We shipped a free audit trail for the largest agent framework in open source. Apache-2.0, no account, one install.
Every tool call, message, and cron your @openclaw agent runs gets recorded to a tamper-evident chain on your own machine.
Nothing leaves unless you point it somewhere. Anchor proof on-chain through Digital Evidence and anyone can verify the record without trusting you. Or us.
Indirect prompt injection through tool use and RAG retrieval is the primary AI attack surface in 2026. A gateway that only inspects the user prompt is structurally blind to it. Gate AI inspects the full payload, including tool outputs being fed back into the model. https://t.co/ZRqfGxp1AL