Back
Governance & SecurityTrust by architecture,
not by policy.AI that can act without your knowledge is AI you can't trust. Synap makes human oversight a structural guarantee — not a setting you have to find.
The proposal systemEvery AI write operation — creating an entity, updating a property, linking two items — goes through a proposal before it touches your data.
Full reasoning, always visible.Every proposal shows exactly what the AI wants to do and why. Not a summary — the complete intended mutation, diff-style, with the AI's reasoning attached.
Approve, reject, or edit.You decide what happens to your data. Approve a proposal to apply it. Reject it with a reason the AI learns from. Edit it before applying — full control at every step.
Every approval is reversible.Applied a proposal and changed your mind? Every AI action can be rolled back at any time. Nothing is permanent unless you explicitly confirm it.
Auto-approve whitelistFor operations you fully trust — like AI tagging a captured URL — you can whitelist specific action types for automatic approval. You stay in control of which categories require review.
The event chainEvery change to your data — whether made by you, by AI, or by a connector — is written to an immutable event log. Not a "last modified" timestamp. A full history of what happened, when, why, and who.
Immutable audit trail.Events are append-only. No change can erase a previous state. You can always answer: "What did my data look like last Tuesday at 3pm?"
Who did what.Each event records the actor: your user ID, a specific AI agent, or a named connector (Google Calendar, GitHub). You can trace every change to its origin.
Time-travel and rollback.Reconstruct any past state of any entity. Roll back a specific change without affecting others. The event chain is the foundation for every undo in Synap.
PermissionsFine-grained role-based access control at every level: workspace, entity type, and individual record.
Role-based access control.Workspace roles (owner, admin, member, viewer) determine what users can see and do. Roles can be scoped to specific entity types or views.
AI permission boundaries.AI agents operate within your permission level. An agent can't do anything you couldn't do yourself — and every AI action is still subject to the proposal system on top of that.
Per-workspace overrides.Each workspace can configure its own auto-approve rules, AI permission boundaries, and connector access independently.
Why this matters
Most AI tools treat oversight as a feature — something you toggle on in settings. We treat it as architecture. The proposal system and event chain aren't optional layers — they're how every write operation works, for every user, always.
This means AI in Synap can be genuinely autonomous — running workflows, restructuring knowledge, connecting data across your tools — while you remain in full, auditable control. Not because we limited what AI can do. Because we made it impossible for AI to act without a trace.
