agent-custody gives AI agents signed receipts per tool call, a Cedar-policy MCP gateway, and a Merkle tamper-evident log with offline and browser verifiers.
Yes, agent-custody is open source under the Apache-2.0 license.
self-hostable free; hosted log free to 10k appends/mo, then $50/mo for 1M (freemium).
Yes, agent-custody can be self-hosted (the source is available under the Apache-2.0 license).
agent-custody is a chain-of-custody layer for AI agents providing signed receipts for every tool call, a Cedar-policy MCP gateway, a Merkle tamper-evident transparency log, and offline/browser verifiers, with SDK adapters for LangChain, the OpenAI Agents SDK, and the Claude Agent SDK. Open source: yes (Apache-2.0); self-hostable; freemium. It is active (~217 commits) at the 0.6.x series.
agent-custody builds a provable record of what an agent did. Every tool call gets a signed receipt; those receipts land in a Merkle-structured transparency log whose tamper-evidence anyone can check with the offline or browser verifiers; and a Cedar-policy MCP gateway sits at the boundary to authorize (or deny) each call against declarative policy before it runs. It plugs into existing agents through SDK adapters for LangChain, the OpenAI Agents SDK, and the Claude Agent SDK, and you self-host it with a container plus docker-compose or Kubernetes manifests. Open source: yes (Apache-2.0); self-hostable; freemium.
Once agents send outreach, edit the CRM, or move money, the question auditors and customers ask is "prove what your agent did, and prove nobody tampered with the record." agent-custody answers exactly that: signed per-call receipts, a Cedar-policy gate deciding what is even allowed at the MCP boundary, and a Merkle transparency log that anyone can verify — including offline or in a browser — with drop-in adapters for the SDKs GTM agents are actually built on. Open source: yes (Apache-2.0); self-hostable; freemium. The honest read: the primitives (policy gate + signed receipts + tamper-evident log + independent verifiers) are the right control surface for accountable agent actions, but at 5 stars it is early, there are no published benchmarks, and the tamper-evidence is a documented design rather than an audited guarantee — pilot the gateway and verifier on a low-stakes action flow before relying on it for compliance.