Skip to content

Activity log & consistency review

The activity log records what your AI queried and saw; the consistency review flags note claims that contradict the rest of your vault. Together they keep your second brain honest — both about what the AI does and about what your own notes say.

AI activity log

Every time a model queries your vault — through the in-app Ask panel, Chat, or a connected MCP client — Altretta records it. The AI activity log is a running history of what the AI queried and saw, so nothing the model reads is invisible after the fact.

This is the accountability layer for AI on your knowledge. Obsidian and Logseq have no notion of it — there's no record of what an assistant touched. Altretta treats AI access to your vault as something worth logging.

Use the log to answer questions like:

  • Which notes did the AI actually read to answer that?
  • Has a connected client been querying parts of my vault I didn't expect?
  • What context has the model seen across this session?

The log pairs naturally with the trust panel in What your AI sees — the trust panel shows the context behind one answer, the activity log shows the pattern across all of them.

Consistency review

Your notes drift. A decision recorded in March gets quietly reversed in June; a fact you wrote one week is contradicted the next. The consistency review is an inbox that flags note claims which contradict the rest of your vault, so you can resolve them instead of letting two conflicting "truths" coexist.

1

Altretta surfaces a conflict

A claim in one note appears to contradict what other notes say. It lands in the review inbox.

2

Verify against the sources

Altretta shows the claim next to the evidence, so you can check the assertion against its sources rather than trusting either note blindly.

3

Resolve it

Decide which is right, update the stale note, and keep your vault internally consistent.

Run the consistency review periodically on fast-moving vaults — project notes, decision logs, research. It's the fastest way to catch a conclusion you changed your mind about but never went back to fix.

Revert an AI action in one click

The activity log doesn't only record what the AI did — for many actions it lets you undo them. Because AI changes are captured as signed actions in the DAG (see Signed history (the DAG)), you can find one in the log and revert it in one click. The revert is itself a new signed action, so undoing an AI edit is just as accountable and traceable as the edit it undoes — nothing is lost from the record.

Not every action is revertible. One-click revert works on the graph-level writes — facts the AI added to or removed from your semantic graph. An action of another kind, such as a direct note edit, a tag change, a new folder or an ingest, is reported as not revertible rather than silently doing nothing. Only actions authored through MCP appear in this log at all.

Prefer to stop questionable writes before they land? Keep your AI read-only — the default — so it can't write at all, and use the Review inbox for notes it proposes.

Honesty, end to end

The activity log and the consistency review complete Altretta's honesty story:

  • Grounded answers cite their sources and admit weak evidence — see Grounded, cited answers.
  • What your AI sees shows the exact context behind an answer — see What your AI sees.
  • The activity log records what the AI queried over time.
  • The consistency review verifies your notes' claims against their sources.

If the activity log or consistency review seems to be missing recent notes, rebuild your vault index so the AI and the checks are working from your current knowledge.