AEVS

AEVS is proof-of-execution for AI agents—signed receipts for every tool call, built by Fetch.ai. Here's the problem: when your agent says "Refund of $142 sent," that's just text describing intent. It doesn't prove which tool ran, what inputs were used, or what the API actually returned. Without AEVS, you're relying on chat history and app logs to reconstruct what happened. That's not proof—that's guessing.

AEVS sits between the agent and its tools. Every invocation gets captured: tool name, inputs, output, timing, errors. Then it's ECDSA-signed (P-256), KMS-backed, and linked to the previous receipt in a hash chain. Tamper-evident. The explorer lets anyone verify the signature independently using a reference_id—no need to trust your codebase. It works with LangChain 0.2+, MCP 1.20+, and Python 3.10 through 3.13. Drop-in SDK, your tools stay the same.

The AI agent industry is exploding but there's a massive trust gap. When agents start making real decisions—moving money, approving transactions, accessing sensitive data—someone needs to prove what actually happened. Not what the model said it did. What actually executed. AEVS is building the audit trail for autonomous agents, and every tool call just got a paper trail.