Talk to a person.

Use the form on this page for anything about your account, billing, a bug, or a feature that is not behaving. Privacy and data-deletion requests go through the same form — say so in the subject. It stores your message and answers with a reference code; quoting that code finds the exact message again.

Including the email address on your account, roughly when the problem happened, and what you were doing at the time is usually enough to find it. If the interface showed you a reference code, send that too — it identifies the exact request in the logs.

“Use your own AI” means your questions are answered by an AI account you already have, instead of KnowMessenger AI. Those answers don't count toward this month's KnowMessenger AI, so reaching its limit doesn't stop you. You switch it on in Settings → Preferences, under AI.

Option 1, your own API key. Create an API key with Anthropic or OpenAI — those are the two providers supported. In Settings → Preferences choose Use your own AI, then “Your key”, pick the provider, paste the key and press Save. The key is stored encrypted in your Token Holder vault, KnowMessenger sends your question straight to that provider, and the provider bills your account for it.

Option 2, your own agent on a desktop computer. In the KnowMessenger desktop app, choose Use your own AI, then “Local agent”, pick Claude Code or Codex, and choose a working folder. Your question runs through that tool on your own computer, under your own login. It needs that tool installed first, and the desktop app has no public download yet.

Where each one works today: on the web, your own API key only — a website can't run an agent on your computer. The phone apps support neither yet. A conversation that uses your own key has to be continued on the web, and an iPhone can never run a local agent.

Being signed out after an hour is not expected. The session refreshes itself silently in the background, so if you are landing on the sign-in page repeatedly, that is a bug worth reporting rather than something to work around.

If the assistant refuses with a message about model access, the model you picked is not enabled on the deployment you are on. Switching to the default model in settings is the immediate fix; tell us which model you wanted.

If your own API key stops working, the provider is usually the one refusing it: the error message names which provider and which failure — a rejected key, an unrecognised model id, or rate limiting — so you can check it at the source before writing in.

If the assistant says it has requested access to your vault and is waiting, the request is sitting in your wallet's data-access tab. Approving it there and asking again is the whole fix.

All of it is self-service. You can delete a conversation, delete a capture session together with its transcript, and delete your whole account — the last one is under Settings → Data, and it is a button, not a request. It shows you what it will visit and what it cannot reach before you confirm. The deletion runs 14 days after you confirm, you can cancel it from the same screen until then, and once it runs it cannot be undone. Write in only if the button will not run for you, or about the things it says it does not reach.

Inside the app, Verify integrity re-derives your activity chain's head from its rows, so a removed or altered row shows up. Signed receipts checked offline by a zero-dependency script belong to the local agent host, which has no download yet — the drag-and-drop verifier shown on the verify page is a labelled preview, not a working control.

KnowMessenger is operated by Mali Franzese. No response-time guarantee is published, and one is not implied here — you will get a real answer from a person rather than an automated acknowledgement.