raw programmable building block to recreate and customize all of those workflows inside their own products.

Control and trust are baked into the design in a way that’s friendly to compliance-heavy teams. Every Finance Search result comes with inline citations, so a developer – or a risk officer – can see exactly which source produced a value and how the model used it in its answer. If the agent pulls an EPS figure or a debt balance, you can click through in the Perplexity UI (or follow the metadata in code) to the underlying SEC filing, earnings transcript, or licensed dataset entry, which is critical if you’re building something that has to survive an internal audit or client scrutiny.

From a developer experience point of view, the stable tool interface is arguably the biggest quality-of-life improvement. Without something like Finance Search, building a serious financial agent usually means juggling multiple APIs, each with different auth, rate limits, schemas, and coverage quirks, then writing custom logic to normalize everything into a single internal format. With the Agent API, you configure finance_search once, pick the model, and let the tool route to whichever providers Perplexity has wired up behind the scenes, adding new data sources over time without forcing you to rewrite your application every quarter.

Zoomed out, there’s also a competitive backdrop. Finance has become one of the main battlegrounds for AI platforms, with companies racing to automate the repetitive parts of equity research, credit work, and portfolio monitoring while still keeping humans in the loop. Perplexity’s bet is that combining strong retrieval (via Finance Search and the broader Search API) with an agent framework and a deep vendor catalog can get surprisingly close to terminal-like workflows at a fraction of the complexity and cost.

If you’re building products for analysts, PMs, or even sophisticated retail investors, this launch effectively moves the baseline: instead of shipping a chatbot that “knows some finance,” you can now give users an agent that reads from the same kind of live, licensed data infrastructure they expect from serious tools, with sources and costs that are much easier to reason about. The official docs include recommended configurations that match Perplexity’s benchmarked setups, so teams don’t have to guess at model choice and tool parameters before they see good results.