See it in action

Tell us who you are and what your team keeps asking. A person replies to find a time. You don't need to prepare anything.

We use these details only to reply to you.

Every claim keeps its paper trail.

People and agents now write into the same company memory. A note is only as good as the process that made it, so the vault records that process with every note: who wrote it, from what, and who checked it.

Why provenance matters

Agents now produce more than any team can read. More output means less trust when nothing records who wrote what, from which source and who checked it. Epistemic corruption is a real threat in agentic workplaces. This is where it usually goes wrong.

  1. Review can't keep up

    Agents write faster than people can review. The pile of text grows. The share that anyone has checked shrinks.

  2. Context explodes

    Each task pulls in more notes, files and past sessions. Nothing in a full context window says which part someone checked, which part is a guess and which part is out of date.

  3. Reasoning gets lost at handoff

    One agent writes a note. Another reads it a month later. The chain of thought is gone, and only the claim is left.

  4. Authorship fades

    A few months later, nobody can tell what a person wrote from what an agent drafted.

  5. Guesses turn into facts

    An agent's hunch goes in as a plain statement. Later, someone cites it as settled.

  6. Agents read their own output

    The next agent treats an earlier agent's draft as company knowledge. Each error feeds the next one.

A claim is worth what the process behind it is worth.
After Alvin Goldman, “What Is Justified Belief?” (1979)

How the vault does it

Each step below runs today, in the open-source Powerhouse stack.

  1. 1

    Every change is an operation

    Edits go into an append-only log as typed operations. The current state is a replay of that log, so you can open any earlier version.

    Thinking in operations
  2. 2

    Every operation is signed

    Renown ties each key to a person or an agent, and the log records the app that made the change. Anyone can check the signature.

    Identity and signing
  3. 3

    The rules live in the code

    Document models reject a bad write before it lands. A rule in a prompt binds the agent that read it. A rule in the reducer binds every writer.

    Actions and reducers
  4. 4

    Every note names its source

    A note records its author and origin: written, imported, extracted or mined from a session. Its link to the source says why, and how sure: grounded, established or speculative.

    Operation history
  5. 5

    Someone else approves

    Notes start as drafts. A reviewer makes them canonical, and nobody can approve their own note. You decide who may approve.

    Document permissions
  6. 6

    Disagreement stays on record

    When one note contradicts another, the vault opens a tension on both. It stays open until a person settles it.

  7. 7

    Nothing is erased

    A newer note supersedes an older one. The old note moves to the archive with its history and links.

Read further

Working paperReliabilism for the Agentic EraWhy checking AI output is not enough, what a signed operation log records, and when it is rational to trust a memory shared by people and agents.Read the paperDownload PDF