Governed project memory for AI work

Keep the useful context. Keep a person in the loop.

Memory Vault is a governed project-memory layer for AI work. It keeps the original material, separates AI extraction and organisation from approval, and gives future work a reliable place to start.

Memory Vault is currently available as a controlled beta.

The boundary

AI can suggest. A person decides what becomes Truth.

Memory Vault does not silently turn AI output into Truth. Material a person or an authorised assistant submits waits for human review. Only approved project knowledge becomes available as reliable future context.

  1. 01

    Conversation or source

    The useful context starts where the work happens.

  2. 02

    Capture

    A person intentionally sends material to Memory Vault.

  3. 03

    AI extraction

    Memory Vault organises possible facts, actions and direction.

  4. 04

    Human review

    Suggestions remain pending until somebody decides.

  5. 05

    Approved Truth

    Accepted knowledge is kept with its evidence.

  6. 06

    Future context

    Authorised assistants can retrieve what was approved.

Project memory, not a black box

Decisions, actions, risks, questions and roadmap context stay connected to the project they belong to, with the source material that supports them.

Assistants stay within their grant

Connected assistants can read approved knowledge and, when permitted, submit pending material. The ChatGPT plugin cannot approve, edit or delete approved Truth.

Reliable because it is reviewable

The vault keeps what arrived separate from what was inferred and what a person accepted. That distinction is the product.