LOKEE AI AND CONTENT PROCESSING POLICY

Effective date: 25 June 2026 (25.06.2026)
Legal URL: https://lokee.app/ai-policy

1. AI is optional and user-initiated

Lokee AI features are optional. AI is used to suggest extraction, classification, relationships, summaries, clustering, entity candidates and similar structured proposals. The user remains the decision-maker.

2. Providers

Managed AI may use OpenAI and Anthropic. BYOK is planned only for local desktop use. User API keys should not be stored as cloud account secrets unless a separate product change and policy update introduces that feature.

3. What may be processed

At launch, AI workflows may process whole uploaded documents, not only short snippets. Users should not submit private/sensitive personal data, prohibited content, confidential third-party content or copyrighted material unless they have the legal right and the necessary safeguards.

4. Review-gated persistence

AI output is advisory until accepted. Extraction/review batches may be stored until the user accepts, rejects or resolves the batch. Nothing should be written permanently to the vault, board, timeline or persistent entity graph without explicit user action.

5. No training of our own models

We do not use vault content to train our own models and do not plan to do so. External AI providers may process submitted content according to their own terms, data-processing settings and retention policies. We should configure provider accounts to minimise retention where available.

6. No general AI logs

Lokee does not intentionally store prompts and AI outputs as general AI logs. The exception is product-state data required for the user-visible review process, such as extraction batches waiting for acceptance or rejection.

7. Credits and limits

Managed AI may require purchasable credits. Coupons and promotional AI credits may be issued. Credits may be consumed when a request is processed, even if the user later rejects the generated proposal.

8. Accuracy and responsibility

AI outputs can be incomplete, inaccurate, biased, inconsistent or unsuitable. Users must verify outputs before relying on them, publishing them or using them for research, fiction continuity, historical interpretation, business use or legal decisions.