Privacy overview

Your documents are not training material.

Canonn does not use prompts, completions, or uploaded documents to train its models. Request-body logging is opt-in and off by default.

This page is a readable overview of how Canonn handles information. The complete, controlling policy is the Privacy Policy. Canonn processes account details needed to provide the service, API credentials stored as non-reversible hashes, billing and usage records, and the documents or messages a customer deliberately submits for inference and retrieval.

Inference and uploaded knowledge

Submitted content is processed to answer the customer's request. Uploaded knowledge files are stored for that customer's tenant until deleted by the customer or according to the applicable agreement. Hosted model inference runs in Canada. Tenant boundaries are applied to stored files, retrieval, quotas, and metering. Customers should avoid submitting personal or sensitive information that is not necessary for their use case.

Logs and retention

Prompt and completion body logging is disabled by default. When a customer explicitly enables it for debugging or evaluation, logged bodies are deleted after 30 days. Operational metadata such as timestamps, request identifiers, status codes, token usage, and abuse-prevention signals may be retained as necessary to operate, secure, bill, diagnose, and improve the reliability of the service. Canonn does not sell personal information or share it with advertisers.

Your controls

Customers can revoke and replace API keys, remove uploaded files, disable optional logging, and request access, correction, export, or deletion where applicable. Revoking a connected Google account is also available through Google's account permissions. Send privacy requests to hello@canonn.ai; include the account email and requested action, but never include an API secret or password.

Consequential use

Canonn is designed for grounded answers but model output can still be wrong. It must not be the sole basis for legal, medical, financial, employment, credit, safety-critical, or similarly significant decisions. A competent person should review the source material and output before acting. See the complete terms, acceptable-use rules, subprocessors, and privacy policy for further detail.