Privacy Policy

Information we collect

Adults only, and information about household members under 18

Wasavi is a service for adults. You must be 18 or older to join the waitlist or create an account. We do not allow accounts for anyone under 18, and the assistant does not message or converse with anyone under 18.

An adult account holder may enter information about household members, including children, so the assistant can help manage the household (for example, a child's practice schedule). That information is collected from the adult account holder, is managed entirely by that account holder, and is used only to provide the service to that household. We do not knowingly collect information directly from anyone under 18. If you believe a person under 18 has created an account or given us information directly, contact us at privacy@wasavi.ai and we will delete it.

Service providers we share information with

We share information with the following service providers only as needed to run Wasavi on our behalf:

We do not sell your personal information. We may also disclose information if required by law, to protect the rights or safety of others, or as part of a merger or acquisition, in which case this policy continues to apply to your information.

AI models and Amazon Bedrock

To answer a request, the assistant sends the information needed for that request to an AI model running on Amazon Bedrock, a service of Amazon Web Services (AWS). We use Anthropic's Claude models (currently Claude Haiku and Claude Sonnet) through Bedrock. Requests go to AWS, not to Anthropic. The model has to read the request to answer it, so at that moment your information is protected by AWS's commitments below rather than by our encryption alone.

What AWS says it does with these requests:

• Not stored. AWS describes Bedrock as zero data retention: by default it does not store model inputs or outputs (Amazon Bedrock abuse detection). We have also set our AWS account's Bedrock data retention setting to "none," AWS's strictest setting, which AWS defines as no request or response data written to durable storage by AWS or shared with the model provider (Amazon Bedrock data retention). It applies in every AWS Region (all in the United States) where our requests can be processed. With this setting, AWS refuses to run any model that requires storing requests, so we cannot use one by mistake.
• No human access. AWS describes Bedrock as zero operator access: no operator of the service can access model inputs or outputs.
• Not shared with model companies. Bedrock runs the models in accounts operated by AWS, and model providers such as Anthropic have no access to prompts or responses (Amazon Bedrock data protection).
• Not used for training. AWS and third-party model providers do not use inputs to or outputs from Bedrock to train models (AWS, Amazon Bedrock FAQs). Wasavi does not train AI models on your information either.

The exceptions AWS describes:

• Automated abuse detection. AWS may use automated systems to detect activity that violates its terms or a model provider's use policies. If they find a potential violation, AWS may ask us about our use of Bedrock and may suspend our access.
• Child sexual abuse material. If AWS's automated systems detect apparent child sexual abuse material in images sent to a model, AWS blocks the request, may store and review the flagged content solely to determine whether it is such material, and may report it to the National Center for Missing and Exploited Children or a relevant authority.

What Wasavi keeps about model use: which model was called, how much text it processed (token counts), and when, recorded under a pseudonymous household reference, to measure cost and reliability. We do not log the content of requests or responses, and Bedrock's optional logging of requests and responses is turned off for our account. AWS keeps its own records of each call (for example, for billing), without the content.

Apart from that review of flagged images, no person at Wasavi or AWS reads your conversations with the model, and nothing in Wasavi watches conversations for emergencies or alerts anyone. If you or someone in your household is in danger, call 911. For a mental health crisis in the United States, call or text 988.

These protections depend on AWS's commitments as published in its documentation and service terms. If AWS changes them in a way that affects your information, we will update this policy, and we will not move your information to a provider with weaker terms without telling you first.

Ideas from the web

Wasavi can suggest things your family might like, such as a new film from a director you talk about, using short web searches. Here is exactly what that involves.

• What is searched. A few words about a topic you've talked about in Wasavi, and what makes an idea fit your family: places and cities (where you live, or where your calendar says you'll be), schools, venues, events and dates. A search can name someone outside your household, such as a public figure, a coach, a club or a friend's school. Wasavi learns these topics only from what the adults say in their own chats.
• What is never searched. The names and nicknames of anyone in your household, including children, as recorded in Wasavi's list of your household's people; email addresses, phone numbers, street addresses, ZIP codes, long account, booking or confirmation numbers, web addresses and map coordinates (an idea's search may be limited to up to 10 named websites, such as a city's events site, checked by the same rules); and a health, money, relationship, religion, politics, legal or sexuality topic together with the name of anyone in your household or anyone else Wasavi knows about. Published code inside Wasavi's secure enclave checks every search against these rules before it leaves, and refuses it otherwise.
• Sensitive topics. Ideas can cover health, money, relationship, religion, politics, legal or sexuality topics you've talked about, such as college savings plans, without anyone's name. Wasavi can search such a topic in general words, such as "pediatric allergist Bellevue". Wasavi's web searches about health never include the name of anyone in your family or anyone Wasavi knows about. The same goes for every other sensitive topic: the published code refuses such a search, and Wasavi is told to leave out everyone else's name too, such as a doctor's. An idea about a sensitive topic appears only in your own private Feed, never in the household's shared Feed, and doesn't send a notification.
• Who sees the search. Amazon Web Services' web search service receives the words of the search. Searches carry nothing that says which household asked.
• Pictures. When an idea has a picture, a Wasavi service fetches it from the public page the search found, without any household identifier, and stores it encrypted with the card. Your devices show it from Wasavi's encrypted store and never contact that website unless you tap the source link.
• Sources. Every idea built on a search shows the pages it came from. Tapping one opens that page in your browser.
• Your activity log. Every search, with its exact words, appears in your activity log before it is sent.
• Turning it off. Any adult can turn web ideas off for the household in Settings, then Proactive help, then Web ideas. Wasavi stops before its next search.

Looking things up while you chat

When you ask Wasavi something in chat that depends on facts that change, such as opening hours, dates, prices, schedules or recent news, it can check them with up to three short web searches before it answers, and shows the pages it used under its answer.

• The same rules. Chat searches follow every rule above: the same things are never searched, the same published code checks every search before it leaves, and each search carries nothing that says which household asked. Chat never fetches pictures.
• What a search can contain. Words from your question, and, in your own private chat, details Wasavi read for you there, such as a venue or date in an email you asked about. A question about health, money or another sensitive topic can be searched in general words, without anyone's name. Never the names or personal details listed above.
• What Wasavi remembers. Wasavi never saves anything from a web page while it answers you.
• Your activity log. Every search appears in the activity log of the chat it was made for (your private log for your private chat) before it is sent.
• Turning it off. Any adult can turn web lookups in chat off for the household, and each adult for their own private chat, in Settings, then Proactive help, then "Look things up on the web in chat".

Cookies and analytics

[To be completed after the tracking audit: list any cookies, analytics, or third-party scripts in use on wasavi.ai and app.wasavi.ai, and how to opt out.]

How long we keep information

• Waitlist information: until you ask us to remove it, or until the waitlist is closed. [Period to be confirmed]
• Account and household information: for as long as your account is open, then deleted within [period to be confirmed] of account closure.
• Server logs: [period to be confirmed].

Your choices and rights

You can ask us to:

• Give you access to, or a copy (export) of, the personal information we hold about you
• Correct inaccurate information
• Delete your information or your account
• Remove you from the waitlist

To make a request, email privacy@wasavi.ai from the address associated with your account or waitlist entry. We may need to verify your identity before acting on a request. We will not discriminate against you for exercising these rights.

International users

Wasavi is operated from, and stores data in, the United States. If you use Wasavi from outside the United States, your information will be transferred to and processed in the United States.

Changes to this policy

If we make material changes, we will update the date above and, where appropriate, notify you by email before the changes take effect.

Contact

Wasavi Technologies LLC
Privacy requests: privacy@wasavi.ai