How Incarna let agents pay BlockRun per inference with AgentCore payments
Incarna, which gives AI agents a persistent identity, used Amazon Bedrock AgentCore payments so its agents can pay BlockRun for model inference one request at a time over x402. The team put the pay-per-inference flow into production on Base.
The problem
Agents often need to buy small things mid-task, such as model calls or API responses, with no person around to approve each purchase. Card networks are not built for sub-cent payments. Building custom rails means deciding where funds sit, how payments are signed, how to support x402, and how to stop an agent from overspending.
How they did it
- Store Coinbase CDP or Stripe Privy credentials as a payment credential provider, with secrets kept in AWS Secrets Manager.
- Create a Payment Manager and connector, and set a default spending limit.
- Create a payment instrument, an embedded wallet the agent pays from, which the end user funds and authorizes to sign through a redirect URL.
- Open a payment session with a budget ceiling and expiry before the agent starts a task.
- When BlockRun answers with HTTP 402 and a price, call ProcessPayment so AgentCore payments checks limits and signs from the agent's wallet.
- BlockRun verifies the signature, serves the inference and records a small per-call charge.
Results
- As reported by AWS, the Incarna team finished the full integration in three days (one to build, two to test), against two to three months originally scoped.
- As reported by AWS, the integration took roughly 200 lines of application code.
- As reported by AWS, during the beta agents processed over 1,000 payments ranging from $0.001 to $0.05 per call, each settled individually on-chain.
- The flow is in production on Base, with spending limits enforced by the infrastructure rather than the model.
As reported by the source (AWS Machine Learning Blog (partner case study)); AgentGid did not measure these figures.
This suits teams already on AWS with agents that must buy small services unattended, and who are comfortable with crypto wallets, Base, and a redirect flow where end users fund and authorize signing. Treat the three-day build, 200 lines of code and 1,000-plus payments as AWS-reported figures from a beta, not independent benchmarks. AgentGid has no listed alternatives for this case, so compare other payment setups yourself.
Similar use cases
Route between coding-agent models with the open-source Weave Router 2.0
The Weave team presented Weave Router 2.0 in a Show HN post. It is an open-source routing model that sits in front of coding agents and switches between LLMs per turn, and the aut…
As reported by the Weave author on Hacker News: on Terminal Bench 4.0 and SWE Atlas, the router had equivalent pass rates to GPT-6 Astra.
Run nanoMuse, an open-source personal agent, across phone, desktop and web
The nanoMuse project is an open-source personal agent that runs on Android, iOS, Windows, macOS, Linux and the browser under one account. It can operate apps through their screens…
The article gives no performance metrics; it is a project README.
How Bordr automated Portuguese tax ID order handling with n8n
Bordr, an online service that helps foreigners relocate to Portugal, used n8n to connect its order, payment, email and form tools into automated workflows. This let the small team…
The Power of Attorney PDF workflow has 18 nodes and the workflow that sends the NIF to the customer has nine nodes, as reported by n8n.
