Privacy Policy

1. Information We Collect

Account information. When you sign in with Google, our authentication provider (Clerk) receives and stores your name, email address, and profile picture from your Google account. We do not set or store passwords — sign-in is Google-only.

Free preview without an account. Visitors in the United States can try a short preview conversation without signing in. No account is created and we do not learn who you are. To stop the preview being abused — it costs us real money each time — our server records a one-way hashed version of your IP address, together with how many previews it has been given, how many minutes of preview time those came to, and whether an account was later created from the same network. That last one is a plain yes/no flag used to see whether previews are useful — it is never linked to an account, an email, or anything identifying you. We keep these records for up to about 31 days, then delete them automatically. We use them to stop the preview being abused and to understand how it is used over time. They are never shared or sold. To be straightforward about what the hash is: it makes the record unreadable at a glance and is not tied to your name or any account, but it is a pseudonym rather than true anonymity — we hold the key that produced it. We also use the country your network reports, to limit the preview to the US. Preview conversations are otherwise handled exactly as described below.

Voice audio and conversations. When you use Companion, your microphone audio is streamed in real time directly from your browser to our third-party AI voice provider, xAI, to generate the AI's spoken responses. Your device connects directly to xAI's voice API using a short-lived access token that our server issues; we do not record or store your voice audio, and we do not keep your conversation transcript unless you have separately agreed to share transcripts with us, described below. Transcripts are displayed live in your browser and exist only for the duration of your session — closing or refreshing the page clears them. The two things that can outlive a conversation are the optional memory summary and the optional transcript sharing described next, each only if you switch it on.

To let you pause and resume a conversation without losing context, xAI temporarily retains a copy of your conversation on its own servers for up to 30 minutes after your last message, after which it is deleted from xAI's systems. See xAI's own privacy policy for details on how it handles this data.

Memory (off unless you turn it on). If you choose to, Companion can keep a short written summary of your conversations so that it doesn't begin from nothing each time — for example, what you have been working through, what you have said helps, and how you like it to talk with you (shorter replies, say, or fewer questions). Those preferences change its manner only; they never change the safety rules it follows. This is off by default for everyone, including people who had an account before the feature existed. It is never switched on for you; you have to choose it, and you can change your mind at any time in Settings.

When it is on, this is what happens. At the end of a conversation your transcript is sent from your browser to our server, used once to rewrite that summary, and then discarded — memory does not store the transcript itself, and leaves no record of it afterwards. The summary is a few sentences at most, and is rewritten in full each time rather than added to — it does not grow into a history. It is held in our own database, hosted in the United States by Neon, linked only to your account identifier. You can delete it at any time in Settings, and you can request a copy of it from us at support@try-companion-ai.com.

Deleting your account deletes your memory. When you delete your Companion account, our systems are notified automatically and the stored summary is deleted with it, along with any transcripts you shared with us. You do not need to delete them separately first.

Nothing you say about suicide or self-harm is ever written into memory. That exclusion is absolute and is not something any setting can change. Separately, health conditions, medication and treatment, religious or spiritual belief, the names of people you mention, specific traumatic events, and legal, immigration or financial details are kept out of the summary unless you switch on "Include sensitive details" — a second, separate choice that stays off even if you turn memory on.

Generating the summary uses the same AI provider as the conversation itself (xAI), under the same terms described above. Memory is only available on an account: the free preview for signed-out visitors has no memory of any kind, and nothing said during a preview is ever written down.

Sharing transcripts to help improve Companion (off unless you turn it on). This is a separate choice from memory, made in Settings, and it is the only circumstance in which we keep a transcript. If you agree, the written transcript of each conversation you have from then on — what you said and what Companion said — is sent from your browser to our server when the conversation ends and stored in our own database, hosted in the United States by Neon, linked to your account identifier. Your voice is never stored, and nothing from conversations you had before agreeing is collected.

An AI system reviews these transcripts to see how Companion responded and to help improve the instructions and safety rules it follows — for example, where it misunderstood someone or gave an unhelpful answer. People at Companion do not read them, except where you ask us to (for example, if you request a copy) or where the law requires it. The review is carried out by xAI, the same AI provider that powers the conversation, processing the transcripts on our behalf. What it produces is a short list of observations about how Companion behaved, checked so that it does not repeat your words or name anything you said; those observations are what people at Companion read, and they are deleted when you delete what you shared. They are not used to train an AI model. We do not sell them and do not use them for advertising. Sharing does not change how Companion remembers you; that is what memory is for.

A conversation in which suicide or self-harm comes up is not kept at all. Before a transcript is stored, the names of people, places and organisations are removed automatically, along with email addresses, links and long numbers such as phone numbers. Finding those names is done by xAI, which already processes the conversation itself. Automatic removal can miss things, so a shared transcript may still contain a detail that identifies you.

You can switch sharing off at any time in Settings, and from then on no further transcripts are saved. Transcripts you had already shared are kept until you delete them: the delete button under "Stored memory" in Settings permanently deletes them along with your memory summary, and deleting your account does the same automatically. Sharing is only available on an account: the free preview for signed-out visitors never shares anything.

Usage and billing information. We store the number of conversation minutes remaining on your account. New accounts are given free minutes on each of their first few days; to tell which day it is where you are, your browser sends its time zone when you open the page or start a conversation. We use it only to work out your local date, and store that date and how many of these free minutes you have been given, not the time zone itself. If you purchase additional minutes, payment is processed entirely by Stripe; we never receive or store your card number or other payment credentials — only confirmation that a payment succeeded and the amount purchased.

Analytics and advertising. Our public pages — the home page, blog, and policy pages — use measurement tools that tell us how people find Companion: Vercel Analytics for aggregate traffic, a Google Ads tag, the Meta (Facebook) Pixel, the Reddit Pixel, and the Microsoft Advertising tag. These record things like the page visited, referring link, approximate location, device and browser, and an identifier the provider may associate with an existing account on their own platform.

These only run if you allow them. We ask before loading the Google Ads tag, the Meta Pixel, the Reddit Pixel or the Microsoft Advertising tag, and they stay off unless you choose "Marketing" or "Allow all" — declining is a single click, and is the default if you ignore the banner entirely. You can change your mind at any time through the Cookie settings link at the bottom of any page. Vercel Analytics sets no cookies at all. Cookies needed to sign you in and keep the site secure (from Clerk and Cloudflare) are always on, since the site cannot work without them.

The Meta and Reddit pixels are disabled once you sign in. Even with marketing consent given, they only run on our public pages — so nothing about the fact that you are using Companion, or how often, is sent to Meta or Reddit.

Embedded videos. Our public pages include demo videos hosted on YouTube. When one of these loads, YouTube (owned by Google) receives your IP address, browser and device information, and the page you came from — this happens as the page loads, whether or not you press play. We embed them in YouTube's privacy-enhanced mode, which defers YouTube's tracking cookies until a video actually plays — though note that one video on our home page starts automatically, so for that one it happens as the page loads. Videos inside blog posts are different: nothing at all is requested from YouTube until you press play on them. Watching a video is subject to Google's own privacy policy.

Advertising platforms never receive anything you say to Companion. The measurement above records page visits only. Google, Meta, Reddit and Microsoft never receive your voice audio or your conversation transcripts. That content is never passed to them, whether or not you use memory or transcript sharing. We do not use any advertising tracker to build a profile of who is using Companion or how often.

2. Third-Party Service Providers

Companion relies on the following third parties to operate:

• xAI — processes your voice audio in real time to power spoken conversations, writes the memory summary if you turn memory on, and reviews transcripts you choose to share.
• Clerk — manages sign-in and account authentication.
• Neon — provides the United States database that stores your memory summary and its settings, if you have turned memory on.
• Stripe — processes payments for additional conversation minutes.
• Vercel — hosts the Companion application and provides aggregate traffic analytics.
• Google — measures the performance of our ads through a Google Ads tag on our public pages, and serves the demo videos embedded there via YouTube.
• Meta (Facebook) — measures the performance of our ads through the Meta Pixel on our public pages, if you allow marketing cookies. It is disabled once you sign in.
• Reddit — measures the performance of our ads through the Reddit Pixel on our public pages, if you allow marketing cookies. It is disabled once you sign in.
• Microsoft — measures the performance of our ads on Bing and the Microsoft Advertising network through the Microsoft Advertising tag, if you allow marketing cookies.

Each of these providers processes data under their own privacy policies and terms, in addition to this one.

3. What We Don't Do

• We do not sell your conversations, transcripts, or voice audio — to anyone, for any purpose.
• We do not record or store your voice audio on our own servers.
• We do not retain your conversation transcripts on our servers unless you have switched on transcript sharing in Settings. If you turn memory on, a transcript is used once to update your summary and is then discarded; without transcript sharing, the transcript itself is never stored by us and never readable by us.
• With memory on, your most recent conversation is kept in your own browser so it is still there if you refresh the page. It stays on that device, is not sent to us, and is replaced by your next conversation. Anyone who can use that browser profile could read it, so on a shared device bear that in mind. Switching memory off, deleting your memory, or signing out erases it immediately.
• We never write anything you say about suicide or self-harm into memory, under any setting.
• We do not use your conversations to train AI models ourselves.
• We do not turn memory on for you, and we do not treat continuing to use Companion as agreement to it.
• We never send advertising platforms information about what you discuss with Companion, and we do not use advertising trackers to profile who is using it or how often.
• We do not exchange your personal information for money. Note that the advertising measurement described in section 1 may count as "sharing" or a "sale" under some state privacy laws, such as California's. You can opt out of it by blocking third-party trackers in your browser.

4. Data Retention

We retain your account information (email, name) and usage/billing metadata for as long as your account exists. If you have turned memory on, your memory summary is kept until you delete it, turn memory off and delete it, or delete your account — deleting your account deletes the summary with it. You can request deletion of your account and associated data via your account menu, or by contacting us at support@try-companion-ai.com.

5. Children's Privacy

Companion is not directed to, and should not be used by, children under 13 (or the minimum age required in your jurisdiction). We do not knowingly collect information from children.

6. Security

We use industry-standard measures — encrypted connections, short-lived access tokens, and reputable third-party providers — to protect your information, but no method of transmission or storage is completely secure.

7. Changes to This Policy

We may update this Privacy Policy from time to time. Material changes will be reflected by an updated "Last updated" date above.

8. Contact Us

Questions about this policy? Email support@try-companion-ai.com.