Pyth and Exa Explore the Financial Market Data Stack for Agentic Search

Why this matters

Financial research tasks need two kinds of information: what the world is saying and what markets are doing.

Exa has built search and agent infrastructure for researching information across the web and more recently across premium data partners. 

Pyth provides structured financial market data, including current prices, historical observations, and candlestick data.

Exa is testing Pyth as a market-data layer for its agentic search, connecting web research with the market data agents need to answer financial questions.

Web search gives agents market information and overall sentiment. Market data gives them market state.

Financial information and market state

An agent can search the web to understand why an asset moved, identify the companies involved in an event, and compare commentary from different sources.

That research is only part of the answer.

A financial question may also depend on the current price, the price at a specific timestamp, the performance over a defined period, or the relationship between several assets.

Those observations need to be retrieved from structured market data rather than inferred from articles, search results, or secondary commentary.

This distinction becomes more important as agents begin supporting financial information analysis.

A useful answer must connect the event with the market data that describes what happened next.

Connecting two layers of financial intelligence

Exa’s agent workflows are designed to search, retrieve, and synthesise information, while Exa Connect allows agents to work with specialist data providers alongside web research.

Pyth’s MCP server gives AI agents access to market-data tools for feed discovery, current prices, historical prices, and candlestick data.

In the testing workflow, the two systems give an agent access to complementary forms of financial intelligence.

• Exa can research the event or question.

• Pyth can provide the relevant market observations.

• The agent can then combine both inputs in a single response.

For example, a user could ask:

What happened to this company after its latest earnings announcement, and how did its stock perform over the following five sessions?

Exa can identify the announcement, summarise the relevant details  and surface the information needed to understand the event. Pyth can retrieve the corresponding market data for the requested period. The agent can connect the two and present the result in a structured format.

The big value is not simply in adding another data source, but giving the agent the right source for each part of the question.

"AI is changing how financial research is done, and that calls for a new approach to market data analysis. By combining Exa's search with Pyth's structured market data, we're exploring what that looks like in practice. We're excited to see where it leads," said Teo Gonzalez , Head of Partnerships at Exa.

From financial research to market analysis

This workflow changes what an agent can do with financial information.

Exa’s agents can connect news, filings, and company research to current and historical market data, enriching the analysis with a direct view of how the market responded.

It can explain an event alongside the price response, compare assets across a common period, or generate time-series data for further analysis.

That creates a foundation for more sophisticated financial workflows, including event-driven research, cross-asset analysis, portfolio monitoring, automated market reports, and prediction-market applications.

Learning how agents consume market data

The Exa–Pyth work is in an early testing phase. The immediate goal is to understand where market data adds the most value inside agent workflows and how agents request it in practice.

Agents may use financial data differently from traditional applications or human analysts.

They may combine news and prices more frequently, request historical snapshots as part of broader research, or compare several markets within a single interaction.

The pilot will help both teams understand those patterns and identify the workflows worth developing further.

It will also provide insight into the technical requirements of machine-native market data, from feed discovery and historical access to response formats and usage patterns.

Building toward more capable AI agents

As the work develops, Exa and Pyth will explore broader applications for agent-native market data.

That may include automated market monitoring, historical event analysis, portfolio workflows, cross-asset research, chart and time-series generation, and other applications that need financial data as part of an automated reasoning process.

Exa already supports sophisticated agentic search and financial research workflows. With Pyth, those workflows can incorporate structured market data alongside the broader research they already perform.

For financial agents, this means enriching web-based research with direct access to the data that describes the market itself. 

Exa and Pyth are exploring what this market-data stack looks like in practice.
Exa gives agents the narrative. Pyth gives them market data.