An AI assistant can infer something a person never explicitly told it. A note about relocating, for example, might become a guess about a job change. That example is fictional, but it captures a question behind Vivendur: if a memory system keeps a guess, can the person responsible see its source and decide whether it should be used?
I’m Azin, Vivendur’s founder. As a doctor and researcher, I worked with sensitive information that limited which AI tools we could use. We’re building a separate memory layer so an organization can decide what each tool receives for a task.
What we have today
We have a technical design and reference implementation. Our team reports checks of source-linked inferences, disclosure rules and removal of tracked derivations. The production core is still being built.
The status page describes our internal reference checks and next milestones. Our verification notes explain their scope and how we plan to publish reproducible results.
Something small you can inspect
We’ve placed a browser-only SHA-256 demonstration on the Architecture & verification page. Edit the sample text and inspect the digest it produces.
This demonstrates hashing, not a working Vivendur receipt, product security or verified deletion. Because the page serves both the text and the digest, it is not independent evidence that a historical statement stayed unchanged. A signed receipt and a trustworthy audit history require more than this example shows.
What comes next
We plan to publish versioned examples and test results, followed by a public receipt verifier and independent review of the production core. Future build-log entries should identify what changed, what readers can reproduce and what each result does not establish.
We’re also accepting design-partner conversations using fictional or redacted workflows. These are discussions ahead of any separately agreed pilot.
— Azin