Heurist Finance Uses Amazon Bedrock to Buy Market Data Per Query

A fintech startup has built an AI-powered investment platform that purchases premium market data on a per-query basis, demonstrating a new model for accessing paywalled financial information without enterprise contracts.

Heurist Finance deployed its system on Amazon Bedrock AgentCore, using the platform's payments capability to buy data from multiple vendors as needed. The architecture combines identity management, isolated code execution, cross-session memory, and cryptocurrency settlement into a single auditable workflow.

The pay-per-query model

Traditional financial data vendors require enterprise contracts that can be difficult to justify before a product reaches scale. Heurist took a different approach: its AI agent purchases only the specific data required to answer each user question.

The system implements the x402 protocol for HTTP-based micropayments. When the agent requests paid data, the merchant returns HTTP 402 with payment terms including amount, recipient, and accepted cryptocurrency. Amazon Bedrock AgentCore payments checks the charge against a per-session spending cap, then signs the payment using USDC stablecoin on the Base blockchain network if approved. The agent retries the request with payment proof, and the merchant releases the data.

This approach gives Heurist access to market prices, macroeconomic indicators, SEC filings, fundamentals, and news from multiple sources without maintaining separate vendor relationships for each.

Personalized research with managed infrastructure

The platform grounds every answer in the user's actual portfolio, watchlist, time horizon, and risk preferences. When a user asks whether a stock is overvalued, the system pulls their holdings from Amazon Aurora PostgreSQL, purchases relevant consensus forecasts, runs correlation and scenario analysis in an isolated sandbox, and synthesizes an answer specific to that investor's situation.

AgentCore Identity carries the authenticated user ID through every service call, ensuring that portfolio data, conversation history, and payment operations remain scoped to one person. AgentCore Memory stores preferences and context across sessions, while AgentCore Code Interpreter executes analysis in a sandbox with no arbitrary network access.

Amazon Bedrock Guardrails filter both input and output. Input filters help block prompt-injection attempts targeting payment and data tools, while output filters enforce Heurist's policy against recommending unhedged single-stock positions.

Why it matters

The architecture demonstrates how managed agent infrastructure can change the economics of data access for smaller companies. By handling identity, memory, sandboxed execution, and payment orchestration, the platform let Heurist's team focus on financial research workflows rather than building infrastructure. The company estimates it reduced agent-system engineering effort by roughly 80 percent compared to building an in-house orchestration stack.

The model also provides predictable per-user marginal costs that support retail pricing, and creates an audit trail connecting every paid data purchase to the resulting analysis. Payment credentials remain in AWS Secrets Manager and are retrieved at runtime, reducing security risk.

What's next

Heurist plans to extend the product with event-driven research tied to earnings calendars, portfolio-aware analysis of market events, and recommendations based on what traders with similar time horizons are researching.

The details were first reported by AWS in a Machine Learning Blog post authored by JW Wang, Joshua Smith, and Chethan Shriyan.