Privacy Policy for Hava

What we collect

When you use Hava, we store:

• Your chat messages and our responses — to improve the

service, fix bugs, and catch gaps in what we cover.

• Your session ID — a random identifier generated per visit.

It doesn't tie to any identity or personal information.

• Technical metadata — which tier of the chat pipeline

answered, how long it took, how many language-model tokens were used.

• A one-way hash of your query — used for rate limiting and

spotting repeated questions, not for identifying you.

If you contribute a business or event, we also store:

• The submission content — name, URL, and any notes you

provide.

• Your email address — only if you choose to provide one for

follow-up.

• A one-way hash of your IP address — used to prevent spam.

We don't store your raw IP address.

In addition to the one-way hash, we keep a plain-text copy of your question in our database so we can read real examples when improving the service — fixing bugs, noticing gaps, and tuning the concierge's voice. Only the operator accesses this; it is not sold or used for advertising.

Why we collect it

We use this data to:

• Make the concierge better — read what people ask, spot what

we can't answer yet, fine-tune the local-friend voice.

• Catch abuse — rate limiting and spam prevention.
• Review contributions — the operator (one person) reviews

submissions before they go live in the catalog.

We do not sell or share this data with anyone else. We do not use it to build profiles of individual users or to target ads.

Who else sees your data (subprocessors)

To answer your questions, we send some of your data to these providers:

• Anthropic — we send your message to Claude Haiku for

filtering, analyzing, and generating responses across the chat pipeline.

• OpenAI — we send your message to OpenAI models (GPT-4.1 mini

and text-embedding-3-small) for extracting hints to maintain session context, generating tags, and computing semantic search embeddings.

• Google Places — when someone contributes a business, we

query Google Places for hours, address, and other details.

These providers have their own privacy and data-retention policies:

• Anthropic: https://www.anthropic.com/privacy
• OpenAI: https://openai.com/privacy
• Google: https://policies.google.com/privacy

We host on Railway, which stores the application logs and database. We use Sentry for error tracking. Both operate under their own policies.

We also use Plausible, a privacy-focused analytics service, to count page views and see which features get used. It is cookieless and does not build persistent profiles of individual visitors; we never send it your chat text or anything that identifies you. Its policy: https://plausible.io/privacy

How long we keep it

Right now, we do not automatically delete old chat messages or contributions. The database retains them indefinitely.

We'll revisit this after six months of operation. If retention becomes a concern you'd like to raise sooner, contact us (below).

Your choices

• You can use Hava without providing an email address or

any personal information.

• Your session gets a fresh random identifier each time you

load or reload the app — no persistent account ties you to past conversations from our side.

• The feedback button at the bottom of each response lets us

know when something's wrong; it's the fastest way to reach us with questions or concerns.

Contact

The fastest way to reach us is the feedback button at the bottom of each chat response — it goes straight to the operator.

You can also reach us by email at hello@havasuchat.com.