A black-box recorder for AI agents
When an agent makes a call based on something it "remembered", you eventually need to answer: who wrote that memory, when, and has it changed since?
Hash chaining in one paragraph
Each log entry stores a hash of the entry before it. If anyone edits or deletes an old entry, the hashes stop lining up and verification fails from that point forward. You do not have to trust whoever runs the database; you can check.
What we log
Memory reads and writes, which agent made them, and when. Our production chain currently has about 800,000 entries that verify.
Why now
The EU AI Act's Article 12 asks high-risk systems to keep automatic logs, and Article 19 says to keep them at least six months. The law does not require hash chaining, but it is the usual engineering answer. More in our Article 12 explainer.