Heurist Finance built an AI investment tool on Amazon Bedrock AgentCore with Claude models
Heurist built the investment chat tool Heurist Finance on Amazon Bedrock AgentCore. Agents orchestrated through Strands use Claude models and purchase premium data on a per-query basis using the x402 protocol in USDC on the Base network.
Heurist built its product Heurist Finance, a chat tool for retail investors, on the Amazon Bedrock AgentCore platform from Amazon Web Services. The Strands tool orchestrates the agent system, and the agents use Claude models from Anthropic available through Amazon Bedrock. The tool combines market data, macroeconomic indicators, company earnings, and news and, according to Heurist, gives retail investors access to analyses such as portfolio construction and scenario analysis, which were previously more commonly available to institutions.
A key part of the architecture is purchasing premium data on a per-query basis through AgentCore payments, because the required data is spread across various paid sources and blanket contracts with providers are not cost-effective for the company. Payments use the x402 protocol in the USDC stablecoin on the Base blockchain, through Payment Manager and Coinbase CDP Payment Connector; each session has a configured spending cap (maxSpendAmount), and if a payment would exceed this limit, the system informs the user and offers alternatives.
Analyses such as correlations, portfolio stress tests, or backtests run in an isolated AgentCore Code Interpreter sandbox without unrestricted network access, which is destroyed once the analysis is complete. User preferences, the state of the investment thesis, and conversation history are stored in AgentCore Memory and tied to a specific user through AgentCore Identity, which, according to Heurist, eliminates the need to build a custom preference store and access controls. Amazon Bedrock Guardrails filters both inputs and outputs — among other things, it is intended to block prompt injection attempts targeting payment and data tools and enforce the company policy against recommending an unhedged position in a single stock.
The source article is cut off midway through the description of an example query about the impact of macroeconomic data on a portfolio. Details can be found in the source article.
Why it matters
The case demonstrates how agent orchestration with Claude models, per-query payments for data in cryptocurrency, and an isolated environment for analyses can be combined in practice into a single auditable system — particularly relevant to development teams designing similar agentic products with paid access to data and requirements for traceability and security.
Relevant practical impact
What this means
For a business
The case shows a concrete reference architecture for companies building agentic financial products: orchestration of Claude models through Strands on the Amazon Bedrock AgentCore platform, payments for premium data on a per-query basis using the x402 protocol in USDC on the Base network, an isolated sandbox for analyses, and auditing across identity, memory, and payments.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.