Archibald312 wires GordonOSS for paid-data accountability
Before letting third-party data feeds plug in, the fork is building a tamper-proof record of every AI call and every tool the system runs.
Archibald312's finance-focused fork of Mike just gained an append-only audit trail. Every time the AI is invoked or a tool is dispatched on a user's behalf, a row gets written to a dedicated log that the database itself refuses to update or delete. The motivation is explicit: paid data connectors are coming, and before outside data flows in, the system needs to prove who saw what and which model handled it.
Rather than duplicating message bodies, the log stores cryptographic fingerprints of inputs and outputs - enough for an investigator to verify what happened without keeping a second copy of sensitive content. User identity is preserved by email so the trail survives if an account is later deleted. A read endpoint lets users page through their own history by project, event type, and time range. Slots are already reserved for future connector and document-movement events.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?