The Service includes features that use artificial intelligence and automated processing to generate suggestions, summaries, organization, estimates, and similar outputs (“AI Output”). AI Output is generated by automated systems, may be inaccurate, incomplete, or unsuitable for your situation, and is not professional advice of any kind, including medical, legal, financial, accounting, tax, or mental-health advice. You are solely responsible for evaluating AI Output and for any decisions or actions you take based on it. Do not rely on the Service for any decision where inaccuracy could cause harm or loss. We make no representation that AI Output is accurate, reliable, or fit for any purpose.
Cloud AI is off when you create an account. Setup may offer an optional Cloud AI disclosure, but Cloud AI stays off unless you explicitly choose Allow Cloud AI. Help improve Capture can be enabled only through its separate Settings disclosure. Kept sends personal content to OpenAI or Anthropic only after you review the applicable disclosure and explicitly allow the relevant permission. Cloud AI authorizes content needed for the cloud-model features you choose; Help improve Capture authorizes only the bounded correction-sample use described above and in the Privacy Policy. Each permission can be turned off independently. Features that require the permission you withdraw may then be unavailable, but account access, synchronization, and non-AI features continue.
Workplace Daily Synthesis uses a separate organization permission that is off by default. An active workspace owner may save permission for future summaries of shared workspace information. Daily Synthesis is not yet available, and saving permission does not start model work. Its data use and permission controls are described in the Privacy Policy.
Some AI features write into your account rather than only answering you. After you allow Cloud AI, opening a goal on iPhone may spend one cloud turn to write that day’s read for that goal. Your account has a daily ceiling on these reads, and under that ceiling a goal is read once a day for the set of modules you have turned on · so if you switch Calendar, Notes, Tasks, Body, or Habits on or off, reopening that goal the same day can spend a second turn on it. Separately, an agent or a connection you have granted access may create a task or a note in your account, or ask you to take one on. Kept records who asked, using the display name carried by the grant you approved rather than anything a model writes, and an item that arrives this way waits for you: work another party marks done counts toward your record only after you accept it. You are responsible for what you grant an agent or a connection, and for what you accept.
Cloud Agents may propose calls to customer-configured MCP servers, and to Kept’s own first-party tools that act on a service you have connected · today, creating one event on a Google Calendar you marked writable. A displayed approval applies only to the exact parameters and destination shown. Approving authorizes Kept to transmit those parameters to the connected service, but it does not guarantee that the service will perform the requested action correctly. A successful server response is an acknowledgment, not independent proof of the external result. Where Kept can read the result back from the service · which it does for the calendar tool · it reports the action as done only after that read matches what you approved. If an outcome is uncertain, Kept may stop without retrying and require you to inspect the external service and record what happened. You are responsible for reviewing approvals, receipts, and material external results before relying on them.
Some external actions cannot be undone. Where an action is not reversible, Kept’s obligation is disclosure and repair, not reversal, and the product does not claim reversibility that does not exist. Irreversible action classes always require your explicit per-action confirmation. If you connect a Google account to Cloud Agents for Gmail, Kept requests read-only access: it requests neither send nor draft-creation authority, because in Google’s model the scope granting draft creation also grants sending. Google therefore refuses any send attempt from that connection, which makes the limit a property of the credential rather than a promise about Kept’s behavior. Published agent capability scores are indexes of verified, settled work as described in product documentation, not intelligence measurements or guarantees of future performance.