Summary: Heurist Finance uses Amazon Bedrock AgentCore payments and the x402 protocol to buy premium investment data per query in USDC on Base, subject to per-session spending caps. Its managed flow also carries identity, cross-session memory, isolated code execution, Guardrails, and an audit trail across user, workload, request, and trace IDs. The company estimates about 80% less agent-system engineering than an in-house orchestration stack, freeing work for research and personalization.
If you are building an AI investment assistant, it has to answer something like “How does today’s PCE release impact my portfolio?” with real data rather than vibes. That data usually lives behind paywalls, and justifying enterprise contracts is hard when you have no users yet. Heurist Finance worked around this by purchasing premium data one query at a time through Amazon Bedrock AgentCore payments, relying on the x402 protocol so each question pays only for what it needs.
The per-query payment problem
Heurist set out to bring retail investors institutional-style workflows: market data, filings, news, deep research, portfolio construction, and scenario analysis, all inside a single chat. The underlying data sources are both fragmented and costly. No one vendor has everything, and paying upfront for contracts makes no sense for a small team.
Per-query data purchasing is the better economic model, but it introduces a production problem of its own. An agent has to spend funds on a user’s behalf while custody, spending limits, and audit requirements are all enforced. Every action has to tie back to a specific user, session, and request. That pulls identity, cross-session state, isolated code execution, payment orchestration, and end-to-end tracing into one working system.
Heurist puts the savings at roughly 80% less agent-system engineering than an in-house LLM orchestration stack would have required, since AgentCore takes care of identity, memory, sandboxing, and payments. Those are months of engineering time that go instead into the research workflows and personalization that set the product apart.
How the payment flow works
A user asks a question, and Heurist Finance requests a paid data feed from a merchant. Back comes HTTP 402 carrying x402 payment terms: amount, recipient, asset (USDC), and network (Base). AgentCore payments then checks the payload against the maxSpendAmount cap on a Payment Session. Should the charge run over that cap, the system informs the user and offers alternatives.
When the charge fits the budget, Heurist calls the Process Payment API, which signs the payment through an embedded crypto wallet scoped to Base. Credentials come from AWS Secrets Manager at runtime. The request goes out again with proof in the X-PAYMENT header, and the merchant responds with the data.
The upshot of this flow is that Heurist pays only for the data a given question requires — no vendor contract, no prepayment. A Payment Manager coordinates a CoinbaseCDP Payment Connector, and every interaction gets its own Payment Session carrying a spending cap.
Identity, memory, and audit in one request
A user’s profile and conversation history can expose beliefs, risk tolerance, time horizon, and positions. Heurist therefore builds identity, access, and audit into every request rather than bolting them on.
AgentCore Identity carries the authenticated user across every service call. Each tool call, payment, and memory operation writes down the user ID, workload identity, request ID, and trace ID, which yields a single audit trail spanning the services. AgentCore memory holds user preferences, thesis state, and conversation history between sessions, with Identity keeping the scope to one user.
AgentCore Code Interpreter performs analysis inside an isolated sandbox that has no arbitrary network egress. Once the analysis is done, the sandbox is torn down. Amazon Bedrock Guardrails screens both input and output, stopping prompt-injection attempts directed at payment and data tools and enforcing policies that rule out recommending unhedged single stocks.
What this means for builders
What Heurist did illustrates how managed agent infrastructure frees a small team to concentrate on differentiated product work instead of the system sitting underneath it. The per-query payment model is the piece that matters most: it delivers governed, auditable access to paid data while sparing the team from building payment infrastructure itself.
The relevance stretches past fintech. Every agent that has to spend money on a user’s behalf — whether buying data, calling APIs, or reaching premium services — runs into the same custody, spending limit, and audit requirements. AgentCore payments offers a pattern for doing that safely.
If you are building agentic products, the takeaway is to seek out managed services that cover identity, memory, sandboxing, and payments as a set. As we argued in an earlier piece on FDEs that build your team’s capability, not your dependency, the aim is to steer clear of building undifferentiated infrastructure. Heurist’s 80% reduction in agent-system engineering is a concrete instance of that principle at work.
Heurist is taking the product in three directions: event-driven research tied to earnings calendars, portfolio-aware analysis of market events, and recommendations drawn from what traders with similar horizons are researching. The foundation already exists, and the per-query payment model is what makes those extensions economically viable.