AI & your privacy

Last updated: 2026-05-25

The short version

• Stateless models. Every AI request is processed by a fresh model instance. The model has no memory of you, your friends, your calendar, or any prior request once the response is returned.
• Your input is never shared with the public. tidbok runs AI workloads on Azure OpenAI Service inside our own Microsoft Azure tenant. Your prompts and the model's responses are not sent to OpenAI, not used to train any public model, and not shared with any third party.
• No training on your content. Microsoft contractually guarantees that data submitted to Azure OpenAI is not used to train, retrain, or improve any Microsoft or OpenAI foundation model.
• You can turn it off. AI personalization, the day planner, and assistant chat can be disabled in Settings → Preferencesat any time. We'll keep your account and calendar fully usable without AI.

How AI is used in tidbok

tidbok uses AI in three narrow ways:

1. Day-planner.When you ask “Plan a family afternoon in Morristown,” we send your prompt plus a list of real, publicly-listed events and venues from our database to the model. The model returns a structured plan. The prompt is discarded server-side after the response is generated.
2. Categorization & deduplication. When new events arrive from public sources (city calendars, library RSS feeds, etc.) we ask the model to pick a category and detect duplicates. Your personal data never touches this flow — only the public event metadata.
3. Personalized recommendations.We compute a small numeric vector that summarizes which event categories you've interacted with. That vector — not your raw clicks, names, or messages — is what the recommender uses. You can delete it at any time from Settings → Privacy → AI personalization.

Stateless: what that actually means

Each AI request to tidbok is a single round-trip:

1. You (or our server) compose a prompt.
2. The prompt is sent over a private TLS connection to a model deployment inside our Azure subscription.
3. The model returns a response in seconds.
4. The model instance forgets the prompt immediately. There is no conversation state stored on the model side, and no shared memory across users.

If you use the assistant's “chat” surface, tidbok keeps a short conversation history in your account so you can follow up. That history is encrypted at rest in our database and is only re-sent to the model on your next request to provide continuity. You can wipe it from Settings → AI assistant → Clear history.

Where the model runs

• Provider: Microsoft Azure OpenAI Service.
• Region:United States (East US) — your data does not leave Microsoft's US data centers.
• Models: gpt-4o for reasoning and text-embedding-3-small for personalization vectors.
• Tenant:tidbok's own Azure subscription. We do not call the public OpenAI API.

What we don't do

• We don't send your data to the public OpenAI API.
• We don't use your data to train foundation models — ours or anyone else's.
• We don't sell or share your AI prompts with advertisers, data brokers, or any other third party.
• We don't use AI to make consequential decisions about you (lending, employment, housing, insurance, etc.) — tidbok is a social calendar.

Abuse monitoring

Microsoft Azure OpenAI performs automated abuse monitoring (e.g. detecting prompts that try to generate illegal content). Per Microsoft's policy, this monitoring is fully automated — no human reviews prompts unless an abuse signal is triggered and a compliance investigation is opened. tidbok does not have access to these logs; they are retained by Microsoft for up to 30 days for operational integrity. See Microsoft's abuse-monitoring documentation for details.

Your controls

Questions

Email privacy@tidbok.com with any AI privacy question. For a full picture of how we handle data in general, see the Privacy Policy and GDPR rights page.