Pay-Per-Inference: Amazon Bedrock AgentCore Payments Now Supports Managed On-Demand Agent Payments
Amazon Bedrock AgentCore payments enables AI agents to pay for model inference and other API services on a per-call basis using managed wallets, with enforced spending limits. Integration with BlockRun and Incarna demonstrates quick setup and fine-grained spending control for agent-driven workflows.
What changed?
Amazon Bedrock AgentCore payments added support for managed, on-demand payments from AI agents to API and inference providers. Using the x402 payment protocol, agents can pay per call, with spending governed at the infrastructure level instead of in code. Incarna integrated this with BlockRun, an inference router offering access to models from over 15 providers, allowing agents to pay for each inference request individually, with settlement in USDC on Base. The integration, completed by the Incarna team in three days, demonstrated per-request payments and enforced spending ceilings.
Why does it matter to an everyday developer?
This capability allows developers to build or connect agents that programmatically purchase inference, API calls, or content on-demand, without embedding sensitive credentials or custom governance in the agent code. Spending limits are enforced outside of agent logic or prompt, minimizing the risk of overspending regardless of application bugs or prompt injection. Integrating per-call payments through a managed wallet (including support for Coinbase CDP) reduces the time and complexity of implementing secure agentic payments. This approach removes the need for individual billing integrations with each provider, supporting rapid adoption of pay-as-you-go multi-provider inference workflows.
What can the developer do now?
How to enable pay-per-inference with AgentCore payments
- 1
Set up AgentCore payments
