Kove is built to cowork with people, not to act on its own. Every product has deliberate limits, every piece of media Kove creates is invisibly marked as AI-generated, and anyone — user or not — can check whether text or an image came from Kove, right here, for free.
Paste a passage or upload an image and the registry answers in one step. Videos found through search are ordinary human-made videos and carry no Kove marking.
Every Kove product has a ceiling on its independence. Kove asks before it acts on your behalf, stops when you stop it, and hands control back at every step. There is no unattended, self-directed mode, and there never will be one — an equal human↔AI working relationship is the whole point of the product.
Kove marks synthetic content at the moment it is created. Generated and edited images carry an invisible, cryptographically signed marker inside the file. Generated speech carries the same marker. Neither changes how the file looks or sounds. Text is covered by an asymmetric registry: every passage over 150 words Kove writes becomes an irreversible fingerprint — no accounts, no prompts, no content, and no way back to the original words. Article 50(2), EU AI Act.
Bennett browses on a server, never on your computer. Downloads and executables never reach your machine, and page content is sanitised before you see it. Bennett is blocked from illegal activity, from bypassing paywalls or logins you have not authorised, and from acting on payments or account changes without you. Sessions are time-limited, one at a time, and end on your word. Secret values you type are hidden from the model and deleted when the session ends.
Kove never holds your passwords. App connections use standard authorisation flows, scoped to the permissions you approve and revocable at any time from the Apps page. Kove only reaches an app when the connector switch is on and the task actually needs it.
The chat model refuses harmful requests — weapons, malware, exploitation, self-harm instructions, targeted harassment and impersonation among them. Image generation runs safety filtering on both prompts and outputs. Kove TTS is for narration and productivity, not for cloning a real person’s voice without consent. Abuse is grounds for account termination.
Your chats are yours. Kove does not train on your conversations; the only content used to improve the product is what you explicitly send with a feedback button, and it is anonymised. The privacy notice has the full picture, including the content registry.