Privacy Policy

1 Who we are and what this covers

This Privacy Policy explains how the person who operates this workspace (the "Operator", "we", "us" or "our") collects, uses, shares and protects information about people who use FLUXBOT. FLUXBOT is the private AI chat workspace at fluxbot.cc and cardinalgo.org, which people join by invitation or through another sign-up option we provide, together with any backup address that runs the same workspace (the "Service"). Every address uses the same accounts and data.

The Operator is a private individual, or that person's small business, who owns the workspace and runs the Service for themselves and for people who join by invitation or through another sign-up option we provide. It is free and it is not a company product. The Operator decides how your information is used; in data protection terms, the Operator is the "controller".

The Operator can give other people the Admin role to help run the workspace. Admins can see the information described in section 4, and admins agree in the Terms to use it only to help run the workspace and to keep it confidential.

This policy doesn't cover the outside companies the Service uses. They have their own privacy policies (see sections 5 and 6).

This policy is a notice about our practices. It is not a contract and does not create contractual rights or obligations. Your use of the Service is governed by our Terms of Use.

2 Information we collect

2.1 Account information

• Your username, display name and role (Admin, Inviter or Member). Your display name (or your username, if you haven't set one) is sent to the AI with every message, so consider using a first name or nickname.
• Your password is stored only as a salted scrypt hash, never in plain text, and admins can't look it up in the app. Admins never see a password you choose. If an admin creates your account or resets your password, they see the temporary password they set; the app will usually ask you to choose your own the next time you sign in, and if it doesn't, change it yourself. Like any website, your password passes through our hosting and network providers when you sign in.
• When your account was created, and who invited you or, if you joined through another sign-up option we provide, that you joined that way.
• Settings admins control for your account: your daily message, picture and web-search limits, how many crew bots you can make, your assigned AI model and whether your account is paused.
• If you turn on 2-step login: the records described in section 2.8.
• Which version of the Terms and Privacy Policy you accepted, and when. The Operator can turn this acceptance step on or off; while it is off, nobody is asked to tick a box.
• If you create invite links: the note you add, when each link expires, and who joined through it and when.

2.2 Your chats and content

• Chats:
  • chat titles and your messages;
  • the AI's replies, including any "reasoning" text the model returns;
  • which model answered.
• Organization: 👍/👎 votes, pins, folders and share links.
• Uploads: images you attach (up to 4 MB each) are stored with their file name, type, size and upload time. Documents are read in your browser; only their file name, size and extracted text are stored, inside the chat message. The original document file is not uploaded.
• AI pictures: pictures you ask the AI to make are stored like uploaded images, with the chat they belong to. They count toward your storage and are deleted with that chat. The picture's description (the one you typed, or the one the AI wrote from your conversation) is kept in the chat, not in the picture file.
• Web search: when an answer used the web, the chat keeps up to 10 links to its sources, with their page titles. We never visit those links ourselves.
• Projects: if you use Projects, each project's name, icon, colour, instructions and settings, and the text of the files you add (PDF, Word or text files, read in your browser), are stored until you delete them or your account is removed. Only the extracted text of a file is kept, not the original file. Chats in a project are marked as belonging to it.
• Crew bots. If you make a bot, FLUXBOT stores its name, emoji, optional picture, description, instructions, conversation starters, the text of any reference files you add (read in your browser; only the text is kept), whether it is shared, and how many times other people have started a chat with it. They are kept until you delete the bot or your account is removed.
  • If you share a bot with the crew, everyone in the workspace can see its name, picture, description, starters, your name and username, and its use count, and chat with it.
  • When someone chats with a bot, its instructions and the most relevant parts of its reference files go to the AI provider with their messages, and the AI may repeat them if asked.
  • You never see other people's chats with your bot, or who used it — only the count. To count a use at most once a day per person, the Service keeps a one-way code made from the person, the bot and the day for about 2 days.
  • When you chat with someone else's bot, your saved memories and custom instructions are not sent with your messages. If you use web search in that chat, the search is made without the bot's instructions, so the bot can't choose what is searched for or which pages are opened; when Groq's search answers, the reply comes from FLUXBOT itself, without the bot.
  • A chat with a bot keeps the bot's name and emoji, even after the bot is deleted.
  • On this device, the app remembers the last few bots you opened, for shortcuts on the new-chat screen (section 13).
• Code previews: not stored. When you press ▶ Preview, the code runs only in your browser, inside a locked sandbox that can't see your account, and is never sent to our server. If you use "Open in new tab", that browser tab keeps a copy of the code until you close it, so it can reload.
• Memories: short facts you ask the Service to remember, and any an admin adds to your account (see section 4). They are added to the instructions of every chat. A message that starts with words like "remember that…" or "don't forget…" is saved as a memory automatically. You can turn memory off in Settings, which stops saving memories and sending them to the AI.
• Settings: custom instructions, theme, voice choice and similar preferences.

2.3 Voice

• Speaking a message. When you use voice input, your recording is sent to Groq to be turned into text. The Service does not store the recording; the resulting text becomes part of your message. Your browser asks for microphone permission, and the microphone is used only when you start voice input.
• Read-aloud. Reply text is either sent to Groq to create speech or read by your browser's built-in voice. Some built-in voices are processed online by your browser or operating system maker, under their own policies.

2.4 Security and usage records (collected automatically)

• Activity log and sign-in history. Each entry has the time, the action (such as signing in, changing a password, an admin opening a chat or an admin changing something in your account), who did it (and to whom) and the IP address. Sign-in entries also record the device type (for example "Chrome on Windows") and approximate location (city, region or country), which comes from the location headers Netlify and Cloudflare add based on your IP address. Some entries include a chat's title (when it's shared or opened by an admin), a project's name (when an admin opens it), a crew bot's name (when you make, change, share or delete a bot, when you report one with the reason you picked, and when an admin changes one, with the reason for hiding it, or opens a private one or one of its files, with the file's name) or the words an admin searched for. A bot's instructions and files are never written to the log, and neither is who used a bot. Other entries note when you turn phone alerts on or off for a device (with the device type), 2-step login events (section 2.8), and when a web search was busy and your question was answered without the web (which search services failed, never your question).
• Failed sign-in attempts, with similar details (including wrong 2-step codes), plus short-lived rate-limit and lockout counters.
• Signed-in devices: device type, approximate place, IP address, and when each was first and last seen. We also keep your last-seen time.
• Daily counts per person of messages (per model), voice use, pictures and web searches.
• Workspace statistics: message counts per person, model and mode per day (all chats with crew bots count together as one "crew bots" mode, never per bot), plus daily totals of pictures, blocked picture descriptions and web searches, and how long they took. These totals contain no text.
• Reply feedback: your vote, any comment you add, and a 300-character snippet of the reply.
• Content-filter records. If an admin turns the filter on and it blocks or flags one of your messages or picture descriptions, we keep your name, which chat it was in, the matched words, whether it was blocked or flagged, and the first 400 characters of your latest message or of the picture description. A picture description stopped by the built-in safety check or by the picture service's own safety model (both always on, for everyone) is recorded the same way. If the filter is in flag mode and you save a crew bot, or a bot file, that matches it, we keep your name, the bot, the matched words and the first 400 characters of the matching text.
• Notifications admins send you, including replies to your support requests, and whether you've read them. Admins also keep a list of the last 50 notifications sent, with their text and recipients. The Service also sends automatic notifications about your account's security, for example a failed 2-step check, a recovery code being used, sign-in being paused after too many attempts or, for admins, a sign-in from a new network (with the device type and approximate place).

2.5 Support requests and problem reports

When you ask for help or report a problem, for example with the Report a problem button, the Service stores a support request (a "ticket") with:

• your message, and the kind of request (bug, question, feedback or other);
• your username and display name, and when you sent it;
• whether it came from the button or from a chat, and which chat, if any;
• the page you were on and the last error message the app showed you, if there was one;
• what you said: for a report filed from a chat, your last message in that chat (up to 1,000 characters);
• your device type (for example "Chrome on Windows"), worked out from your browser's user agent;
• its status (open, being worked on or closed) and any replies.

Reports the AI files for you. While chat reports are turned on (the default), the AI is told that if you say something in FLUXBOT itself is broken, or ask to reach the team, it should end its reply with a one-sentence summary of the problem. The app then sends that summary to the admins as a ticket, together with your last message, the chat it came from and the last error shown, and a note under the reply tells you it was sent. The AI decides when to do this, so it may file a report you didn't mean to send, or miss one you wanted; you can ask an admin to delete a ticket. Admins can turn chat reports off.

Reporting a crew bot. Reporting a bot sends a support request with the bot's name, its maker's username, the reason you picked and anything you wrote. If you tick "Include this chat", admins can also open that chat, but only while "Admins can read chats" is on. Like every support request, it goes to the admins, not to the bot's maker (unless the maker is an admin).

Who sees tickets.

• All admins can read every ticket, including any message quoted from your chat, even while "Admins can read chats" is off.
• Admins get a notification with your display name and the start of your message, and the activity log records the first 80 characters of it.
• Replies reach you in your 🔔 notifications. They are signed "FLUXBOT Support" or "Admin 001" and so on, never with an admin's real name.
• You can see your own tickets and the replies to them.

Email. If you email the Operator, for example at the support address shown in the app or the contact address in section 16, the Operator receives your email address, your message and any attachments through the Operator's email provider. Those emails are kept in the Operator's mailbox, outside the Service.

2.6 Information our providers receive when you connect

• Netlify and Cloudflare process your IP address, browser user agent, the addresses you request and other technical request data, so they can deliver and protect the Service.
• Google Fonts receives your IP address and user agent when your browser loads fonts from it. The app's own code is served from this site itself, except Cloudflare's Turnstile script, which the sign-in page loads from Cloudflare only if the Operator turns Turnstile on.
• Cloudflare Turnstile. If the Operator turns on Turnstile bot protection, Cloudflare checks signals from your browser to tell people from bots, and the Service sends Cloudflare your IP address with the check.
• Public code hosts, only for code previews. When you preview code that uses outside libraries or fonts, your browser downloads them directly from jsDelivr, cdnjs (run by Cloudflare), unpkg, esm.sh, Tailwind's CDN or Google Fonts. Those hosts see your IP address and browser type, as with any website, but not which page you came from. The code itself isn't sent to them, except that esm.sh receives the names of the code packages a preview imports.
• Websites you open from an answer. Source links under web answers are ordinary links. If you open one, that website sees your visit like any other, under its own policies. The link is opened without telling the site which page you came from.

2.7 Phone and desktop alerts (optional)

Alerts are off until you turn them on for a device in Settings → Notifications. Your browser then asks for permission.

• What we store for each device: the push address your browser gives us and its two encryption keys, a device description (for example "Chrome on Android"), when alerts were turned on and last delivered, a failure count, and which sign-in turned them on. You can have alerts on up to 10 devices. We also store which kinds of alerts you want, and a short record of which account a browser's push address belongs to.
• What an alert says: only what kind of thing happened, for example "🛟 Support replied to your report", "🔐 Security alert" or "📣 New message from FLUXBOT". It never contains your chats, messages, support requests, replies, the text of notifications, names or email addresses. It also carries your unread count for the app badge and which page of this site to open.
• How it is delivered: through your browser maker's push service: Google Firebase Cloud Messaging (Chrome and most Android browsers), Apple (Safari, iPhone and iPad), Mozilla (Firefox) or Microsoft Windows Push Notification Services (Edge). The alert's contents are encrypted so that only your device can read them. The push service sees that an alert was sent, its size and time, and our server's address.
• When alerts stop: when you turn them off, sign out on that device, sign out everywhere, change your password, when your account is paused, when 2-step login is turned on or reset, and when the push service says the address no longer works. Device records are deleted when your account is removed. They are never included in workspace backups.

2.8 2-step login (optional)

Anyone can turn on 2-step login in Settings → Account. The Operator can require it for admins.

• What we store while it is on: the secret key for your authenticator app (encrypted), when you turned it on, the time step of the last code you used (so a code can't be used twice), and your 10 recovery codes, stored only as one-way (scrypt) hashes and marked when used.
• Short-lived records: a half-finished setup is kept for 10 minutes. After you type the right password, a one-time sign-in ticket, stored only as a hash, lasts 5 minutes.
• Who sees the secret: it is shown to you once, during setup. It is never shown to admins, never included in account copies, logs or ordinary backups, and it leaves the server only in a sealed form if the Operator chooses to include it in a backup (section 4).
• Security records: turning 2-step login on or off, a reset, a recovery code being used (with how many are left), new recovery codes, and failed 2-step checks (with the time, IP address and approximate place) are recorded in the security log. You get a notification when someone types your password correctly but then enters a wrong 2-step code three times, and when a recovery code is used. For admin accounts, the Operator is told too.

2.9 What we don't collect

• Payment details.
• Analytics, advertising identifiers or tracking pixels.
• Precise (GPS) location.
• Your contacts, or your email address unless you email the Operator.
• Information about you from data brokers or other websites.

If other people mention you in their chats, that information is stored in their chats and handled as their content.

3 How we use it, and why

We don't use your information for advertising or profiling, or for automated decisions that have legal or similarly significant effects on you. The Operator does not train AI models on your content. Section 6 explains what the AI providers themselves may do.

4 What admins can see and do

• Chats and projects. While the workspace setting "Admins can read chats" is on (the default), admins can:
  • see a list of everyone's chats and their titles;
  • open and read any chat, read-only, including AI replies, uploaded images, AI pictures with their descriptions, and web sources;
  • search for words inside the messages of everyone's recent chats, and see short excerpts around the matches;
  • see which project a chat belongs to, and open your projects: their instructions and the text of their files.
  You are told about this on the sign-in and sign-up page, when you first sign in, and under the message box. Opening a chat, opening one of its images, opening a project or one of its files, and searching inside messages are recorded in the activity log (kept 90 days). Seeing the list of chats and filtering it by title or name are not. When the setting is off, admins can't list, open or search chats, or open projects, through the admin tools. If it is turned back on, all existing chats and projects become readable again, including ones written while it was off.
• Crew bots. Admins can always see bots that are shared with the crew, including their instructions and reference files, and can hide, edit, make private or delete any bot. Private bots can be opened by admins only while "Admins can read chats" is on, and each time is logged. When an admin changes your bot, you get a notification that doesn't say which admin did it. Admins see how many times a bot was used, never who used it; they can read people's chats with a bot only as described above for chats.
• People and security. Admins can see:
  • names and usernames, and roles;
  • last-seen times and usage counts, including pictures and web searches;
  • whether you have 2-step login on, since when and how many recovery codes you have left (never your secret key or codes);
  • signed-in devices, IP addresses and approximate locations;
  • sign-in history;
  • reply feedback and content-filter records, including blocked or flagged picture descriptions;
  • support requests (section 2.5), whatever the "Admins can read chats" setting;
  • invite links and who joined through them;
  • storage used.
• Actions. Admins can:
  • pause, sign out, reset or remove accounts, and change names, roles, limits and models;
  • reset someone's 2-step login, which signs that person out everywhere (only the Operator can do this for admins and inviters);
  • answer, close and delete support requests;
  • send notifications;
  • turn features off, or put the workspace in lockdown;
  • sign everyone out;
  • set the content filter and "safe mode";
  • give the AI workspace-wide instructions and modes;
  • delete old or unused images to free up storage.
• Account tools. Admins can manage what is inside other people's accounts (only the Operator can change the Operator's own account). They can:
  • rename or delete your chats (deleting a chat also removes its images, AI pictures and share link);
  • delete your uploaded images;
  • view or delete your projects (deleting a project keeps its chats);
  • see your crew bots (private ones only while "Admins can read chats" is on), hide them or delete them, and set how many you can make;
  • view, add or delete your memories;
  • view or reset your settings, including your custom instructions;
  • clear your notifications and reset your message count for the day;
  • set a new password for you;
  • wipe everything in your account (chats, uploads, pictures, projects, crew bots, memories, settings, notifications and daily counts) while keeping the account itself, its signed-in devices, its alert devices and its 2-step login;
  • download a copy of everything in your account (account details, chats, projects with their file text up to 3 MB in total, your crew bots with their reference-file text up to 1 MB in total, memories, settings, devices, the device types that get your alerts, notifications and a list of your uploads).
  Viewing your memories, settings, projects and private crew bots, and downloading a copy of your account, work only while "Admins can read chats" is on. Deleting works either way. A downloaded copy is a file on that admin's own device, which the Service can't control or delete.
• How admins appear to you. You are never told the name of the admin who did something.
  • Notifications and support replies from the Operator appear as "FLUXBOT" or "FLUXBOT Support", and so do automatic notices such as a support request being closed. Notifications and replies from other admins appear under a number, such as "Admin 001" or "Admin 002".
  • Most changes admins make to your account, such as the account tools above (other than password changes) and changes to your role, limit or model, are not announced to you. If an admin other than the Operator resets or sets your password, you get a notification from their "Admin" number. When the Operator does it, you are not notified.
  • Changes admins make to your account are recorded in the activity log (kept 90 days) under the admin's real username, which the Operator and other admins can see.
• Backups. Only the Operator (the workspace owner account) can download or restore a full backup of the workspace.
  • While "Admins can read chats" is on, a backup includes everyone's chats, projects (with their file text), crew bots (with their file text), memories and settings, and optionally uploaded images, AI pictures and bot pictures. While it is off, chats, projects, crew bots, memories, settings and images are left out. A very large backup may leave out some images, project files and bot files to stay within the host's size limit.
  • Backups always include accounts (with password hashes, never passwords), invite links, share links, notifications, support requests, reply feedback, content-filter records, statistics, workspace settings, failed sign-in records and the activity log (with IP addresses).
  • Optionally, a backup can include the workspace's AI provider API keys.
  • Optionally, a backup can include 2-step login records. They are then sealed with a key derived from the workspace's encryption key, so only this workspace can restore them.
  • Backups don't include device lists, phone-alert devices, daily usage counts or last-seen times.
  Each download is logged. A downloaded backup is a file on the Operator's own device. The Service can't control or delete it, and it is kept for as long as the Operator keeps it. Restoring an older backup can bring back content, including content you deleted after that backup was made.
• The Operator. The Operator runs the storage and holds the encryption key, so the Operator could technically access stored data even when chat viewing is off. For example, this could happen during maintenance, restoring a backup, investigating abuse or responding to a legal requirement.

5 Who receives your information

5.1 Service providers

These companies process information so the Service can work. They follow their own terms and privacy policies.

5.2 Other people in the workspace

• Admins can see what section 4 describes, including your support requests.
• If you create a share link for a chat, signed-in members who open it see that chat (including its AI pictures and web sources) and your display name, until you turn the link off. A shared chat never shows which project it belongs to, or that project's instructions or files.
• If you create an invite link, anyone who opens it sees your display name and any note you add.
• If you share a crew bot, everyone in the workspace sees its name, picture, description and conversation starters, your display name and username, and how many times it has been used, and can chat with it (section 2.2). They don't see your chats with it, and you don't see theirs.

5.3 Legal reasons and safety

We may disclose information if we believe in good faith that it is needed for one of these reasons:

• to comply with a law, subpoena, court order or other legal process;
• to enforce our Terms;
• to detect or stop fraud, abuse or security problems;
• to protect the rights, property or safety of the Operator, users or others.

Apparent child sexual exploitation will be reported to the National Center for Missing & Exploited Children.

5.4 If the workspace changes hands

The Operator may hand the workspace over to another person or organization, or may become unable to run it. In that case, the information may pass to whoever takes over. We will ask them to keep following this policy for existing information, or to tell you about any changes and ask you to accept them.

5.5 With your permission

We may share information in other ways when you ask us to or agree to it. Other than as described in this policy, we don't give your information to anyone else.

6 What AI providers do with your messages

What is sent. Each time you send a message, the chosen AI provider receives all of the following:

• your message;
• recent chat history;
• your display name (or your username if you didn't set one);
• the system instructions: the workspace name, today's date, the workspace instructions admins set, the selected mode, the safe-mode instruction if it is on, the instruction for filing problem reports (with the support address) while chat reports are on, short notes about the app's features (pictures, web search, code previews), and your custom instructions;
• your saved memories, unless you turned memory off (or turned it off for that project);
• in a project chat: the project's instructions and the parts of its files that best match your latest message;
• in a chat with a crew bot: the bot's name and instructions and the parts of its reference files that best match your latest message;
• any attached document text and images.

Crew bots. A chat with a bot someone else made leaves out your custom instructions, your saved memories, the problem-report instruction and the notes about the app's features, so the bot's maker can't reach them through the AI. A chat with your own bot is sent like a normal chat. Bot text goes to the same AI providers as any other chat text; there is no new outside company.

If "auto-switch" is on and the first provider fails, the same content is sent to backup providers, which may include a provider you didn't choose. One message can therefore reach more than one provider.

Pictures. When you make a picture, only its description goes to Cloudflare's Workers AI: the words you typed, or a description the AI wrote from your conversation when it offered to make one. Descriptions are checked first: a built-in safety check that is always on, the workspace's content filter and safe mode can stop a description before anything is sent, and then a safety model at Cloudflare reads it before the picture is made. There is no automatic check of the finished picture.

Web search. When you turn on web search for a message, it always goes to Groq's GPT-OSS models with Groq's built-in browser search, even if you picked a model from another provider. Groq receives a shorter version of the usual information: the system instructions, up to your last 15 memories, up to 3,000 characters of project text, and up to your last 8 messages (at most 6,000 characters in all), without images. The AI writes search queries from it, and Groq's search partner Exa receives those queries. If Groq's search is busy and the Operator has turned on the backup search, Tavily receives your latest message as a search query, and its results go to your usual model with the rest of the chat. If no search works, your usual model answers without the web and the answer says so. Search results come from public web pages, which can be wrong, out of date or written to mislead, so the app never acts on instructions found in a web answer.

Projects. In a project chat, the project's instructions and the most relevant parts of its files go to the AI with every message, up to a limit for each provider. Files are read in your browser, and only their text is stored and sent.

Your own AI. The Operator may connect an AI model that runs on their own computer. It shows up in the model menu under the name the Operator gives it, such as "My PC", and auto-switch can use it as a backup. A message answered by that model is processed on that computer, not by an outside AI company. It gets there through an encrypted tunnel, usually Cloudflare Tunnel. The bridge software that links the computer to the Service doesn't save messages, but the Operator controls that computer and the AI software on it, which may keep its own logs.

The summaries below come from each provider's own pages as checked on September 24 and 25, 2026. Providers can change their terms at any time, and the Operator does not control them.

Groq

• Retention. Groq says it does not keep inputs or outputs by default. It may log them temporarily, for up to 30 days, to fix reliability problems or investigate suspected abuse. Data it keeps is stored in Google Cloud in the United States.
• Training. Groq's Services Agreement says it won't use inputs or outputs to train models without the customer's permission.
• Zero Data Retention. Groq offers this as an account setting.

Google (Gemini API, free tier)

• Product improvement. Google's terms say that for unpaid services it uses the content submitted and the responses generated to provide, improve and develop its products and machine-learning technologies.
• Human review. Human reviewers may read, annotate and process inputs and outputs, after Google disconnects them from the account.
• Sensitive information. Google tells developers not to send sensitive, confidential or personal information to its unpaid services.
• Retention. Google also keeps prompts and outputs for 55 days to detect misuse. It states no retention period for data used to improve products.

OpenRouter and its upstream hosts

• OpenRouter itself. It says it doesn't store prompts or responses unless the account owner opts in to logging, and doesn't use them for training. It does keep request metadata such as token counts and timing.
• Upstream hosts. OpenRouter passes requests to the host that runs the chosen model. For some free models, the host logs prompts and responses or uses them to train its models. Policies differ by model and change over time.
• Settings. OpenRouter account settings can exclude hosts that may train on data. Even so, assume that a free model's host may keep your messages.

Cloudflare Workers AI (pictures)

• Training. Cloudflare says it does not use customer content sent to Workers AI to train AI models or to improve Cloudflare's or other companies' services, unless the customer explicitly agrees.
• Storage. Cloudflare says inputs and outputs are stored only if the customer uses its storage services for them, which the Service doesn't. The finished picture is stored by the Service itself (section 2.2).

Exa (web search through Groq)

• Product improvement. Exa's privacy policy says search queries are used to improve its products, including training and fine-tuning its models, and it gives no retention period. Queries that reach it through Groq may be covered by Groq's own agreement with Exa, which the Operator can't see. Assume the search queries may be kept and used this way.

Tavily (backup web search, only if turned on)

• Retention and use. Tavily's privacy policy says it keeps query data for as long as needed to provide and improve its services, and may use parts of it to improve its answers to future queries.

7 No selling for money, no ads

• No selling for money, no ads. We don't sell your personal information for money and we don't share it for targeted advertising. Be aware: some AI providers we use on their free tiers, namely Google's Gemini API and some model hosts reached through OpenRouter's free models, may use what they receive to improve their own products as a condition of offering the service for free. The search services behind web search (Exa through Groq, and Tavily if the Operator turns it on) say they may do the same with search queries (see section 6). Some privacy laws could treat that as a "sale" or "sharing" of personal information. Choosing other models and not using web search reduces this, but because of auto-switch it may not avoid it completely; it stops only if the Operator turns those providers or web search off for the whole workspace or moves to paid tiers. You can ask the Operator to do that.
• No ads or data brokers. There are no ads, no targeted advertising and no data brokers.
• Sensitive personal information. We use it only to provide the Service. For example, your username and password are used to sign you in. We don't use it to infer characteristics about you.

8 How long we keep information

• When clean-up runs. Automatic deletion runs when an admin opens the admin console, at most once a day. If no admin opens it for a while, records stay until the next time one does.
• Removing an account deletes that person's chats, uploads, AI pictures, projects, crew bots (with their files and pictures), memories, settings, notifications, devices, phone-alert devices, 2-step login records, daily counts, last-seen time, share links, support requests and the invite links they created. Entries about the person in the activity log, feedback, content-filter records and statistics are deleted when their own retention periods end.
• Restoring a backup. If the Operator restores a backup made before you deleted something, that data may come back. Tell the Operator and ask for it to be deleted again.
• Copies of one account. A copy of your account that an admin downloads (section 4) is kept on their device for as long as they keep it, outside the Service's control.
• Caches. Deleted data may stay for a short time in provider caches.
• Longer holds. We may keep information longer where the law requires it, or to resolve disputes or investigate abuse.

9 How we protect information

• Encrypted connections. HTTPS only, with HSTS.
• Passwords. They are hashed with salted scrypt and never stored in plain text.
• Stored data. The Service is set up to encrypt the contents of the records and uploaded images it stores with AES-256-GCM before saving them, using a key kept in the server's settings and never sent to browsers. Some technical details, such as record names that include usernames and dates, and basic file details such as size and type, are not encrypted. Netlify also encrypts stored data at rest.
• Sign-in sessions.
  • On the live HTTPS site, the sign-in cookie is signed, HttpOnly, Secure and SameSite=Strict.
  • Signing out ends the session on the server.
  • You can see your devices and sign them out, and admins can sign everyone out.
• Protection against guessing and bots. Sign-in has rate limits and lockouts. Bot protection uses a hidden trap field, timing checks and optional Cloudflare Turnstile.
• 2-step login (optional). Codes from an authenticator app (TOTP), each usable only once, plus 10 one-time recovery codes stored only as hashes. Wrong codes are rate-limited, and turning 2-step login on signs out your other devices.
• Code previews. Previewed code runs in a separate, locked sandbox with no access to your account, your cookies or this site's data, and it is blocked from loading anything except the public code hosts in section 5. Browsers can't block every way code can reach the internet (for example, real-time connections), so don't type passwords or personal details into a preview.
• Pictures and alerts. AI pictures are checked to be real image files and served with the same protections as uploads. Alerts are encrypted so that only your device can read them, and the push addresses and keys never leave the server.
• Access checks. The Service is designed to check permissions on the server for every request, so that people can reach only their own records and chats shared with them, and only admins can use admin tools.
• Input checks. Input is validated, and only allowed fields are accepted.
• Security headers. Strict headers include a Content-Security-Policy and blocking of framing by other sites.
• API keys. AI provider API keys (including the picture and backup-search keys), and the key for the Operator's own AI, are kept on the server, encrypted, and are never shown in full in the app (admins see only the last four characters). Only the Operator can change them. They leave the server only if the Operator downloads a full backup and chooses to include them.
• Your own AI. The bridge on the Operator's computer only listens on that computer, accepts requests only with its secret key, and lets through only model lists and chat replies.
• Logging. Sign-ins and sensitive admin actions are logged.

No guarantee. No website, app or storage system is completely secure. We can't guarantee that information won't be accessed, disclosed, changed or destroyed, including at third-party providers.

Your part. Use a strong password you don't use elsewhere, consider turning on 2-step login and keep its recovery codes somewhere safe other than your phone, sign out on shared devices, and tell an admin if something looks wrong.

If there is a breach. If we learn of a security breach that affects your personal information, we will notify you when and as the law requires. Because the Service usually doesn't have your email address, notice may be given in the app, on the sign-in page, or in another way the law allows.

10 Your choices and rights

10.1 Things you can do yourself in the app

• View and delete your chats, including their AI pictures.
• View, change and delete your projects and their files.
• Make, change, share, make private and delete your crew bots and their files, and report someone else's bot.
• Delete saved memories, or turn memory off in Settings (or just for one project).
• Change your settings and custom instructions.
• Turn off share links.
• See your signed-in devices and sign them out.
• Clear notifications.
• See your support requests and the replies to them.
• Change your password.
• Turn 2-step login on or off, and make new recovery codes (unless the Operator requires 2-step login for your role).
• Turn phone and desktop alerts on or off for each device, choose which kinds you get, and see which devices get them.
• Choose not to use voice features, pictures, web search, code previews or crew bots.
• Clear this site's data in your browser to remove local preferences.

10.2 Requests to the Operator

You can ask the Operator to:

• give you access to, or a copy of, your personal information;
• correct it;
• delete your account and its data, or a support request;
• explain how it is used.

Contact the Operator (see section 16).

• Verification. We'll verify your request, usually by asking you to confirm it from your signed-in account.
• Timing. We'll try to respond within 45 days.
• Authorized agents. An agent can make a request for you with proof of authorization.
• When we can't comply. We may decline where the law allows. Examples: we can't verify you, the request would reveal someone else's information, or we must keep the information. Deleting data can't reach backup files the Operator has already downloaded, or copies AI providers hold under their own policies. Where practical, we'll tell you about any such limits.
• Appeals. If we decline, you can ask us to reconsider. If you're still unhappy, you can contact your state attorney general.
• No penalty. We won't treat you differently for using these rights.

10.3 California and other US states

Many state privacy laws, such as California's CCPA/CPRA, apply only to businesses that meet certain thresholds. A free, non-commercial workspace run by an individual may not be covered. We offer the choices above to everyone anyway.

In the last 12 months we collected these categories of information, from you, your device, admins and our hosting providers (for approximate location), for the purposes in section 3:

• identifiers (username, display name, IP address, and push addresses for devices where you turn on alerts);
• account login credentials (username, a hashed password and, if you turn on 2-step login, an encrypted authenticator key and hashed recovery codes);
• internet or other electronic network activity (activity log, device type, usage counts);
• approximate geolocation (city, region, country);
• audio (voice recordings, passed to Groq and not stored);
• the content of your messages, uploads, picture descriptions and AI pictures, projects, crew bots and support requests, which may include personal information you choose to enter.

We disclosed these categories to the service providers and AI providers listed in section 5. As section 7 explains, some free-tier AI providers and search services may use what they receive for their own purposes.

10.4 If you are in the EEA, the UK or Switzerland

The Service is meant for people in the United States and is not offered to people in the European Economic Area, the United Kingdom or Switzerland. One reason is that Google's Gemini API terms allow only paid services to be offered to users there, and the Service uses Google's free tier.

If you use the Service from there anyway, data protection law gives you these rights:

• to access, correct or erase your data;
• to restrict how it is used, or to object, including to use based on legitimate interests;
• to receive your data in a portable form;
• to withdraw consent;
• to complain to your local data protection authority.

Contact the Operator to use them. The Operator has not appointed a representative in the EU or the UK.

11 Children and minors

• Adults only. The Service is for people aged 18 or older. It is not directed to children or teens, and we don't knowingly collect information from anyone under that age. If we learn that an account belongs to someone under the minimum age, we will delete the account and the data the Service stores about it. We can't delete copies already sent to AI providers.
• Google's age rule. Google's terms don't allow its free API to be used by anyone under 18, or in services likely to be used by people under 18. This is one reason the Service is for adults only.
• Parents and guardians. If you believe someone under the minimum age has an account, please contact the Operator.

12 Where information is stored

The Service is run from the United States:

• Netlify stores the database in a single US region (US East by default) and runs the server code in the US.
• Cloudflare handles traffic at a location in its global network, usually one near you.
• Groq stores any data it keeps in the US.
• Google, OpenRouter and the model hosts it uses may process data in the US and in other countries.
• Cloudflare Workers AI makes pictures in its global network. Exa, Tavily and the browser push services may process data in the US and in other countries.
• The Operator's own computer, if the Operator connects their own AI, processes messages wherever that computer is.

By using the Service, you understand that your information is transferred to and processed in the United States, where data protection laws may differ from those where you live.

13 Cookies and local storage

The Service uses only what it needs to work:

• no analytics, advertising or tracking cookies;
• no tracking pixels;
• no third-party cookies set by the Service.

You don't need to accept a cookie banner. If you block cookies, you won't be able to sign in.

When its bot protection or challenges run, Cloudflare may set its own strictly necessary security cookies, such as __cf_bm or cf_clearance. Turnstile, if it is turned on, may use similar storage. These fall under Cloudflare's policy.

Google Fonts receives your IP address and user agent when fonts load (see section 5). The app's own code libraries are served from this site, not from a third party. The exceptions are the optional Cloudflare Turnstile check on the sign-in page, and libraries that code previews load from public code hosts inside their sandbox.

14 Do Not Track and Global Privacy Control

Do Not Track. We don't track you across other websites or show ads, so Do Not Track signals don't change anything.

Global Privacy Control. The Service doesn't act on Global Privacy Control signals automatically. We don't sell personal information for money; see section 7 for how some free-tier AI providers may use your messages, and how to avoid it.

15 Changes to this policy

• Updates. We may update this policy. The version number and effective date are shown at the top of this page.
• Accepting a new version. The Operator can turn acceptance in the app on or off. While it is on and the version changes, the app asks you to review and accept the new version before you can keep using the Service, and records which version you accepted and when. While it is off, the new version applies from its effective date, or from when it is posted if that is later.
• Small fixes. Typo or formatting fixes that don't change what we do may be made without a new version.
• Earlier versions. The Operator keeps copies of earlier versions. Ask the Operator if you need one.

16 Contact

Operator

the person who operates this workspace

Contact address

[email protected]

To make a privacy request or ask a question, contact the Operator at the address above. For help with the app itself, you can also use Report a problem in the app.