When an AI writes for you, "did it actually do what I asked" matters. Daneshbase makes every change visible and reversible — without making you babysit the agent.
Three review modes
Pick the level of oversight that matches your trust:
- Silent — the agent applies its edits and you read the result. Fast, low-friction; right for low-stakes notes you'd rewrite anyway.
- Auto — clean edits go through automatically; conflicts and contradictions queue for you. Sensible default; trades a tiny bit of speed for a safety net.
- Manual — every change waits in the review queue. No edit lands without you clicking approve. Right when the wiki matters.
The mode is per-base, set in settings. Switch any time.
A review queue, not a notification feed
When an edit waits for you, it lives on a dedicated review page. Each entry shows the doc that's about to change, the diff, the raw source that justified the edit, and which agent proposed it. Approve, reject, or skip — the queue stays organized so a few days of backlog doesn't drown you.
Every change is auditable
Whether you reviewed it or not, every changeset records the doc, the diff, the timestamp, the raw source, and the agent identity. Want to know who edited your brewing notes last Tuesday? It's there.
Reversible
The audit log isn't read-only. Bad edit slipped through? Revert it directly from the history view. The previous version comes back; the bad agent's changeset stays in the log with its reverted annotation so you don't lose the trail.