Approvals and the activity log
Which AI actions wait for a person, how to approve or deny them, and how proof receipts record everything consequential.
AMOS lets the AI act freely where it is safe, asks a person first where it matters, and keeps a receipt for everything consequential.
How AMOS decides what needs approval
Every action is checked against your company's policy before it runs. It ends in one of three outcomes:
- Runs: routine work, such as reads, goes ahead.
- Runs with a note: the action runs and is marked so you and the AI can see it was a policy-checked write.
- Waits for approval: the action does not run. It is parked with its exact details until a person decides.
A few actions always wait for a person, for everyone including owners, such as setting billing keys and tearing down environments. Owners can tighten the rules for anything else.
Approving or denying
- When something is waiting, the AI gets a review link and can show it to you in the conversation.
- You decide on that review page, or in the approvals list at Settings > Approvals in the AMOS web app.
- Approving runs the parked action once, with the authority of the person who originally asked for it. Denying records the refusal, and nothing runs.
The AI cannot approve its own work. Deciding requires a signed-in person. API keys and ordinary AI connections cannot approve. Approvals in AMOS Desktop use a one-time challenge signed by your installation.
The activity log: proof receipts
Every consequential action leaves a proof receipt: what was intended, what was checked, the evidence, and the outcome. Receipts identify who acted and which connection was used, without revealing any credentials. You can review them on the Receipts page of the AMOS web app, and they can be exported to support clients, boards, auditors or regulators.
Staying in control
Owners can deny pending work, pause an automation, deactivate a user or API key, disconnect an app, and inspect receipts at any time.