I built AHP because I kept seeing the same problem — AI agents are making real decisions in production (approving refunds, calling APIs, managing infrastructure), but there's no standard way to verify what they actually did. Every framework logs differently. Most don't log at all. As agents move into healthcare, finance, and critical infrastructure — and as agents from different organizations start transacting with each other — this becomes a serious reliability and security gap. A hallucinated tool call or a compromised agent shouldn't be invisible. AHP is a common observability protocol for AI agents. It hash-chains every action (HTTP, MCP, A2A, LLM inference, authorization) into a tamper-evident, append-only record. If anything is modified, deleted, or reordered — verification fails instantly. It's framework-agnostic with Python and TypeScript SDKs. Auto-instrumentation means you can drop it into any existing agent system with no code changes. I'd love your feedback — especially on what conformance level (basic recording, cryptographic signing, or independent witnesses) makes sense for your use case. And if you're building agents in production, I'd love to hear what observability gaps you're dealing with today. GitHub: github.com/iamanandsingh/agent-history-protocol

Agent History Protocol

Open standard for AI agent observability

Open standard for AI agent observability