Privacy Policy

Effective Date: August 6, 2026
Previous version: Privacy Policy of July 13, 2026

If you only have one minute, read this

This is a long document, and some of it is dull. That is what legal precision costs. Read it properly if you can, or open it in Chat and let your assistant in Phi Browser summarize it and answer your questions. It is a good first use of the AI, and your privacy is worth ten minutes.

We are not interested in who you are. Not what you browse, not what you search for at two in the morning, not the things you would rather nobody knew. That is yours, and it stays yours. What we want is to build the best AI browser there is, and for you to stay because you like using it.

You do not have to take that on trust, because our business does not depend on you being interesting. We do not sell ads. We do not sell personal data. We do not train AI models on your content, and to be blunt about why: training a model takes a fleet of GPUs we do not own and a bill we could not pay. We are browser people. We make a browser. We would rather you paid us for a better one than have an advertiser pay us for you.

If we wanted your secrets, you would have a problem. We do not, so you do not. But intent is the weakest thing we could offer you, and good intentions have never stopped a breach or a leaking log. What stops those is built rather than promised: what we never collect, what never leaves your Mac, what a switch actually turns off, and how briefly anything survives. That is what the rest of this is made of, and it is what you should hold us to.

What does this Privacy Policy cover?

At Phinomenon Inc., we built Phi Browser for people with privacy in mind. This policy explains, in plain language, what personal data we collect, why, who we share it with, how long we keep it, and the rights you have over it. It is a transparency document: it describes our practices, and it is not a contract you "accept" by using Phi. Your rights under privacy law exist regardless of anything written here.

Signing in is the choice. The first screen offers "Explore Phi without signing in": take it and Phi is a plain browser, with no account, no AI, no memory, no sidebar chat and no Phi Link, and most of this policy describes things that simply do not happen. Sign in or create an account and you get the AI browser, with the AI features on, because that is what the account is for. You can still switch them all off with one setting.

This Privacy Policy covers how we handle personal data that we gather when you use Phi Browser and our related services, including our account service (account.phibrowser.com), our AI service (ai-service.phibrowser.com), our connector service (ai.phibrowser.com), and our domains phibrowser.com, phi.cc and phi.to, all operated by Phinomenon Inc. Two more of our hostnames appear in Phi's network traffic and are covered by this policy too: g.phibrowser.com, where the product measurement described below is received (our self-hosted analytics), ota.phibrowser.com, which serves Phi's application updates, and sentinel.phibrowser.com, which serves component updates for Phi Sentinel. Generally, "personal data" means any information that identifies or relates to you. This policy does not cover the practices of companies we don't own or control or people we don't manage.

We're always working to make Phi better, which means this Privacy Policy might change from time to time. When we make material changes, we will notify you in advance by placing a notice on our website or sending you an email, and where the law requires your consent for a new use of your data, we will ask for it rather than assume it.

What we do NOT collect and things we NEVER do?

Before we list what we do collect, here is the more important list, the things that never reach us at all:

• The pages you visit and your browsing context. Your browsing stays in your browser.
• Your prompts, your AI conversations, and the responses the AI gives you. When you use AI features through Phi Cloud, the content needed to answer is sent to the provider of the model that serves your request, which may be OpenAI, Google, Anthropic, or SpaceXAI, or the host of one of our open-weight models. It passes through, it is not retained by us, and our relay keeps no copy of requests or responses, including in logs. These providers do not use API traffic to train their models by default, and wherever a provider offers an additional no-training control we have turned it on; that processing is a relationship between you and the provider, with us as the facilitator.
• Your browser memory. Phi's memory of your browsing is created and stored on your device and remains your property. We do not collect it or upload it; see "What are Phi's AI products and features?".
• Chat logs through Phi Link. With our official Telegram bot we are a pure proxy: messages are delivered, never cached, never stored. With your own bot, the traffic never touches us at all.

And with the data we do hold, here is what we never do:

• We do not sell your personal data for money, to anyone, ever.
• We do not show you ads inside Phi Browser, and Phi Browser itself does not use advertising cookies or trackers.
• We do not train AI models on your content, full stop.
• We do not share data with data brokers or advertising networks. The one exception is the advertising conversion pixel on our public website, which measures whether an advert led to a download and loads only if you consent to it; see "Cookies and Similar Technologies".
• We do not build advertising or behavioural profiles of you from product measurement, and we do not use it to make automated decisions about you. The anonymous baseline described in the next section is counted, not profiled.

What personal data do we collect?

Data you provide to us:

A plain word about this whole category: this is data you hand us, knowingly. We ask, you answer, and for everything that is not strictly needed to run your account you can decline and Phi keeps working. Nothing here is taken from you quietly.

• Account or contact data (only if you create an account): Signing in is optional. If you choose to create an account, we collect your email address, password, display name, avatar, user ID, your year of birth from our sign-up age gate (we ask for the full date of birth only if the year alone cannot establish that you are an adult), and related account records such as invitations and activation status. Our authentication service is operated by Auth0 (Okta, Inc.), and this is the full extent of what it holds: your sign-in credentials and the account records just listed. The display name is whatever you choose to call yourself: it can be anything, and nothing requires it to be your real name. If you sign in with a social account, the name it carries over is only a starting point, and you can change it right at setup or any time after. Auth0 never holds your payment card details, browsing history, or AI conversations. If you explore Phi without signing in, none of this data exists, because there is no account.

• Payment data: If you choose to purchase paid features, we collect payment data such as payment card type and billing details. Payment data is predominantly processed and stored by our payment processor, Stripe.

• AI usage metering: When you use AI features through Phi Cloud we record, for each request, the model used, token counts, and the resulting cost. This is metering data, not content: it never includes what you asked or what the model answered. We collect it because model tokens cost us real money on a free product, and it is what lets us keep a free tier while stopping abuse (see the fair-use terms in our Terms of Use). It is recorded in Langfuse, self-hosted on our own infrastructure, so no third party receives it.

• AI and Context data: created by you, kept on your device. Your AI memory, agent state, Context, and the inputs and outputs of AI features live on your machine. We do not collect or store them. What our servers keep for agent features is thin task metadata, meaning which task ran, when, and its status, never the content or your memory. If you use Phi Cloud, the content needed to answer a request transits to the model provider to generate the response and is not retained by us (see "What are Phi's AI products and features?").

• Connector and synced data: If you connect a third-party account (for example email, documents, or other services) through our connector pipeline at ai.phibrowser.com, we store the OAuth tokens needed to access that account and a synced copy of the content you choose to connect. You can disconnect a connector at any time, which immediately revokes and deletes our OAuth tokens and deletes the synced copy.

• Phi Link (messaging pairing): Phi Link connects a messaging app such as Telegram to Phi, and it comes in two shapes with different privacy footprints, so we describe both.

  • Our official Telegram bot: we store the pairing record that links your chat to your Phi, and we act as a proxy for the messages in between. Proxy means exactly that: the content passes through and is delivered, and we do not cache or store the chat logs that flow through it.
  • Your own Telegram bot: you can instead create your own bot, in which case traffic flows directly between Telegram's service and your Phi Browser and does not route through us at all. That relationship is between you and Telegram under Telegram's terms; we just make the connection possible.

• Feedback: What you send when you click the "Send feedback to Phi" button, answer a survey, or rate an AI response, including any attachments you choose to add.

Data we collect automatically:

• Product measurement, governed by one switch: "Help improve Phi's features and performance." The switch controls two things: detailed usage statistics and crash reports. Underneath both, a small anonymous baseline always runs. We measure at all because otherwise we cannot tell which features are worth building on and which are quietly wasting your time.

  • On. Detailed usage statistics through Chromium's built-in telemetry, and crash reports to Sentry (Functional Software, Inc.): a stack trace, device and OS information, log excerpts, and a memory snapshot of the process that failed. A crash dump captures what was in memory at the moment things broke, so it can incidentally contain fragments of page content or text you had typed. We look in it for the fault and nothing else, we keep it for 90 days, and while you are signed in it carries your account identifier (the Auth0 subject ID; Auth0 holds your sign-in credentials and whatever display name you chose, never payment details).
  • Off. Both of those stop, from that moment. A crash then produces a number and nothing else: we learn that one happened, not what you were doing, not what was on screen, and not what was in memory.
  • Why crash reports are worth leaving on. A crash report is the difference between knowing that something broke and knowing why. The failures that genuinely ruin a browser, the ones that lose your tabs or wedge a window, almost always happen on a machine we do not have, in a state we cannot reproduce, and the stack trace is the only witness we get. If you want Phi to become more stable, this is the most useful thing you can hand us. We would rather ask you for it plainly than bury the switch somewhere you will not find it.
  • The baseline, which never switches off. Counts, and nothing more: how many installations turned a feature on, and how many used it. Setting Phi as your default browser, using Spaces, using Profiles are the kind of thing, and features we build later get counted the same way. They go to PostHog. Every one of them is a number with no identifier attached, which is why none of it is personal data and none of it can be traced back to you. In neither position do we profile you, sell any of it, or try to work out who you are.
  • Why the baseline stays on: Phi is free and will always have a free tier. That is not free for us: the AI costs real money in tokens, and we pay that bill every month whether or not a feature turns out to matter to anyone. A handful of anonymous counts is how we find out whether it did, and what to spend the next month building. It is a small thing to ask in return for something we hand over for nothing.
  • If you want a browser that never contacts us at all, you can have one. The macOS client is open source under the Apache License 2.0, so build it from source and you get exactly that, with our blessing. Just know that it then flies solo: no updates from us, no Phi Cloud, no AI tokens on our bill, and when something breaks it stays broken until you fix it. That is a legitimate choice and we will not talk anyone out of it. If you would rather we do the work, this is the trade: we keep it running, patched and paid for, and in return we need to know enough about how it runs to keep it running well.

• IP/device data: Your IP address, device ID, macOS version, and browser version, collected automatically. On its own none of this says who you are. An IP address is where a request came from, a device ID is a random string, and a version number is a version number. We use it for two things: keeping the service secure, which means spotting abuse, rate-limit evasion and fraud, and knowing in aggregate which macOS and browser versions people actually run so we support the right ones. That second use is a count like any other in the baseline. We do not try to work out who you are from any of it, and these logs are kept for 30 days. European law still treats an IP address as personal data, and so do we, which is why it is listed here rather than quietly left out.

What is required, and what is optional:

• An account itself is optional. You can choose "Explore Phi without signing in" on the first screen and use Phi as a plain browser indefinitely, providing no account data at all. AI features, memory, sidebar chat, and Phi Link require an account and are unavailable in that mode.
• Required to have an account: Your email address and a password (or a social sign-in) are contractually required to create and secure a Phi account. If you do not provide them, we cannot create an account for you.
• Required for paid features: Payment data is contractually required if you purchase paid features. Without it, we cannot process the purchase, and the free features remain available.
• What an account gives you: the AI features. Signing in is how you choose them, and one setting switches them all off again.
• Off until you act: data connectors and messaging pairing (Phi Link) do nothing until you connect an account or pair a channel. Your name and avatar, survey responses, and marketing emails are optional in the same way. If you never provide these, the only consequence is that the specific feature does not work; nothing else about Phi is degraded.
• Age is the one thing we must ask. No data you give us is demanded by a specific statute, but we are only permitted to offer Phi to adults, and the only way to honour that is to ask your age before anything else. That is why the age gate comes first and why we keep it to a year of birth.

A note on other people's data (if you are not a Phi user): Someone using Phi can sync or relay content that contains personal data about you, for example the other side of a conversation with the Phi user, email the Phi user received from you, or a document that mentions you. You did not choose this, so here is where you stand:

• What we may hold about you: your name, email address, username or handle, the content of the communication itself, and whatever else the synced or relayed material happens to contain. It can incidentally include sensitive details, for example a health matter or a political opinion mentioned in an email. We do not seek any of it, we do not profile you with it, and we process it only as an inseparable part of the content the Phi user chose to sync. Where such content reaches us, we rely on the fact that it was manifestly made public by you or is strictly necessary to deliver the feature the Phi user asked for, we apply no further processing, and we delete it with the connector or pairing.
• Where we got it: from the Phi user's connected third-party account or paired messaging channel, not from you directly.
• What we do with it: only provide the connector or relay feature to the Phi user who enabled it. Nothing else. It is deleted when the Phi user removes the connector, removes the pairing, or deletes the account.
• Why you did not hear from us: we have no relationship with you and no reliable way to reach you except through the data itself, so contacting you individually would be impossible or disproportionate. This policy is your notice instead.
• Our lawful basis: our legitimate interest in providing the connector and relay features the Phi user asked for. The Phi user's decision to switch a feature on cannot by itself be a lawful basis for processing your data, which is why we keep this strictly limited to delivering that feature and nothing more.
• Your rights: you do not have to wait for the Phi user to act. Write to us (see Contact Information) and exercise your rights directly, including erasure. Because we hold no account for you, we will ask for enough information to verify who you are and to find your data, for example the handle or email address that appears in the content, and never more than that.

How do we use personal data?

• To provide you with Phi:

  • Creating and managing your account.
  • Processing purchases for paid features.
  • Providing AI research, automation, memory, document generation, and translation features.
  • Syncing and relaying the data sources and channels you connect.
  • Fraud protection, security, and debugging.

• To improve Phi:

  • Internal analytics, testing, research, and product development.
  • Diagnosing crashes and errors.
  • Identifying usage trends to improve user experience.

• To stay in touch:

  • Replying to support requests.
  • Sending you updates or news about Phinomenon Inc. (only if you've given us permission).

What are Phi's AI products and features?

Signed in, Phi's AI features are on, and one setting switches all of them off at any time. When they are on, they are orchestrated by Phi Sentinel, a separate component of Phi Browser that is only activated if AI features are enabled. Sentinel is a small menubar application that supervises Phi's AI services on your machine: it manages the local processes, acts as the permission gate for anything sensitive (screen capture, file access, and similar are guided by Sentinel and scoped to folders you choose), and hosts the settings for everything below.

You then have two ways to run AI, and they have very different privacy shapes. You can also use a mixture of both.

• Option 1: Phi Cloud. We make models from OpenAI, Google, Anthropic, or SpaceXAI, plus selected open-weight models, available through our service. The model capacity behind this is purchased by us from OpenAI and OpenRouter.

  • Content flows through, it does not stay. Your input, the necessary context, and the model's output are transmitted to the provider of the model that serves your request. Each provider processes that content under its own API terms, and your use of a given model is also subject to those terms. We act as the intermediary in this exchange: we do not retain the content, and what we record is the metering described above (model, token counts, cost). For this relay we act as your processor: we transmit what you asked us to transmit, on your instruction, and keep none of it. The AI provider is an independent controller for its own handling under its API terms.
  • Fair use. Because tokens are a real cost on a free product, Phi Cloud is subject to the fair-use limits in our Terms of Use. Metering, not content, is how those limits are enforced.

• Option 2: Private AI. If you prefer that we are not in the loop at all, enable Private AI in Sentinel and bring your own AI:

  • Your own API key from a vendor of your choice. Requests then go directly from your machine to that vendor under your own agreement with them. We are not a party to that relationship and receive neither the content nor the metering.
  • Your own models. Sentinel can run models on the same Mac (on Apple Silicon it supervises OpenAI-compatible local servers for chat, background memory tasks, and embeddings, bound to 127.0.0.1 so only your machine can reach them), and it works equally with LM Studio for MLX models or Ollama, whether those run on this Mac or on your own AI rig elsewhere on your network. As long as Phi can reach the endpoint, where the model lives is your choice, and in these modes you manage your own AI infrastructure. Nothing goes to us, and no account is involved.

Your memory is yours. Phi's memory of your browsing is generated and stored on your device and will remain your property. We do not collect it, we do not upload it, and it is not part of either option above. It is also not a black box: you can always see what is in it, edit it, delete entries, or wipe it entirely, because it is on your own disk. Phi is privacy-centric and local-first, and this is the clearest expression of it.

We actively keep your secrets out of it: the AI Guardrail. Being local is not by itself enough, because a browser sees things you would never choose to keep. So Phi Sentinel runs an AI Guardrail, a model that runs on your own Mac and inspects text on its way into memory before anything is written. If it finds a credential, it refuses that write outright: API tokens and authorization values, passwords, private keys, crypto seed phrases, database connection strings that carry credentials, password hashes, and recovery or backup codes. It applies the same check before Phi shows you a generated greeting, so explicit content and sensitive browsing are not surfaced back at you on screen.

Four things about it matter more than the feature itself:

• It never leaves your Mac. The text being checked goes to a service on your own machine over a local connection. It is not sent to us, to Phi Cloud, or to any model provider. Screenshots are never sent to it at all.
• It never edits your data. There is no masking and no silent cleaning. A write is either kept exactly as it happened or refused whole, so Phi does not quietly rewrite your own record of what you did.
• It records nothing about what it found. No matched text, no page address, title or search query, no position in the text. What we can see is how often the guard allowed, refused, or could not run, and nothing about the content of any of it.
• It deliberately leaves ordinary personal information alone. Names, phone numbers, email addresses, places and dates are what make a memory useful and they pass through. This is aimed at secrets, not at your life.

It is the first line of defence, not the only one. Protection here is tiered, and the Guardrail is the layer closest to you: fully local, running before anything is written, and the only one of the three that works on your own machine. Above it, Phi Cloud applies its own online guardrail to the requests it serves, as is standard for a hosted AI service. Above that again, each model carries the safety and alignment work its own vendor built into it. Three layers, and the one nearest your data is the one that never phones home.

We will also be straight about what the local layer cannot do. It reads text, so a secret that exists only inside an image is not something it can catch. If a check cannot complete within the time allowed, the write proceeds rather than failing, because a slow filter must never become a reason your own record of what you did goes missing. And it lets ordinary personal information through on purpose. Treat it as a strong first line that we run at our own cost because we think it should exist, not as a promise that a secret can never reach your disk.

Your control: You can switch AI features off in Phi Browser settings (which also deactivates Sentinel), disconnect individual connectors and pairings, and delete local AI data from your machine at any time. To see or edit what Phi remembers, open the memory page in Phi. The only AI-adjacent data that lives on our servers is what the connector and Phi Link sections above describe, plus that thin agent task metadata.

Who we share personal data with?

We do not sell your personal data. We disclose it only to the recipients described below, or to parties you authorize. Not all recipients have the same role, and we describe each honestly rather than labeling them all the same.

Our processors. These providers process personal data on our documented instructions, under a data processing agreement with us:

Stripe (payments). Stripe, Inc. processes your payment and billing data (collected by Stripe directly). Under Stripe's service terms, Stripe acts as our processor for some of this processing and as an independent controller for other parts (for example fraud prevention, compliance, and improving its own services). Stripe's own privacy policy applies to the processing it controls.

Independent controllers. These recipients decide their own purposes and means for at least part of the processing. They are not our processors under a DPA for that processing, and their own privacy policies apply to it:

We will keep this list current. When we add or replace a service provider that processes personal data, we will update this policy, and we email users about important changes. Our tooling changes as the product does, and services we have stopped using are decommissioned and their data erased. There is a blunt commercial logic to that worth stating plainly: we do not pay for services we do not use, and once we stop paying, the data goes with them.

Your data may also reach other parties, in situations you control or the law compels:

Your Phi memory is exposed as an MCP server on your own machine, so you can share it with another AI tool if you choose. If you do, that is your action and your decision, not a disclosure by us. Beyond that:

• Parties you authorize: Third parties you engage with through Phi, or when you use a social sign-in to create an account.
• Legal obligations: If required by law, court order, or to protect the rights and safety of Phinomenon Inc. or our users.
• Business transfers: If Phinomenon Inc. undergoes a merger, acquisition, or bankruptcy, your data may be transferred to the new owners. We would notify you before your data becomes subject to a different privacy policy.

How to delete or export your data?

Deleting your account and data is something you do yourself, inside Phi. Go to Settings, then Additional Browser Settings, then your name under You and Phi, then Delete account and data. That is the route. You do not need to email us, fill in a form, or ask to be forgotten: the button does it.

Deleting your account permanently removes the account and the data stored with it, and it cannot be undone. On our side that covers every system: your Auth0 account record, analytics events in PostHog, crash reports in Sentry, connector data and OAuth tokens (which we also revoke), messaging pairing records and relayed content, the agent metadata we hold server-side, and payment data not subject to statutory retention. Your Phi memory is not in that list because it was never on our servers: it lives on your Mac, and it goes when you remove it or remove Phi. We issue matching deletion instructions to our processors, and deleted data ages out of backups within 90 days, as described in Data Retention.

Uninstalling and deleting your account are two different things, and you probably want both.

Uninstalling cleans up your Mac. Phi ships an uninstaller, and you do not have to hunt for it. Drag Phi to the Trash and Phi Sentinel notices, then offers to remove what Phi leaves behind: Phi's application support and cache directories for the browser and for Sentinel, and any local AI models that were downloaded. You can also start it from Sentinel directly. It asks you what to remove rather than assuming, and it only ever touches directories that belong to Phi.

Uninstalling does not close your account. It is a local clean-up, so the account and everything listed above stay with us until the account itself is deleted. If you have already uninstalled and now want the account gone, install Phi again, sign in to the same account, and use the button above. Nothing is lost by having uninstalled first, and there is no deadline.

If you have never signed in, there is nothing on our side to delete. Without an account we hold no account data about you at all; the only trace is the short-lived server logs described earlier, which expire on their own within 30 days. Your data is on your Mac, and removing Phi removes it.

Exporting your data. European law gives you the right to receive your personal data in a portable form, and we are building this as a self-service export that packages your data and emails it to you. When it ships, it will appear on the same settings page as deletion. Until then, ask us (see Contact Information) and we will put the package together for you.

International Data Transfers

Phinomenon Inc. is based in the United States, and the service providers listed above process data primarily in the United States. Telegram is an exception: it is an independent service with its own global infrastructure and its own privacy policy; messages you exchange through a paired Telegram channel are processed by Telegram under its own terms, and the transfer safeguards described here apply to our own processors, not to Telegram. If you use Phi from the European Economic Area, the United Kingdom, or Switzerland, your personal data is transferred to the US.

We make those transfers lawful in the standard way: every provider listed above is engaged under a written data processing agreement that includes the European Commission's Standard Contractual Clauses, the legal mechanism for sending personal data outside Europe. Where a provider is additionally certified under the EU-US Data Privacy Framework, that certification applies too. Phinomenon itself is not certified under that framework, which does not change the protection you get, because the contractual clauses cover these transfers either way. You can request a copy of the relevant safeguards by contacting us.

Cookies and Similar Technologies

Our public website, phibrowser.com, uses cookies, pixel tags, and local storage. We group these technologies as follows:

• Strictly necessary: Required for the site to function. These are always active.
• Analytics: Help us understand how visitors use the site. We use Google Analytics (Google LLC) and PostHog (PostHog, Inc.).
• Advertising measurement: Lets us measure whether our advertising works, for example which campaigns lead to downloads. This uses a conversion pixel provided by the advertising platform we are running a campaign on, which records page views and download events and shares them with that platform for attribution. The platforms may change over time; the purpose does not.

Advertising measurement runs only if you accept it. The advertising pixel is set only after you accept via Cookie Settings, and declining stops it. Our own analytics carry no advertising identifiers and are not used to track you across other websites. You can change your choice at any time through the Cookie Settings link in the footer of phibrowser.com, and you can always remove or block cookies in your browser.

These technologies are used on our websites only. Phi Browser itself does not use advertising cookies. Note that Phi Browser does send product analytics and crash reports as described above; those are separate from website cookies and are covered in the sections above.

Permissions Phi Asks For on Your Mac

macOS permissions are granted to an application as a whole, so the prompt you see is often much broader than the use we make of it. We would rather explain each one than let the wording speak for us.

• Full Disk Access. Phi asks for this only so that it can import your existing data from Safari during setup (bookmarks, history, and similar). macOS does not offer a narrower permission for reading another browser's profile, so the prompt it shows covers far more than we use.

  • It is optional. If you decline it, or grant it and later revoke it, Phi keeps working normally; you simply cannot use the Safari import.
  • We do not read your Mail database, Messages, Time Machine backups, or any other file this permission technically exposes. The import reads the browser data you asked us to import, and nothing else. None of those other files are read, copied, transmitted, or stored by us.
  • You can revoke it at any time in System Settings, Privacy and Security, Full Disk Access.

• Importing from other browsers. Importing from Chrome, Arc, or a bookmarks file does not require Full Disk Access.

• Other permissions (such as notifications, microphone, camera, or screen recording) are requested only when you use a feature that needs them, and are handled by macOS in the usual way.

Passwords, and the agent vault

Phi has no password manager of its own, and that is deliberate. Chromium ships one, and we took it out. On its own it is local only: no sync between your devices, no breach monitoring, nothing watching for a password that has turned up in a leak. The part that makes it genuinely useful is Google Password Manager and Chrome Sync, and those are Google's own services, which Google restricted to Google Chrome in March 2021. No third-party browser can offer them. We would rather remove a half-equipped password manager than ship one and let you assume you were covered when you were not.

So use a real one. If you have no strong preference, iCloud Passwords is already on your Mac, it syncs across your Apple devices, and it warns you about compromised passwords. If you already use 1Password, Bitwarden, Dashlane or anything else, keep it; their extensions run in Phi like they do in any browser.

Whichever you pick, your passwords are not ours. We do not store them, sync them, see them, or send them anywhere. A password manager is a separate product from a separate company, and its privacy policy and terms govern what it does with your data, not this policy. We provide the browser it runs inside, and nothing more.

The agent vault. Phi includes an optional feature for a specific problem: when an AI agent works on your behalf, it sometimes reaches a sign-in page, and the alternative to a vault is you pasting a password into a chat. It is off unless you turn it on, and it works like this:

• The vault is Bitwarden's, not ours. You sign in to your own Bitwarden account. Bitwarden is the provider of that service and its own terms and privacy policy apply to it. You choose which server holds your vault: bitwarden.com (United States), bitwarden.eu (European Union), or your own self-hosted Bitwarden. That choice decides where your vault lives and whose law reaches it, so it is worth making deliberately rather than accepting a default. If you self-host, your vault never touches Bitwarden's servers either.
• Your vault never comes to us. No part of it is sent to Phinomenon, stored by us, or synced through our services. We never see your master password.
• Phi itself does not hold the keys. The vault is unlocked inside a separate helper process, and the unlocked key and all vault decryption stay there rather than in the browser. The two talk over a private channel with no name on the filesystem, so no other program on your Mac can connect to it, and your session is held in the macOS Keychain locked to Phi's own signature.
• The agent gets one credential at a time, and only when you say so. It never receives the vault, or a list of what is in it. It asks for one credential, at the moment it needs it, you approve that request, and the access is written to an audit log you can read.

If you never enable the agent vault, none of the above happens and no Bitwarden connection exists.

Data Security

We aim to protect your personal data through a system of organizational and technical security measures, including encryption in transit, access controls, and encryption at rest, least-privilege access, and regular dependency and vulnerability review. However, no method of transmitting data over the internet is 100% secure. You should help protect your data by selecting a strong password and limiting access to your device.

If you believe you have found a security vulnerability in Phi, please report it to us (see Contact Information).

Data Retention

We keep personal data only for as long as necessary for the purposes described in this policy. Our standard retention periods are:

These periods are our operating defaults; where a law requires us to keep something longer (for example financial records), the legal period applies.

Backups: Our systems are backed up on a rotating schedule. When data is deleted from our live systems, it ages out of backups automatically within 90 days. Backups are not used to restore data that you have deleted, except where restoring an entire system after a failure makes it temporarily reappear, in which case we re-delete it.

When we delete your account, we also instruct our processors listed above to delete the associated personal data they hold for us, in line with our data processing agreements.

Children's Personal Data

Phi is for adults. We do not knowingly collect personal data from children under 18, Phi is not directed at them, and the sign-up age gate described above refuses account creation to anyone under 18. If we learn we have collected personal data from a child under 18, we will delete that information as quickly as possible. If you believe this has happened, write to us (see Contact Information).

U.S. State Privacy Rights

If you reside in certain U.S. states (such as California, Colorado, Connecticut, Texas, Utah, Virginia, and others), you have specific rights regarding your personal data.

• Access: You have the right to confirm if we are processing your data and access a copy of it.
• Deletion: You can ask us to delete the personal data we have collected. You can also do it yourself in Phi; see "How to delete or export your data".
• Correction: You can ask us to correct inaccurate personal data.
• Portability: You can ask for a copy of your data in a machine-readable format.
• Right to limit sensitive personal information: To the extent we process sensitive personal information (such as account sign-in credentials), we use it only to provide the service, and you may ask us to limit any other use.
• Non-discrimination: We will never discriminate against you (for example by degrading the service or charging more) for exercising any privacy right.
• Authorized agents: You may designate an authorized agent to submit requests on your behalf; we will verify the agent's authority and your identity.
• Global Privacy Control (GPC): Some browsers send an automatic "do not sell or share my data" signal. We honour it. If your browser sends GPC, we treat it as an opt-out of advertising measurement and of any sharing for cross-context advertising, and you do not have to do anything else. You can still set your preferences by hand, and a choice you make yourself takes precedence over the signal.

Right to appeal a refused request (Colorado, Connecticut, Virginia, Texas, and other states with appeal rights). If we decline to act on your request, we will tell you why in writing. You may appeal our decision by replying to our response, or by writing to us with the subject line "Privacy request appeal", within a reasonable period after receiving it. We will respond to your appeal in writing within the timeframe your state's law requires, explaining the reasons for our decision. If we deny your appeal, our response will include a way for you to contact your state Attorney General (for example through the AG's online consumer complaint portal) to submit a complaint.

California disclosures (CCPA/CPRA). In the preceding 12 months we collected these categories of personal information, from you directly and automatically from your device, and from third-party sources you connect: your connected third-party accounts (connector sync), messaging platforms you pair (Telegram relay), and social sign-in providers, for the business purposes described in "How we use personal data," shared with the recipients named in "Who we share personal data with," and retained per the Data Retention table:

"Selling" or "Sharing": We never sell your personal data, and no money changes hands for it. We should flag one technicality: some U.S. state laws define "sharing" so broadly that running an advertising pixel counts, even though nothing is sold. That is the only thing in this policy that could fall under that definition, it only happens if you accept cookies, and you can turn it off at any time. You can opt out at any time via the "Do Not Sell or Share My Personal Information" link in our website footer, by rejecting cookies in the consent banner, or by sending a GPC signal from your browser, which we honour automatically.

To exercise these rights, contact us (see Contact Information).

European Data Subject Rights (EU, UK, and Switzerland)

If you live in the European Union, the United Kingdom, or Switzerland, you have rights under the GDPR, the UK GDPR, and the Swiss FADP. Phinomenon Inc. is the controller of your personal data.

Lawful basis for processing. We only process your data when we have a lawful basis:

• Contractual necessity: Account data (creating, securing, and managing your account), payment data (processing your purchase), and the AI features, which are what the account is for: signing in is how you ask for them, so the basis is our contract with you rather than your consent. This covers AI content relayed through Phi Cloud when you invoke a feature, transmitted to generate the response and not retained. We do not stretch this basis to the optional things below.
• Legal obligation: Retaining payment and transaction records for tax and accounting purposes.
• Consent: The things that cannot happen until you act. Data connectors and messaging relay (Phi Link) stay off until you connect an account or pair a channel. You are shown what that involves first, the act of connecting or pairing is your consent, and disconnecting or unpairing withdraws it. Consent is also the basis for analytics and advertising cookies on our websites and for marketing emails. In-app product analytics and crash reporting run on the switch described above, which lives in Settings and can be changed whenever you want. You may withdraw any consent at any time, as easily as you gave it, without affecting prior processing and without any detriment beyond losing the feature itself.
• Legitimate interest: Crash and diagnostic data outside the EEA/UK/Switzerland scope above (our interest: fixing defects and keeping Phi reliable), feature usage data where consent does not apply (our interest: understanding and improving the product), and IP/device data and security logs (our interest: securing the service and preventing abuse). You can object to processing based on legitimate interests (see Your Rights). For telemetry the control is the switch itself: it belongs to that installation of Phi rather than to your account, so turning it off there is what stops it. The anonymous baseline is not personal data, so there is nothing in it attached to you to object to. You can ask us for a summary of our balancing assessment (see Contact Information).

What you are required to provide: see "What is required, and what is optional" above; in short, only the minimum account data (and payment data for purchases) is contractually required, everything else is optional, and the only consequence of not providing optional data is that the related feature will not work.

Your rights. You have the right to:

• Access (Art. 15): Get confirmation of what we process about you, a copy of it, and the actual recipients it has been disclosed to.
• Rectification (Art. 16): Have inaccurate data corrected.
• Erasure (Art. 17): Have your data deleted. You can do this yourself in Phi; see "How to delete or export your data".
• Restriction (Art. 18): Have us pause processing of your data in certain situations, for example while a dispute about accuracy is resolved.
• Portability (Art. 20): Receive the data you provided to us in a structured, commonly used, machine-readable format, and have it transmitted to another controller where feasible.
• Objection (Art. 21): Object to processing based on legitimate interest, and object at any time to any direct marketing.
• Withdraw consent (Art. 7(3)): For any processing based on consent.
• Complaint (Art. 77): Lodge a complaint with your local supervisory authority (in the EU, the data protection authority of your member state; in the UK, the ICO; in Switzerland, the FDPIC). We would appreciate the chance to address your concern first, but you do not have to contact us before complaining.

Rights of other people (Art. 15(4), Art. 20(4)): Connector-synced content and relayed messages usually contain personal data about people other than you. When we fulfil an access or portability request that covers such content, we will redact or exclude other people's personal data where disclosing it would adversely affect their rights and freedoms, and we may refuse the affected part of a request on that ground. We will tell you when we have limited a response for this reason and why.

Automated decision-making: We do not make decisions based solely on automated processing that produce legal or similarly significant effects about you.

How we handle requests: Write to us at any address in Contact Information. We respond within one month (extendable by two further months for complex requests, in which case we will tell you why). Exercising your rights is free of charge, unless a request is manifestly unfounded or excessive (for example repetitive), in which case we may charge a reasonable fee or refuse the request as Art. 12(5) allows, explaining why. To verify your identity we ask only for what is proportionate to the request and the risk it carries (Art. 12(6)), and we will not demand excessive documentation. For access, correction, restriction, and similar requests, a request sent from the email address on your account is normally sufficient for most requests; for requests that would export or delete large bodies of synced third-party content (connector archives, message relays) or delete your account, we additionally require you to confirm through a signed-in session or a second confirmation loop, because the stakes of impersonation are higher. For destructive or irreversible requests, above all deletion, we apply stronger verification: we confirm the request through a verification email loop to your address on your account (deletion proceeds only after you confirm), we require you to re-authenticate in Phi where your account still exists, and where we still have reasonable doubts about your identity we may ask for proportionate additional evidence.

Our EU and UK representatives: because Phinomenon Inc. is established outside Europe, we have appointed a representative in each region under Article 27, so you can raise anything in this policy locally rather than writing to the United States. Both are listed under Contact Information. Using them is your choice, not a requirement.

Contact Information

If you have any questions about this Privacy Policy or your rights, please contact us:

Email:

• General inquiries and security reports: hi [at] phi.cc
• Privacy inquiries and rights requests: privacy [at] phi.cc
• Data protection contact: dpo [at] phi.cc

Mailing address (US):
Phinomenon Inc., Attn: Data Protection Contact
324 S Diamond Bar Blvd, Unit Num 717
Diamond Bar, CA 91765, United States

UK representative (UK GDPR Article 27):
Fyde Innovations Ltd
124 City Road, London, EC1V 2NX, United Kingdom

EU representative (GDPR Article 27):
BizLegal Limited, trading as EU Rep
27 Cork Road, Midleton, Co. Cork, Ireland (company number 635921)
Appointed under Article 27 as the statutory contact point for EU supervisory authorities and data subjects. For anything in this policy, the direct route is the email addresses above; the representative exists so that a local point of contact is always available.