Workspace controls

Autonomy, models and connections

Control what ACC may do automatically, how model providers are routed, and which repeated actions may earn narrowly scoped trust after your explicit approval.

Autonomy policy

Auto = ACC continues. Approval = a human decision is required. Deny = blocked. Deny always overrides learned trust.

Daily AI budget

Estimated ceilingPrevents new runs after the configured daily limit is reached.

AI routing

Loading…

Routing policy

ModeAuto follows ACC's configured provider order. Manual puts your chosen provider first.

Preferred providerOnly configured providers are selectable.

FallbackIf enabled, ACC may try another configured provider after a provider/model failure.

Allow fallback

Why this order?

Loading routing policy…

Cost-first and quality-first routing are intentionally not offered yet. ACC needs trustworthy per-model cost and outcome telemetry before those modes can optimize anything rather than guess.

Connections

Progressive autonomy

ACC may recommend narrower approval rules after repeated successful approvals. It never grants itself permission. Current scoped trust is limited to exact Gmail recipients; new recipients still require the normal policy.

Scoped trust can only turn Approval → Auto for the exact trusted recipient. A global Deny still blocks the action, and Spend/Delete are never affected by recipient trust.

Recommendations and trusted recipients

Loading trust evidence…