superbot.gg privacy notice

Version 2026-11-04. What superbot collects, where it goes and how long we keep it. For superbot, this notice replaces the general EZO Labs Inc privacy policy at ezolabs.io.

The short version

Who we are. EZO Labs Inc is the controller.
Why. To sign you in, sync, run requests, bill you and stay secure.
What we keep. Account, payments, uploads, logs and chats, which we can read.
Sealed on your device. Passwords, sessions and SSH keys, read by default. We cannot read them. Turn any off in Settings.
Chat text. Kept 180 days to fix bugs, including model calls the relay carries. Synced chats stay until deleted.
Training. Never on your app chats. Model API requests (beta) are kept 180 days to improve our models, even when anonymized (details).
Not sold. We do not sell your personal data (details).
Your rights. Get a copy, correct, delete or move your data, at /account or [email protected].

In detail: account and billing, device reading, connections, chats and files, calls and teams, logs, who processes it and your rights.

Short view

1. Who we are

The controller of your data is EZO Labs Inc, 1007 N Orange St, 4th Floor, Suite 5765, Wilmington, DE 19801, USA. For a privacy request or question, email [email protected].

2. What we collect, and what happens to it

The relay on your computer reads some kinds of data from your device, and each item's When line says when. Our servers create the rest as you use superbot. Open a group to see each item: what it is, when we collect it, where it goes and how long we keep it (Kept).

Account and billing

Sign-in, payment and invites 7 items

Account and billing identity

Your account id, email address, sign-in session token, device records (name, kind, a short network fingerprint and push token) and payment records.

We keep a record of account events.
It covers sign-ups, sign-ins and beta codes.
It covers trial starts and revoked sessions.
It covers devices and keys you revoke by agent.
It covers devices you link and key changes.
Each has a time and the kind of app.
It may add a method or a reason.
A revoke names the device or key.
It also counts erased accounts.

We keep a daily snapshot of your account.
It holds your plan, tier and subscription status.
It holds when the period ends and if it cancels.
It holds your recharge status.
It holds the account kind and when it began.
It holds no balance or card detail.
A second one counts your linked devices.
We use both for product analytics.

When: When you create the account, add a device or pay
An account event when it happens.
The snapshot once a day.
Where: Held on our servers. Card details are held by Stripe; we keep only Stripe's customer, subscription and payment-method references.
Events and snapshots are in our own telemetry store.
A Bitcoin, Lightning or Monero payment goes to our own wallets, and our billing service keeps the invoice of every crypto payment.
Kept: Until you delete the account. Billing records are kept after that (see "Deleting your account" below); we have no automatic deletion schedule for them.
Events and snapshots are kept 730 days.
Deleting the account removes yours sooner.
The count of erased accounts stays 730 days.
It is a time and a pseudonym, not your id.

Apple, Google, TikTok and X sign-in

What Apple, Google or X sends us.
One record for each sign-in.
A new sign-in replaces it.
It is for display and analytics.
We never match accounts with it.

Apple: your name and email.
Apple sends the name only once.
We keep it after that.
Whether the email is a relay.
Whether Apple verified it.
Apple's real-person signal.
Plus other fields about you.

Google: your name and email.
Whether Google verified it.
Your given and family names.
Your photo link and language.
Your Workspace domain, if any.
Plus other fields about you.

TikTok: your account id (union_id).
Your display name and your photo.
We keep the id and the name.
The id links this sign-in to you.
We keep nothing else TikTok sends.

TikTok also sends an access token.
We revoke it at sign-in.
We never store it.

X: your public profile.
Name, username, bio, location.
Website, photo and banner.
Follower and following counts.
Post and like counts.
Verified status and type.
If X verified your identity.
If it is a parody account.
Your subscription type.
Subscriber count.
Verified follower count.
When you joined X.
If your posts are protected.
Links in your bio and website.
Where X withholds your account.
Your email, if X confirmed it.
Other profile fields X sends.

Your pinned post and latest post.
Up to 20 of your recent posts.
Only your own, never reposts.
Never your home timeline.
Never anyone else's posts.
Each with its text and time.
Its language and public counts.
Its links, mentions and hashtags.

When: Sign-ins after you accept this.
TikTok sign-in from 17 November 2026.
Where: Held on our servers.
We can read it.
Kept: Until you delete the account.
Or until you unlink that sign-in.
Your export includes it.

superbot API keys

We keep an encrypted copy of each API key.
It lets you copy the key again.

When: When you create or roll a key.
Where: Sealed under a key held for your account.
Our servers can open it.
It is never written unsealed.
Kept: Until you revoke the key.
Or until you delete the account.
Backups drop it within 14 days.
Your export shows the key masked.

Auto-recharge settings

Whether auto-recharge is on, the amount per charge and the monthly cap.

When: Off until you turn it on at /account/billing
Where: Held on your account. Stripe makes each charge on the payment method it saved at checkout.
Kept: Until you delete the account.

Beta invites and referrals

Invite codes and invited emails (stored as keyed hashes), the email on a pending pass, and your referral code and its uses.

When: When you are invited, redeem a code or refer someone
Where: Our servers.
Kept: A pass email is kept until you delete the account. We have no automatic deletion schedule for pass or referral records.

Profile picture

The image you upload as your avatar.

When: When you upload one
Where: Our servers. Anyone with its link can open it.
Kept: Until you replace it or delete the account.

Waitlist and sign-up attribution

A sign-up email address, an invite code, and the channel a visit came from.

When: When you join the waitlist or sign up through a link
Where: Our servers. When an ad brought you in and you sign up, the ad platform named in section 5 receives your hashed email and the ad click id. From 26 October 2026, when you make an account on our website, Meta receives your hashed email (section 4).
Kept: We have no automatic deletion schedule for waitlist or attribution rows; ask us to delete them.

What the relay reads from your device

On by default, and you can turn each off 7 items

Browser cookies and site storage

The cookies and site storage of each browser profile on your device, sealed on your device before it syncs.

When: Session import is on by default in new installs. It runs every 12 hours, and when you ask. You can turn it off in Settings > Privacy in the desktop app, with Browser sign-ins, which also turns off the password scan, or, for this alone, with Sync to your other devices in that switch's details.
Where: Sealed on your device with a key only your devices hold. Our servers store the sealed copy and cannot open it. We can see which site each row is for.
Kept: Until you delete the row or the account. See "Deleting a vault item" below.

Saved passwords

Entries from 1Password, Bitwarden, LastPass, KeePass, Chrome, Chromium and Firefox. Your macOS Keychain is looked up only for the site being signed into, never listed.

When: The password scan is on by default in new installs. It runs once a day, and at startup when the last scan finished more than 12 hours before. You can turn it off in Settings > Privacy in the desktop app, with Browser sign-ins. The Passwords and SSH keys row there shares that switch.
Where: Sealed on your device with a key only your devices hold. Our servers store the sealed copy and cannot open it. We can see the site, the password manager and the username of each row.
Kept: Until you delete the row or the account. See "Deleting a vault item" below.

SSH private keys

The private keys in your ~/.ssh folder, keyed by their public fingerprint.

When: Read by the password scan (the same setting), which is on by default in new installs. It runs with that scan. You can turn it off in Settings > Privacy in the desktop app, with Browser sign-ins.
Where: Sealed on your device with a key only your devices hold. Our servers store the sealed copy and cannot open it.
Kept: Until you delete the row or the account.

MCP server headers

The authentication headers of MCP servers you add, and of servers set up in your other apps if you allow config import.

When: Imported headers are read while config import is on. Headers you add yourself are sent when you add them.
Where: Imported headers are sealed on your device, and we cannot read them. Headers you add by hand reach us in cleartext over an encrypted connection; we store them encrypted, and our servers can read them.
Kept: Until you delete the server or the account.

Imported chat histories

Your Claude, Codex and Cursor conversation files, and chat-export archives in Downloads, Desktop and Documents.

When: Chat import is on by default in new installs. It runs every 5 minutes. You can turn it off in Settings > Privacy in the desktop app, with Vendor chats, or for one app under Choose apps.
Where: Not sealed. Stored on our servers as readable text, so search and your other devices can use it. No redaction runs over the messages.
Kept: Until you delete the import or the account.

Session presence

Whether your browser is signed in to a short fixed list of delivery and restaurant services (yes or no per site).

When: At startup, while session import or the password scan is on. Both are on by default in new installs.
Where: Sent to us as a list of those sites. It is not sealed.
Kept: We keep it in memory only while your device is connected. Deleting your account removes it.

Imported-chat search index

A numeric embedding made from each imported chat history, so search can find passages in it.

When: When a chat history is imported
Where: OpenAI makes the embedding, and the result is stored on our servers. When you search your imported chats, or Superbot searches them for you, the search words go to OpenAI to be embedded the same way, and that embedding is not stored. This is part of running the search you asked for, so it rests on our contract with you, not on a separate consent.
Kept: Deleted with the account.

Connections and publishing

Services you link or publish to 9 items

Model-connection credentials

The sign-in or API key for each model provider you connect.

When: When you connect a provider
Where: A sign-in sealed on an enrolled device stays sealed, and we cannot read the sign-in itself. Your device unseals it to mint a short-lived access token, and our servers send that token to the vendor to make the request. An API key you paste is stored encrypted, and our servers decrypt it to make each request that uses it.
Kept: Until you disconnect the provider or delete the account.

Custom endpoints and external APIs

The address and key of each endpoint or API you add, such as OpenRouter, DeepSeek, Hugging Face, Perplexity or any compatible URL.

When: When you add one
Where: Stored encrypted under a key we hold. We hold your key so we can call the endpoint for you: our servers decrypt it and make each request, so the request passes through us.
Kept: Until you remove it or delete the account.

Git provider tokens

GitHub and GitLab tokens you paste into the relay or grant through a device sign-in. A token held by the gh or glab command-line tool stays on your computer and is never uploaded.

When: When you connect Git with a pasted token or a device sign-in
Where: The relay sends them to your Superbot account over an encrypted connection, and we store them encrypted under a key our servers hold, so our servers can read them. This is what lets your phone list your Git repositories while your computer is offline.
Kept: Until you disconnect Git or delete the account. See "Deleting a vault item" below.

Connected accounts

Sign-in tokens for Google, GitHub, Notion, Slack and similar services you connect, and what the assistant reads through them for a task.

When: When you connect one
Where: Stored encrypted. When a task uses the connection our servers make the call, so we see what the service returns.
Kept: Until you disconnect it or delete the account.

Published apps

The access grant for your Cloudflare or Railway account. Your app's code, content, logs and analytics stay in your account with that provider.

When: When you connect Cloudflare or Railway
Where: The grant is sealed on your device, and we cannot read it. Deploys run from your computer. A deploy to a superbot.sh or ezo.dev address sends the Cloudflare account id, subdomain and Worker name to us to reserve the address (see "Branded subdomain claims").
Kept: Until you disconnect the account. See "Deleting a vault item" below.

Build requests on X

The public post that tagged @ezolabsinc, the handle and numeric id of the X account that posted it, and the app built from that post.

When: When a post tags @ezolabsinc on X
Where: We read the post from X through the official X API. The app is published as a Cloudflare Worker behind a superbot.sh address, and its files are stored on our servers.
Kept: The post and the author id are kept until you or the author ask us to delete them. The app stays live for at least 7 days, and may be removed after that when capacity needs the room.

superbot.sh addresses

For each visit to an app at a superbot.sh address: the full address including its path and query, the method, the visitor's country, the status, the timing and the size. Not the IP address, cookies or headers.

When: Always, for apps on a superbot.sh address
Where: Cloudflare's analytics and our own log store.
Kept: Our copy 90 days.

Weekly usage meters

Weekly usage figures for each model provider account the desktop app reads them from, and the account's display label, which may be a sign-in email.

When: Whenever the desktop app sends them
Where: Our servers, so your other devices can show your usage limits.
Kept: Replaced each time the desktop app sends them, and cleared when it sends none. Deleted with the account.

Branded subdomain claims

For an app published at a superbot.sh or ezo.dev address: the address, the Cloudflare account id, the workers.dev subdomain and the Worker name.

When: When you publish an app to a branded address
Where: Our servers, in the hostname registry the router serves the address from.
Kept: While the claim is live. A released address is held 90 days before a sweep purges it. Deleting your account de-identifies the row.

Your chats and files

What you write, save and make 12 items

Sync data

Your chats with the full text of every message, projects (including the folder path on your computer), rules, skills and context documents, MCP servers, settings, spaces, shares and access grants.

When: Whenever a signed-in device saves a change. Chats sync while Sync chats across devices is on. It is on by default, and you can turn it off.
Where: Stored on our servers, encrypted at rest under a key we hold, so we can read it. It is not end-to-end encrypted.
Kept: No automatic deletion. Deleting an item stops it syncing; see "Deleting a vault item" below.

Project folder mirror

The files in the folder of each project in your project list, as a git repository.

When: On by default. The relay mirrors a project's folder on a day you use the project, including a project it found in your other apps, such as your code editor's recent folders, and whenever you press Sync now. You can turn it off for all your projects with Sync projects across devices in Settings > Privacy. To turn it off for one project, click the project's name at the top of its page and turn off Sync this project.
Where: Stored on an encrypted disk on our servers, and our servers can read it. The copy on your computer is not encrypted.
Kept: We have no automatic deletion schedule. Deleting your account removes it.

Finished replies

A copy of each finished reply and the id of its chat session, so an app that dropped its connection can still pick the reply up.

When: For every reply, for every user
Where: Our servers.
Kept: Up to 24 hours, then deleted.

Uploads and attachments

Files you upload to a tool, and files you attach to a chat.

When: When you upload or attach one
Where: Our servers.
Kept: A file you upload to a tool 7 days by default, and never more than 30 days. An attachment to a chat is kept with that chat, until you delete it or the account.

Screenshots and screen recordings

Screens the assistant captures when a task asks it to, and recordings you start.

When: Computer use starts at Ask in new installs, so the assistant asks you before it captures your screen. Off asks you to turn it on first, and Always captures without asking. You choose with the Computer use button by the message box. A recording starts only when you start one.
Where: Uploaded to your workspace.
Kept: Until you delete the artifact or the account.

Monitoring summaries

Summaries of mail, messages and calendar items the assistant prepares from your own connected accounts.

When: Off unless you switch monitoring on
Where: Held on your account.
Kept: Until you delete them or the account.

Cloud agent runs

The instructions you give a cloud agent, its status, summary and output files, and the repository and access you give it.

When: When you start a cloud run
Where: Runs in containers on our servers.
The run record is stored there too.
Kept: No automatic deletion.
We keep metered usage.
It backs charges already billed.
Your account id becomes a pseudonym.
Deleting your account removes the rest.

Super mode runs

The run record of a super mode message, and a working folder for its agents.

When: When you use super mode
Where: Our servers, which send each leg to its provider.
Kept: Working folder 24 hours. The run record stays until you delete your account.

Media generations

The prompt you give and the image, video, audio or 3D file that comes back.

A file shared on superbot.sh records, for each request, the file id, the visitor's country, the referrer host and whether it was a person or an automated preview.
No IP address or browser string is stored.
These records are kept for three months.

When: When you ask for one
Where: Sent to the provider that makes it (section 5). The file is stored on our servers and served from an unguessable link that anyone with the link can open.
Kept: We have no automatic deletion schedule for generated files.
A file is served with a one-year cache, so a copy a browser or network cache already holds can stay that long.
A file reported to us can be removed by an operator, and its link stops serving; you have no removal control of your own.

Shared chat links

A chat you choose to share by link.

Opening a chat link on superbot.sh records a one-way hash of the link, the visitor's country, the referrer host and whether it was a person or an automated preview.
No IP address or browser string is stored.
These records are kept for three months.

When: When you share one
Where: Anyone with the link can open it.
Kept: The link stops working after 30 days by default, and never lasts more than 365 days.

Long-term memory

Facts the assistant remembers about you and items you share, such as a contact card, held in your account profile.

When: As you chat, and when you share an item
Where: Our servers, in your account profile.
Kept: Until you delete it or the account.

Tool-search queries

A numeric embedding made from the words the agent searches its tools with, so it can find the tool a task needs. The search can carry words from your request. The name and description of each tool it searches are embedded too, including the tools of MCP servers you add.

When: When the agent searches its tools. Never for an incognito key or under zero data retention, and never on a turn that read data we got from Google APIs, such as Gmail.
Where: OpenAI makes the embedding. This is part of running the request you asked for, so it rests on our contract with you, not on a separate consent.
Kept: We do not store the embedding. It is held only in our server's memory, at most until the server restarts.

Calls, voice, rooms and teams

Voice, phone and shared spaces 5 items

Voice sessions

Your live microphone audio, and the text of the assistant's spoken replies.

When: Only during a voice session you start
Where: Audio streams through our servers to Deepgram for transcription. Reply text goes to ElevenLabs for speech. Both run under our accounts with them.
Kept: We do not store the audio. The transcript is kept with the conversation. A session log (ids and times) has no automatic deletion schedule.

Calls, text messages and faxes

The numbers, times, duration, status and outcome of each call, and the text of each message and fax record.

When: When you place or receive one
Where: Telnyx carries them.
ElevenLabs voices the calls.
Twilio carries text messages in rooms.
The records are stored on our servers.
Kept: No automatic deletion.
We keep call and fax consent as proof.
Each holds a number, a time and a scope.
We keep the number and time of a STOP.
That number stays opted out.
Your account id becomes a pseudonym.
Deleting your account removes the rest.

Voice clones

A recording you upload to make a voice clone, and the speaker's consent record.

When: When you make a clone
Where: The recording goes to ElevenLabs. We keep a fingerprint of it and the consent record.
Kept: Until you delete the account. Deleting it removes our fingerprint and the consent record, and asks ElevenLabs to delete the voice.

Rooms

Messages, members and email invites in a shared room.

When: When you join or create a room
Where: Our servers. Other members of the room can see them.
Kept: Messages 180 days. The room text-message log 90 days.

Teams and organizations

Team name, members, roles and directory groups, signed policy revisions and the receipts devices send when they apply them, the team's pooled rules, skills and MCP servers, the team vault, and the team's audit log of what each member's agents did with its credentials and tools.

When: When you create or join a team
Where: Pooled rules, skills and MCP servers are encrypted under a key our servers hold, so we can read them. Team vault items are sealed under a team key only members' devices can open, so we cannot read them. The team's owners and admins can read its audit log.
Kept: Until you delete the account. Deleting it removes you from each team and deletes a team you were the only member of. A team with other members keeps its own records, and items you added to its vault are revoked, but their encrypted copies stay. Its audit log keeps each entry one year, and keeps yours with your email and account id replaced by a pseudonym. The last owner must hand over ownership first. A member the team removes is kept as a removal record.

Logs and diagnostics

Records of how it runs 10 items

Request logs

For each request to our servers: the time, path, status, timing, browser and network fingerprints, your account and device ids, and your IP address as a keyed hash. The path is recorded as sent, so a one-time code, invite token or other credential in a link can appear in it. Also the words you send to a tool and the reply it gets, for every user.

When: Always, while you use the service. There is no setting that turns it off.
Where: Our own log store.
Kept: 180 days.

Model request records

For each model call our servers make: the model, the timing and the outcome, and the reply and any error text, for every user.

When: Always, while you use the service. There is no setting that turns it off.
Where: Our own log store.
Kept: 180 days.

Product telemetry

Device events (install, locale, time zone, a count and sample of your installed apps, heartbeats). The prompt, reply and any error of each turn, for every user. We use this text to fix bugs. We never use it to train or fine-tune any model. A Model API request that passes through to a model has its own row.

When: Automatically, for every user. There is no setting that turns off chat text.
Where: Our own telemetry store.
Kept: Device events 180 days. The latest record per device 730 days. Conversation text 180 days.

Relay and tool logs

The full request and reply of each model call the relay carries for you, and the arguments and results of each tool it runs on your computer, including jobs our servers send to it. Also the timing and size of each piece of a streamed reply, and the journal of each agent session the relay runs, with its messages. The text in these records (requests, replies, tool arguments and results, and session messages) is kept for every user.

When: Automatically while the relay runs. Share usage counts, in Settings > Privacy in the desktop app, is on by default, and turning it off stops the relay sending these records.
Where: Our own telemetry store. It is not end-to-end encrypted.
Kept: Request and reply bodies 180 days. Streamed-reply timing 90 days. Agent session journals 180 days. Tool calls 90 days. Request summaries and relay events 400 days.

Model gateway logs

The prompt of each model call we send through our model gateway, including its instructions, message history and tool definitions, and metadata for each request (model, timing, size and outcome).

When: Always, for calls through our gateway.
Where: Our own gateway log store, for debugging. It is not end-to-end encrypted.
Kept: Prompts about 15 days, and up to about 45 days while storage partitions age out. Request metadata 200 days. Deleting your account does not remove them sooner. Model API requests are also kept 180 days, under superbot Model API requests.

superbot Model API requests

We keep the text of each Model API request.
We keep the reply too.
It covers calls on your own vendor key.

A request keeps its system prompt.
It keeps your messages and tool results.
It keeps tool definitions and settings.

A reply keeps its text and reasoning.
It keeps each tool call and its input.

We drop user and cache ids a client adds.

We use it to evaluate our own models.
We also use it to train and improve them.

The Model API is /v1/chat/completions, /v1/messages and /v1/responses.
It covers a request that passes through to a model.
A request the superbot agent answers is conversation text instead.
A request the agent fails to answer is kept here.
A chat turn or an MCP tool call is not part of it.

Data we got from Google APIs, such as Gmail, is left out.
So are the model replies that follow it in the session.
We never use either to train.

When: During the beta, for every request the Model API passes through to a model.
Anonymize my data does not stop it.
It is off for an incognito key.
It is off under zero data retention.
Where: Our own telemetry store.
We can read it.
It is not end-to-end encrypted.
Kept: 180 days, then deleted.
Deleting your account erases its rows.
With Anonymize on, a row has no account.
Deleting your account does not reach it.

Crash reports

Error and crash reports from the desktop and mobile apps, sent without your account or device identity. No request bodies.

When: On by default in released apps. You can turn it off with Send crash reports.
Where: Sentry. We set it to discard IP addresses and location.
Kept: Sentry's retention applies.

Server error reports

Error reports from our servers: the error type, the stack trace (no message text), the route pattern (never the path as sent), and the request id. The report also names the environment, the release and the service, and for a process-level error its kind. Never request bodies, headers, cookies, query strings, prompts, messages, files, credentials, or any user, device or IP identifier.

When: Automatically when a server process hits an error. There is no setting that turns it off.
Where: Sentry. We set it to discard IP addresses and location.
Kept: Sentry's retention applies.

Bug reports

What you write, and any screenshot, pasted image or screen recording you attach.

When: When you send one
Where: Our servers.
Kept: 180 days, or until you delete it or the account. A summary without the attachments is kept 730 days.

Web-page captures and session logs

Pages the search and fetch tools retrieve, when they are added to the shared page catalog, and the log of an agent session.

When: When the agent reads a page or runs a session
Where: Our sidecar services. The page catalog holds example bodies with secrets removed, keyed by a hash of the account, not the account id.
Kept: Session logs 7 days. The page catalog has no automatic deletion schedule.

Why, and our legal basis. Each group of rows says what its data is for. We rely on the performance of our contract with you to sign you in, sync your devices, run your requests and media, carry your calls, run your teams and publishing, and bill you. We rely on our legitimate interests in keeping the service working and secure to keep request, model, gateway and crash logs, and to detect and stop abuse. We rely on our legitimate interest in improving our own models to keep the text of superbot Model API requests during the beta. We rely on your consent to read from your device, and on a legal obligation for billing, tax and opt-out records. This is the lawful basis for the processing (GDPR Art. 6 and 13(1)(c)).

Withdrawing consent. Where we rely on your consent, you can withdraw it at any time by turning the setting off in Settings > Privacy, or by emailing [email protected]. Withdrawing it stops future reads and does not affect what we did before it (GDPR Art. 7(3)).

Retention. Where an item gives a period, it is deleted on that timer. Otherwise it stays until you delete it, and deleting your account removes it only where the item says so. The law may also require us to keep some records.

Deleting your account
It also reaches our analytics and logs.
Rows tied to your account are hidden.
That takes up to a day.
Physical removal takes up to 30 days.
Our application log files are scrubbed.
We keep the billing ledger, for tax law.
We keep payment events behind invoices.
We keep your deletion request record.
It proves we carried it out.
Stripe keeps its payment records by law.
We keep the record of a charge still owed.
It is needed for billing.
We delete it once the charge settles.
Under a pseudonym, not your id or email
We keep metered usage.
It backs charges already billed.
A team audit log keeps your agent actions.
The team owns that record.
We keep call and fax consent as proof.
Each holds a number, a time and a scope.
We keep the number and time of a STOP.
That number stays opted out.
We keep each charge record for billing.
It has time, charge id, quantity, outcome.
We keep the record of a withdrawal.
It has the contract, amount and dates.
We keep a count of erased accounts.
Each holds the time of the erasure.
Rows stored without your account
They come from anonymized data.
Or from data sent without signing in.
Your account cannot look them up.
Deleting it removes any that name your id.
They expire on their own timers.

The steps to delete your account, and a way to ask for it without signing in, are at /legal/delete-account.

Deleting a vault item revokes it at once, but its encrypted copy stays in an append-only record until you delete your account. We can see which sites and account names you hold credentials for, not the secrets.

Anonymize my data is on by default in new installs of the app, and you can turn it off. It removes the link to your account from analytics and log rows: while it is on, those rows are stored with no account id, and your device id is replaced by a keyed hash. Your own activity history, the Activity card and your Relay call log, is shown only to you and is erased with your account, so it keeps the link to your account. The key is held only in memory and discarded at each UTC midnight, so one day's rows cannot be joined to the next. Most rows carry no location, time zone or IP address. The text of your chats is still kept, and it can identify you if you type things that identify you. The text of superbot Model API requests in the beta is kept too, without your account. Two records keep a raw or keyed value, without your account: the latest record for each device, and our security events. A security event keeps a keyed IP value that is not rotated, so it can be recognised from one day to the next, because our security team needs the address. It does not cover your account, billing, the chats and credential vault you choose to sync, the keys of custom endpoints you add, a bug report you file, or model gateway logs. It also does not apply to a team's audit log, which records your agent actions with your email and account id for the team's owners and admins. With it off, telemetry is tied to your account and device.

superbot Model API requests, in the beta. The superbot Model API is the /v1/chat/completions, /v1/messages and /v1/responses endpoints. While the beta runs, we keep the text of each request one of them passes through to a model, and its reply, for 180 days. We use it to evaluate, train and improve superbot's own models. A request keeps its system prompt, messages, tool results and tool definitions. A reply keeps its text, reasoning and tool calls. This covers calls on your own vendor key. It covers accounts with Anonymize my data on too, with their rows stored without the account. It does not cover a request made with an incognito key, or by a member of a team with zero data retention. Mark a key as incognito to opt that key out. When a request carries data we got from Google APIs, such as Gmail, that data and the model's replies that follow it in the session are left out of this capture and never used to train. A chat turn or a call through the superbot MCP tool is not a Model API request. A Model API request the superbot agent answers is kept as conversation text (Product telemetry), which we never use to train, but one the agent fails to answer is kept here. Deleting your account erases the rows tied to it, as above. A row stored without your account cannot be tied back to it, so deleting the account does not reach it. That row is deleted when its 180 days end.

Training. Apart from superbot Model API requests in the beta (above), superbot does not use your conversations, prompts, files or code to train or fine-tune any model. When you connect your own vendor account or key, that vendor's terms and your settings with them apply. Some models we run for you are provided by vendors (section 5) under their own terms. Data we get through the Superbot browser extension is never used to train any model, and the beta capture above leaves it out (section 3).

People on your team, in superbot relay. If you use the Team tab to give other people keys, superbot relay on your computer counts what each person does: prompts, chats, tokens and what the calls cost. If you turn on prompt history, it also keeps the text of each person's prompts and the replies, up to 4,000 characters each, in a file that only your account on that computer can read. Prompt history is off until you turn it on, and you can turn it on only after you confirm that the people on your team were told. You choose how many days the text is kept (30 unless you pick another number), you can turn it off for one person, and when you turn it off you can delete everything already kept. Nothing is kept for a key that is incognito or under zero data retention. We do not receive this copy from your computer. A request a person sends through our servers, on superbot credits or the hosted API, is handled as the rest of this notice describes, including the beta capture, which leaves out an incognito key. A person on your team who wants to know what you keep about them should ask you. For what we keep, write to [email protected].

The apps use no advertising or product-analytics service. Our web pages may use Google Analytics (section 4), and our website at superbot.gg may also load the TikTok Pixel to measure our TikTok ads (section 4). From 5 November 2026, it may also load the X Pixel to measure our ads on X (section 4). From 26 October 2026, it may also load the Meta Pixel to measure our ads on Facebook and Instagram (section 4).

3. What the relay reads

In new installs, the relay reads from your browsers, password managers, ~/.ssh folder and other AI apps by default. You can turn any of them off in Settings > Privacy in the desktop app, with the switch named for each:

Session import: browser cookies and sessions, and whether you are signed in to a short list of sites. Switch: Browser sign-ins, or Sync to your other devices in its details for the sync alone.
Password scan: saved passwords and SSH private keys. Switch: Browser sign-ins, the same switch as session import.
Chat import: your chat histories from other AI apps. Switch: Vendor chats.
Config import: your other apps' MCP servers, rules, skills and context files. Switch: AI app setup.

Cookies, sessions, passwords and SSH keys are sealed on your device with a key only your devices hold before they sync. Our servers store the sealed copy and cannot open it. Chat histories and the list of signed-in sites are not sealed, and we can read them.

How often each runs, where it goes and how long we keep it is under what the relay reads from your device. Turning a setting off stops future reads. Delete what was already synced in your vault. An existing install keeps any answer you already gave.

The Superbot browser extension

What it is. The Superbot browser extension lets Superbot do the web tasks you ask for in background tabs of your browser that use your existing sign-ins. If the Superbot app is on the same computer, the extension connects your browser to it, which this notice calls Companion mode. If the app is not installed, or you choose Standalone in the extension's Settings, the extension works on its own and connects to our servers instead, which this notice calls Standalone mode. A choice in its Settings always wins: Companion never goes online by itself, and Standalone never contacts the app, whether or not the app is installed. You can ask for a task in the Superbot app, or in the extension's own popup or side panel, where you can also watch it work. It does nothing until you press Allow on its welcome page. Pressing Allow is your consent, and you can withdraw it at any time, as described below. If you pressed Allow before the extension could work without the app, it asks you once more, and no task or page leaves your browser before you agree.

What it reads. It works only in the tabs it opened for Superbot, and in a tab you hand to Superbot yourself. In those tabs it reads the page's text and structure, screenshots and a live picture of the tab, the tab's web address and title, the addresses the tab goes to, the text of any dialog the page shows, and whether the page has loaded or crashed. Elsewhere in your browser it reads only which browser it runs in and its version, how many windows are open, and when you switch to or away from one of Superbot's tabs. While the extension's popup or side panel is open, it also reads the title and web address of the tab you are on, so it can name that page on its Ask card. It passes them to the Superbot app in Companion mode, or to our servers in Standalone mode, only when you send a task or an addition to a task. It does not read the pages of your other tabs, your history, bookmarks, your own downloads or saved passwords.

Save downloads. The extension has a Save downloads switch. It turns on when you press Allow on the welcome page and your browser grants the permission it asks for there, and it stays off if you refuse. You can turn it off, or back on, in the popup's Settings at any time. If you installed the extension before this change, it keeps the choice you had. While it is on, when you ask Superbot to save a file from a page, your browser downloads that one file into a Superbot folder in your Downloads folder. The extension then reads that download's name, size, type and progress, and nothing about any other download. Superbot refuses programs and installers, and files over 500 MiB. Turning the switch off stops it at once.

File uploads. Superbot attaches a file to a page only when you ask it to, and only after you turn on the extension's access to files in your browser's extension settings. The Superbot app copies that file into a private temporary folder on your computer, and the extension hands the page that copy, as if you had picked it yourself. The site receives the file when the page sends it. The app deletes the copy the next time it attaches a file an hour or more later.

Standalone mode. Standalone mode is the extension working on its own, because the app is not installed or because you chose Standalone in the extension's Settings. If the app is not installed and you have chosen nothing, the first time you send a task it makes a guest account for your browser, so you can start without signing up, on the same free credits a guest gets on our website. If guests are not being admitted, it asks you to sign in with a short code you enter on our website. If you chose Standalone, it asks you to sign in with that code first, and it makes a guest account only if you press Skip for now, which it offers only while guests are being admitted. When the free credits are used up, it asks you to sign up, and once you sign in to your new account in the extension, the chats from your guest account move into it. If you are signed in and run out of credits, it takes you to the billing page on our website. A guest account and its chats are deleted if the extension goes 28 days without using it. The extension opens its connection to our servers only while a task is running or its popup or side panel is open, and for 30 seconds after, and it closes the connection when you press Disconnect. Everything on that connection is encrypted. Share sign-ins, Save downloads and File uploads need the Superbot app, so in Standalone mode the extension does none of them.

Where it goes. In Companion mode, the extension sends what it reads only to the Superbot app on your computer, and it makes no network requests of its own, except the download of a file you asked Superbot to save. In Standalone mode, it sends what it reads over an encrypted connection to our servers and nowhere else, it never contacts the app, and it makes no other network requests of its own. From there:

The live picture is shown in the app so you can watch the work, and it is not stored. In Standalone mode, it is made in your browser for the extension's popup and is never sent anywhere. When Superbot takes a screenshot of the tab it may use the latest picture, and that screenshot is handled like any other page Superbot reads.
Page content and screenshots that Superbot reads become part of your task, like the result of any other tool. So does a task you type into the extension's Ask box, which goes to the Superbot app on your computer with the title and web address of its page. In Standalone mode, it goes to our servers the same way, with the title and web address of its page. They are sent to our servers and to the AI provider that handles that turn (Model vendors, section 5).
In Standalone mode, the extension also tells our servers which browser and version it runs in, its own version, how many windows are open and its random install id, so the connection can be set up and kept alive.
Shared cookies stay on your computer, in Superbot's built-in browser. If session import is on, that browser's sign-ins are sealed on your device before they sync, as this section describes above.
A file Superbot saves stays on your computer, in the Superbot folder of your Downloads, until you delete it. Its name, size and place go to the agent like any other tool result, and Superbot reads the file itself only when your task needs it.

How long it is kept. The extension itself keeps three settings in your browser: whether you pressed Allow, whether Share sign-ins is on, and your Companion or Standalone choice, if you made one. It also keeps one task for a short time. If a task you type in its Ask box cannot run yet, because you have to sign in or sign up first, or because it can run only in the Superbot app, the task waits in your browser with the title and web address of its page, and runs when it can. It holds one task at a time, so a newer one replaces it, and a task that has waited 24 hours is deleted and never sent. Once it has worked in Standalone mode, it also keeps your Superbot sign-in: a random install id, the tokens that keep you signed in, the email address of your account if you signed in to one, and, for a guest, a key that moves the guest's chats into your account when you sign up. Signing out deletes the tokens, the address and the key. The install id stays until you remove the extension, and removing the extension deletes everything it stored. Its other working state, such as Pause, which tabs Superbot opened, how far a running task has got, a sign-in code while you sign in and the last task you sent with its page's title and web address, stays in your browser's memory and is gone when you quit the browser or reload the extension. What Superbot reads through it is kept like the rest of the chat: in your chat history until you delete it (Sync data, under chats and files in section 2), each screenshot as a file in your workspace until you delete it or your account, and its text in our logs of the turn for up to 180 days.

How we use it. We use data we get through the extension only to provide and improve the feature you see: doing the web tasks you ask for in your browser, and keeping that working and secure. We do not sell it, and we do not use or transfer it for advertising. We transfer it only to carry out your task (through the AI providers and other processors we use, including those named in section 5), to comply with the law, or to protect against fraud, abuse or security threats. If our business is merged, acquired or sold, we transfer it only after asking you first. We never include it in the de-identified or aggregated data described in section 7, and we never use it to train any model, including under the beta capture of superbot Model API requests (details). No one at EZO Labs Inc reads it except with your consent (for example, when you attach it to a bug report), when it is needed for security, or when the law requires it. We also count, for each account, how far a new use of the extension gets: that it signed in, that it connected, that a task started, that the agent took a first step and that it ran a first action. A count holds no page, address, task or answer, and we read the counts only to see whether the extension works for people and to fix it where it does not.

Limited Use. The use of information received through the Superbot browser extension adheres to the Chrome Web Store User Data Policy, including the Limited Use requirements.

Stopping it. To stop Superbot in a tab, press Cancel on the "Superbot started debugging this browser" bar your browser shows while Superbot works in that tab. To stop the extension, turn it off from its button in your browser's toolbar, or remove it on your browser's extensions page. Either one stops future reads. In Standalone mode, Disconnect in the popup closes the connection to our servers at once, and Sign out in the popup's Settings ends your sign-in there as well. What Superbot already read stays in your chats until you delete them or your account (Deleting your account).

4. Cookies

superbot sets only these first-party cookies: sbk, sbfl, sblk, sbnh, sbinv, sbp, sbsi, sbfn, sb_ghlink_csrf, sb_handoff, sbtrial, sbev, _ga, _ttp, _tt_enable_cookie, ttclid, ttcsid, ttcsid_ followed by our pixel id, _fbp, _fbc, _twpid, _twclid, _twsid. Strictly necessary: sbk holds your sign-in session for up to 90 days; sbfl, sblk, sbnh, sbinv, sbp, sbsi, sbfn, sb_ghlink_csrf only carry the state of a sign-in, invitation or account-linking step and expire within 1 day; sb_handoff carries the words you typed in the home page ask box to the web app, in a new chat, and expires within 10 minutes; sbtrial ties this browser to the free trial API key it asked for, so asking again gives the same key, and expires when that trial ends, within 28 days. Analytics: sbev, _ga, which last 2 minutes and 13 months. Advertising: _ttp, _tt_enable_cookie, ttclid, ttcsid, ttcsid_ followed by our pixel id, which the TikTok Pixel sets on our website at superbot.gg from 23 October 2026, never in the apps, and which last 13 months from last use (see TikTok below). _fbp, _fbc, which the Meta Pixel sets on our website at superbot.gg from 26 October 2026, never in the apps, and which last 90 days from last use (see Meta below). _twpid, _twclid, _twsid, which the X Pixel sets on our website at superbot.gg, never in the apps, and which last 390 days from last use (see X below). We also keep first-party values in your browser's local and session storage for preferences, page state and our own landing-page experiment. None is shared with advertisers, except the ad click id in sb-attr: when an ad brought you in and you sign up, we send it to that ad platform (section 5). The only ones that leave your browser are sb-attr, sb-vid, sb-sid, sb-internal, sb-ga-force: our own pages read sb-attr, sb-vid, sb-sid, and Google Analytics reads sb-attr, sb-internal, sb-ga-force. Clearing cookies signs you out.

Google Analytics

Our web pages (superbot.gg and this site) may use Google Analytics 4 to count visits and sign-up steps: the pages you view, with invite codes, one-time codes, tokens and email addresses removed from every address before it is sent, the links and buttons you click, your approximate location and your browser and device type. Email addresses and codes are never sent. In the EEA, the UK and Switzerland analytics storage is off by default, so Google Analytics sets no _ga cookie there. When you sign up, start as a guest in the web app or confirm your place on the waitlist on the web, our server may also send Google Analytics that step, with the page it came from, the campaign that brought you (if any) and the id in your browser's _ga cookie, so it is counted even when the page's own report is lost. It does this only for a browser that already has that cookie, and never in the EEA, the UK or Switzerland, or when your browser sends Global Privacy Control. A guest start is sent with no account or guest id. All of Google's advertising features are off: no Google signals, no ad personalization, no ad storage. The apps do not use Google Analytics. To opt out on any site, install Google's opt-out browser add-on.

TikTok

From 23 October 2026, our website at superbot.gg may load the TikTok Pixel to measure our ads on TikTok: which ad clicks lead to visits and sign-ups. It runs only on superbot.gg's web pages, never in the apps. It sends TikTok the pages you view, the buttons you click, when you submit a form such as the waitlist form or finish signing up, the TikTok click id in the link you arrived by, your IP address and your browser and device type. When you submit a form, it also sends the email address or phone number in it, hashed in your browser first, so TikTok can match the visit to a TikTok account. It sets these first-party cookies on superbot.gg, each kept 13 months from its last use: _ttp, _tt_enable_cookie, ttclid, ttcsid, ttcsid_ followed by our pixel id. TikTok can also set or read its own cookies on its own domains, such as _ttp on tiktok.com and _pangle on pangle-ads.com, and its script may keep values in your browser's storage for superbot.gg. TikTok may also use this data to choose the ads it shows you, under its own terms. The pixel is never loaded in the EEA (which includes Iceland, Liechtenstein and Norway), the UK or Switzerland, or when your browser sends Global Privacy Control. To opt out, turn on Global Privacy Control in your browser, or block cookies for superbot.gg. If you use TikTok, you can also stop it using your activity on other websites for ads in its ad settings.

X

From 5 November 2026, our website at superbot.gg may load the X Pixel to measure our ads on X: which ad clicks lead to visits, so X can report on them and build its website-visitor audience from them. It runs only on superbot.gg's web pages, never in the apps. It sends X the pages you view, the X click id in the link you arrived by, your IP address and your browser and device type. It sets these first-party cookies on superbot.gg, each kept 390 days from its last use: _twpid, which identifies a returning browser to X; _twclid, which holds the X ad click id a visit arrived with; and _twsid, which records your session on our site, so X can tell one visit from another. X can also set or read its own cookies on its own domains, such as x.com, and its script may keep values in your browser's storage for superbot.gg. X may also use this data to choose the ads it shows you, under its own terms. The pixel is never loaded in the EEA (which includes Iceland, Liechtenstein and Norway), the UK or Switzerland, or when your browser sends Global Privacy Control. To opt out, turn on Global Privacy Control in your browser, or block cookies for superbot.gg. If you use X, you can also stop it using your activity on other websites for ads in your personalized-ads settings.

5. Who else processes your data

Each processor receives only what its purpose needs:

Stripe: card payments, subscriptions and auto-recharge.
Public blockchain services, public Monero nodes and our Lightning provider: only when you pay with crypto. A Bitcoin, Lightning or Monero payment goes straight to our own wallets, and our billing service watches the chain to see it arrive. For an on-chain payment, our billing service reads the sending address from the blockchain and checks it against the US Treasury's sanctions list, and a payment from a listed address is never credited. The pay page tells you who sees what, for the kind of payment you pick. For an on-chain payment it says: "To see your payment arrive, we look this address up on public blockchain services, which can see it." For Monero it says: "Our wallet asks public Monero nodes about new blocks to see your payment; they never see this address." For Lightning it says: "Our Lightning provider routes this payment and can see its amount."
Plisio: a third-party crypto payment processor (plisio.net), only when you pay through "Pay with another coin", to process that payment. It receives the amount, our order number and the name of what you are buying, and on its own page the payment you make and the email address you type there, if you type one.
Resend: sign-in codes, invitations, account email and the automation results you ask to have emailed.
Sentry: app crash and error reports, and error reports from our servers (the error type, stack trace, route and request id; no request content, no message text, no account, device or IP identifier).
Apple: push notifications.
Expo: mobile push notifications.
Your browser's push service: a notification to the superbot web app goes through the push service your browser uses (Google, Mozilla, Apple or Microsoft), encrypted so that service cannot read it.
Cloudflare: the network in front of our service, and the router and analytics behind superbot.sh addresses.
Our hosting providers: the servers that run our service and our house model.
Telnyx: calls, SMS and fax. Twilio: text messages in shared rooms.
Deepgram: speech-to-text for voice sessions and cloud dictation, which receives your live microphone audio.
ElevenLabs: voices for voice sessions and calls, voice cloning, and generated sound and music. In a voice session it receives the text of each spoken reply.
Model vendors: with Smart Routing on, a turn we route to a third-party AI provider can carry the messages you send, the photos and files you attach, your approximate location (when location is on) and your profile (your name, and the notes and memory Superbot keeps about you). The providers are Anthropic, DeepSeek, Google Gemini, Hugging Face, Meta, Moonshot AI (Kimi), OpenAI, OpenRouter, Perplexity, xAI and Z.ai (GLM). To turn Smart Routing off, use the Smart Routing switch in the SUPER menu. Then Superbot's own agents answer, and a model you pick yourself still goes to its vendor. Some connections use a shared client identity that is not registered to superbot. Our house model is self-hosted and falls back to Anthropic, but not when Smart Routing is turned off.
Media vendors: OpenRouter, fal, OpenAI, Google, xAI, Meshy, Tripo and ElevenLabs receive the prompt, and any picture you give them to work from, for an image, video, audio clip or 3D model, under our own accounts with them. Google also receives a video you attach in the desktop app, so Superbot can describe what happens in it.
Search engines: the web-search and news tools send a query's search terms to Yahoo, DuckDuckGo, Bing, Brave and Google News. When superbot looks for pictures to guide something you asked it to build, the search terms go to Brave or Serper (which searches Google), and for a game to Steam and itch.io too. When it makes a video thumbnail or a Shorts cover, its search for similar YouTube thumbnails sends the search terms to Serper. Roblox: a search for pictures of Roblox games or Creator Store items sends its search terms, and the id of any Roblox game you linked, to Roblox. Your account details are not sent.
Research and archives: a search for papers sends its search terms, or a paper's DOI, to OpenAlex, arXiv, Crossref and Unpaywall. When superbot looks for an older copy of a web page, the page's address goes to the Internet Archive (the Wayback Machine), Common Crawl and archive.today. Your account details are not sent. A search for code or repositories sends its search terms to GitHub, under our account or under your own GitHub connection if you have one.
Lookup services: a weather lookup sends the place you ask about, which can be where you are, to Open-Meteo, and then that place's coordinates. A lookup of a topic or a word sends it to Wikipedia and Wiktionary (the Wikimedia Foundation) and to the Free Dictionary API, and a photo search sends its search terms to one or more of Wikimedia Commons, Pixabay and Unsplash. A currency or market lookup sends the currencies, ticker, coin, company or data series you ask about to one or more of Frankfurter, Yahoo Finance, CoinGecko, Cboe, Alpaca, the US Securities and Exchange Commission (SEC) and the Federal Reserve Bank of St. Louis (FRED). Your account details are not sent.
Design references: when superbot looks for icons, screens of real apps or interface components, the search terms go to Iconify, Mobbin and 21st.dev. Your account details are not sent.
Software sources: when superbot looks for a tool or package to install for a task, the search terms and package names go to the MCP Registry, Smithery, npm, PyPI, Homebrew, crates.io, Docker Hub, GitHub, OSV (which lists known vulnerabilities) and the package's own registry or website, and OpenAI embeds the words of what it is looking for. Your account details are not sent.
TypeSafe: when this is turned on, it reads a web search's terms and results and monitoring snippets to rank their relevance; it also reads verification challenges on pages the agent visits for you. It is off until it is turned on, and with it off none of that leaves.
WorkOS: single sign-on when a team signs in through its own identity provider; it handles the sign-in and returns the profile.
Trello (Atlassian): a bug report you file is mirrored to a public board, which carries what you wrote, any screenshot or recording you attached, the build and platform, and when you filed it.
Reddit and X: when an ad brought you in and you sign up, a conversion event carrying your hashed signup email, the ad click id and, for Reddit, the browser's user agent is sent to that ad platform. From 5 November 2026, the X Pixel on our website at superbot.gg (section 4) also sends X what section 4 describes.
TikTok: from 23 October 2026, the TikTok Pixel on our website at superbot.gg (section 4). From that date, when a TikTok ad brought you in and you sign up, a conversion event carrying your hashed signup email, the TikTok click id, your IP address, the browser's user agent and the page you signed up on is sent to TikTok.
Meta (Facebook and Instagram ads): from 26 October 2026, the Meta Pixel on our website at superbot.gg, and our server's copy of the same page views, waitlist joins and sign-ups (the Conversions API), which carries the page address, your IP address, the browser's user agent, the Meta cookies and, for a sign-up with an email address, that address hashed (section 4).
Our residential-proxy vendor: outbound web fetches from the region you work in.
Google Fonts: a few marketing pages load fonts from Google, which sees your IP address.
jsDelivr: a shared chat or recipe page that shows math loads its math script and fonts from jsDelivr, and an app superbot builds for you can load code libraries from it. jsDelivr sees the reader's IP address. Google site icons: a link preview shows the website's icon from Google, which sees the website's domain name and your IP address.

Some processors are in the United States. Transfers out of the EEA or the UK rely on the Standard Contractual Clauses, or on the EU-US Data Privacy Framework where the processor is certified under it.

6. Services you connect

Model providers and endpoints you connect (including each leg of a super mode message), MCP servers and connected accounts such as Google, GitHub, Notion and Slack, services whose API key you give superbot (such as Duffel for flights, Roblox or Modrinth), Cloudflare and Railway when you publish, and sites the agent signs in to for you each receive your data because you connected them. Each is a separate controller under its own terms. Disconnect a credential in your account to stop future requests. A website, web address or API you ask superbot to open, read, call or send something to receives that request from our servers, sometimes through our residential-proxy vendor (section 5). So does the website a link in your chat points to, when superbot shows a preview of it.

Gmail. When you connect Gmail, superbot reads only the messages you ask it to, through the gmail.readonly scope. Google's API Limited Use requirements apply: we use Gmail data only to provide the features you see, we do not transfer or sell it, we do not use it for advertising, and no human reads it except for security, to comply with law, or with your consent. The beta capture of superbot Model API requests leaves Gmail data, and the model's replies that follow it in the session, out (details). For a request that read your Gmail, Calendar or Drive data, and for seven days after in that chat, our operational records (the request log, the model-call ledger, tool-call and conversation records, and our model gateway's log) keep a marker and a size in place of what it returned and the replies written from it, never the text.

YouTube. When you connect YouTube, superbot uses YouTube API Services to do the jobs you ask for on your own channel: reading your channel and your videos and their numbers, uploading a video, editing a video's details, setting a thumbnail, adding captions, adding videos to a playlist and moderating comments. What we read goes to our servers and to the AI provider that handles that task, like any other tool result (section 5). We use it only to do the job you asked for and to show you the result. We do not sell it, we do not use it for advertising, and we never use it to train any model. When superbot edits a video you already had, it keeps a copy of the parts it is about to change, so you can put them back. That copy is kept for 30 days, then deleted. YouTube's own terms cover your channel: by connecting it you agree to the YouTube Terms of Service, and Google's privacy policy says what Google does with the data it holds. You can stop superbot's access whenever you like. Disconnect YouTube in superbot and we delete what we stored within 7 days; remove our access on Google's permissions page and we delete it within 30 days. For a question or a complaint about how we handle it, write to [email protected].

Discord. The superbot application on Discord runs /ask for the superbot account you have linked to your Discord account. To link, you sign in to Discord and we read one thing from it, your Discord user id. We drop the access token Discord gives us at once and never store it. If you add superbot to a server, we also keep the server's id and name and which superbot account added it. If you set a default project for a channel with /settings, we keep the channel's id and the project's name. For each command we keep a record of its interaction id, the server and channel it came from and whether it was answered, so a command Discord delivers twice runs once. That record names no account and holds no message text. We delete it after 24 hours. For each command a linked person runs, that person's account also gets an automation record of the command's name, the server, the channel, the Discord user id and the outcome, never the text. The text you type after /ask becomes an ordinary chat turn on your superbot account, in a session for that channel. It is handled like a chat you type in the app (section 2): it goes to the AI providers in section 5, and we never use it to train any model (Training). superbot does not read the other messages in a channel or a DM. The superbot application posts the answer in the channel or in a thread, so everyone who can see that channel can read the command and the answer. The same goes for a message you ask your superbot agent to post to a server you added it to. We use the records above only to run the application, and we share them with no one. Discord is a separate controller of what passes through it, under Discord's privacy policy. We keep your Discord link, the servers you added and your channel defaults until you remove them or delete your account. To stop it, disconnect Discord in superbot, and a command from your Discord user no longer runs on your account. A record that the link ended stays until you delete your account. Disconnect Discord server and the servers you added are removed from our records. The application stays in a server until a server admin removes it. You can clear a channel's default project with /settings. Deleting your account erases your Discord link, the servers you added, your channel defaults and your automation records. Your Discord link, the servers you added and your channel defaults are in the export of your account data (section 7). The command record names no account, so it is not in the export, and deleting your account does not touch it. For a question about how we handle it, write to [email protected].

7. Your rights, and how to make a request

You can ask for a copy of your data, or ask us to correct, delete, restrict or move it, or object to a use of it. Use the export and delete controls at /account, or email [email protected] (also if you cannot sign in).

Export your data at /account gives you a copy of your account and sign-in data, your synced data (chats with the full text of every message, projects, rules, skills, context documents, MCP servers, settings, spaces, teams, shares and access grants), the chat histories you imported from other apps with every message, the conversation transcripts superbot keeps for your chats, your phone and text records, your calls, your billing account (balance, plan and credit buckets), your usage records, payment events and deletion requests, your purchase contracts with the waiver you gave for each (what you bought, what you paid, its payment reference, your answer to the 14-day waiver and any refund), and the records that survive deletion (call and text consent, STOP opt-outs and charge records). Your relay messages, device events, bug reports, account events and daily account snapshots are capped at 1,000 of the newest rows of each. Your phone calls, phone messages, text log, usage records and purchase contracts are capped at 5,000 of the newest rows of each. Imported chat histories are capped at the 5,000 most recently updated chats, and imported messages and conversation transcripts at 100,000 messages each. A field your device stored under a name that marks it as a secret, such as a token, hash or password, is left out of the copy. Rows stored without your account are not in it.

The export includes sign-in data.

Requests are free, and we may confirm your email first. We answer within 30 days in the EEA and the UK (GDPR Art. 12(3)) and within 45 days everywhere else (CCPA §1798.130). We extend only by telling you before the deadline. You can also complain to your data protection authority: in the UK, the Information Commissioner's Office (ico.org.uk); in the EEA, your national supervisory authority (GDPR Art. 77, DPA 2018 s165).

We do not sell your personal data, and we make only four kinds of advertising report. The first is the signup an ad brought us: the ad platform named in section 5 receives a hashed signup email and the ad click id, Reddit also receives the browser's user agent, and TikTok also receives your IP address, the browser's user agent and the page you signed up on. The second is the TikTok Pixel on our website at superbot.gg, which sends TikTok what section 4 describes. TikTok receives neither before 23 October 2026. The third is the Meta Pixel on our website at superbot.gg and our server's copy of its events, which send Meta what section 4 describes. Meta receives neither before 26 October 2026. The fourth is the X Pixel on our website at superbot.gg, which sends X what section 4 describes. X receives nothing from that pixel before 5 November 2026. None is made in the EEA, the UK or Switzerland, or when your browser sends Global Privacy Control. Nobody else receives your data for advertising. We may make de-identified or aggregated data from how superbot is used, data that cannot reasonably be linked to you or your household, and use, share or sell it. We keep that data de-identified, do not try to re-identify it, and require anyone who receives it to promise the same in writing (CCPA §1798.140(m)). Before we first sell any, we email you as section 11 describes. That data never includes anything we get through the Superbot browser extension, which we never sell and never use to train any model (section 3).

If you are not a superbot user. Data about you can reach us, for example as the author of a post that built an app, or in a call or message someone sent you. You can ask us for a copy of it, or to correct or delete it, at [email protected] (GDPR Art. 14).

8. Automated decisions

No decision about you, such as your billing, your account standing or deleting your data, is made solely by automated processing with a legal or similarly significant effect (GDPR Art. 22). If an automated step gets something wrong, ask us through section 7 and a person will review it.

9. Children

superbot is for people aged 13 or older, and you must be 18 or older to buy. We do not knowingly collect data from anyone under 13. If you believe a child has given us data, write to [email protected] and we will delete it.

10. Security

Only your devices can read

Sealed on your device under a key only your devices hold, so we cannot read them:

browser sessions
saved passwords
SSH keys
imported MCP headers
publishing grants
team vault items
model sign-ins sealed on an enrolled device

We can read

Encrypted at rest under keys our servers hold:

synced chats and settings
pasted API and endpoint keys
a copy of each superbot API key you create
Git tokens (sent to your account so your phone can list your repositories)
MCP headers you add by hand
connected-account tokens
project mirrors

Stored as readable data:

imported chat histories
logs and telemetry

If a breach is likely to put your rights at risk, we tell the supervisory authority within 72 hours and tell you without undue delay (GDPR Art. 33 and 34). Report a security issue to [email protected].

11. Changes

We may change this notice at any time, and publish each change here with a new version date. A minor change, such as a correction or a clarification, takes effect when we publish it.

A material change is emailed to you before it takes effect: at least 30 days before if you live in the EEA or the UK, and at least 7 days before everywhere else. Material changes include any new recipient or processor of your data, any new transfer of it out of the EEA or the UK, any new purpose, any new class of data and any longer retention. The email links to the new version, says what changes and the date it takes effect. Where the law needs your consent, we ask for it.

If you do not accept a material change, you can delete your data or account before it takes effect, and a consumer can end a paid plan free of charge with the refund in section 12 of the terms.

Change log. Version 2026-10-12: during the beta, the text of superbot Model API requests and replies is kept 180 days and used to evaluate, train and improve superbot's own models, including from accounts with Anonymize my data on. It takes effect on 2026-11-06.

Version 2026-10-15: section 5 now says what a routed turn can carry to a third-party AI provider (the messages you send, the photos and files you attach, your approximate location and your profile), names Google Gemini, and says how to turn Smart Routing off. It also says that a voice session sends its microphone audio to Deepgram and the text of each spoken reply to ElevenLabs. It also says that a house turn that is down does not fall back to Anthropic when Smart Routing is turned off. The short version says which copy of chat text is kept 180 days. The recipients, purposes and retention periods are unchanged.

Version 2026-10-16: section 4 now says that our website at superbot.gg may load the TikTok Pixel to measure our TikTok ads, what it sends to TikTok, the cookies it sets and how to opt out. Section 5 names TikTok, which also receives a conversion event when a TikTok ad brought you in and you sign up, and section 7 says which advertising reports we make. Neither reaches TikTok from the EEA, the UK or Switzerland, or when your browser sends Global Privacy Control. The apps are unchanged. It takes effect on 2026-10-23.

Also in version 2026-10-16: section 3 now describes the Superbot browser extension, including what it reads, where that goes, how long it is kept, the Share sign-ins switch and how to stop it. Sections 2, 3 and 7 now say that data we get through the extension is never sold, never part of de-identified data and never used to train any model. What the extension reads goes to the same places, and is kept as long, as the rest of a chat. This applies only if you install the extension, and it takes effect when we publish it.

Version 2026-10-17: section 3 now says how the Superbot browser extension saves a file you ask for (the Save downloads switch, off until you turn it on, into a Superbot folder in your Downloads) and attaches a file you ask for to a page (a private temporary copy on your computer, once you allow the extension access to files). Nothing else it reads or sends changed. This applies only if you use the extension.

Version 2026-10-18: section 6 now says how superbot uses YouTube API Services when you connect YouTube: what it reads and does on your own channel, the 30 days it keeps a copy of what it changes, the YouTube Terms of Service and Google's privacy policy, and how to stop its access. This applies only if you connect YouTube.

Also in version 2026-10-18: section 2 now covers signing in with TikTok. TikTok sends us your account id (union_id), your display name and your photo. We keep the id and the name, and nothing else TikTok sends. The access token is revoked at sign-in and never stored. This applies only if you sign in with TikTok. It takes effect on 2026-11-17.

Also in version 2026-10-18: section 5 now says that Smart Routing is turned off with its switch in the SUPER menu, instead of a card that asks before the first send. What a routed turn can carry, the providers and the house-model fallback are unchanged.

Also in version 2026-10-18: section 6 now says what the superbot application on Discord keeps (your Discord user id, the servers you added it to, a channel's default project, and a record of each command, never the message text), that the text you type after /ask is an ordinary chat turn that we never use to train any model, and how to stop it and have it deleted. This applies only if you use superbot on Discord.

Also in version 2026-10-18: section 6 now says how long superbot on Discord keeps what it stores. The record of each command is deleted after 24 hours. Your Discord link, the servers you added and your channel defaults are in your data export. Deleting your account now erases your channel defaults too, project names included. This applies only if you use superbot on Discord. It takes effect when we publish it.

Version 2026-10-19: section 4 now says that our website at superbot.gg may load the Meta Pixel to measure our ads on Facebook and Instagram, what it sends to Meta, the cookies it sets and how to opt out, and that our server sends Meta a copy of the same page views, waitlist joins and sign-ups through Meta's Conversions API, with the email address of a new account hashed. Section 5 names Meta for this, and section 7 says which advertising reports we make. None of it reaches Meta from the EEA, the UK or Switzerland, or when your browser sends Global Privacy Control. The apps are unchanged. It takes effect on 2026-10-26.

Also in version 2026-10-19: section 3 now says that the Superbot browser extension takes a task from its own popup or side panel, reads the title and web address of the tab you are on while they are open and passes them on only when you send a task, keeps a task you type for up to 24 hours when the Superbot app is not installed, and reads your browser's version. It also says that Share sign-ins and Save downloads turn on when you press Allow and your browser agrees, and that an earlier install keeps its choice. Nothing else it reads or sends changed. This applies only if you use the extension.

Version 2026-10-21: section 2 now says that if you delete your account while a charge is still owed, we keep the record of that charge until it settles, and then delete it.

Also in version 2026-10-21: section 5 now says that Resend also sends the automation results you ask to have emailed, and that a search for pictures of Roblox games or Creator Store items sends its search terms, and the id of any Roblox game you linked, to Roblox.

Also in version 2026-10-21: section 5 now says where a search for pictures to guide something you asked superbot to build sends its search terms (Brave or Serper, and Steam and itch.io for a game), and that a search for YouTube thumbnails sends them to Serper. Section 4 now names the sbtrial cookie, which ties your browser to a free trial API key you asked for, for up to 28 days.

Also in version 2026-10-21: section 2 now says that when you search your imported chats, or Superbot searches them for you, the search words go to OpenAI to be embedded, and that this embedding is not stored.

Also in version 2026-10-21: section 5 now also names these services that superbot's tools send your words to, and says what each receives: OpenAlex, arXiv, Crossref, Unpaywall, GitHub, the Internet Archive, Common Crawl and archive.today for research; Open-Meteo, the Wikimedia Foundation, the Free Dictionary API, Pixabay, Unsplash, Frankfurter, Yahoo Finance, CoinGecko, Cboe, Alpaca, the SEC and FRED for lookups; Iconify, Mobbin and 21st.dev for design references; the registries searched for a tool to install, and OpenAI for the words of that search; your browser's push service; jsDelivr and Google's site icons. It also says that media vendors receive any picture you give them and that Google describes a video you attach in the desktop app. Section 4 now says our server may send Google Analytics a sign-up or waitlist step from the web. Section 6 now covers services whose API key you give superbot, and the websites you ask superbot to open or that a link preview reads.

Also in version 2026-10-21: section 4 now says that the step our server may send Google Analytics includes starting as a guest in the web app, which is sent with no account or guest id.

Version 2026-10-28: section 4 now says that our website at superbot.gg may load the X Pixel to measure our ads on X, what it sends to X, the cookies it sets and how to opt out. Section 2 says when, section 5 names X for the pixel as well as for the sign-up report, and section 7 says which advertising reports we make. The pixel reaches X from none of the EEA (which includes Iceland, Liechtenstein and Norway), the UK or Switzerland, or under Global Privacy Control. The apps are unchanged. It takes effect on 2026-11-05.

Also in version 2026-10-28: section 3 now says that the Superbot browser extension can work without the Superbot app. When it works on its own, because the app is not installed or because you choose Standalone in its Settings, it connects to our servers and sends them the tasks you type, the title and web address of their pages and what it reads in the tabs it works in, makes a guest account for your browser the first time you send a task (if you chose Standalone, it asks you to sign in first), and keeps a Superbot sign-in and a random install id in your browser. It also says that we count how far a new use of the extension gets. Share sign-ins, Save downloads and File uploads still need the Superbot app. With the app, as Companion, nothing changes. This applies only if you use the extension on its own. It takes effect on 2026-11-29.

Version 2026-11-03: section 5 now says who else can see a payment you make with crypto, such as Bitcoin, Lightning or Monero: public blockchain services, public Monero nodes and our Lightning provider, as the pay page tells you for the kind of payment you pick. It also says that our billing service checks the sending address of an on-chain payment against the US Treasury's sanctions list, and that Stripe processes card payments, subscriptions and auto-recharge. Section 2 now says that a crypto payment goes to our own wallets and that our billing service keeps its invoice. This applies only if you pay with crypto, and it takes effect when we publish it.

Version 2026-11-04: section 5 now names Plisio (plisio.net), a third-party crypto payment processor, which processes a payment you make through "Pay with another coin", and says what it receives. Sections 2 and 5 now say that Bitcoin, Lightning and Monero payments go to our own wallets, and section 2 that our billing service keeps the invoice of every crypto payment. This applies only if you pay with crypto, and it takes effect when we publish it.

Also in version 2026-11-04: section 2 now says that the project folder mirror is on by default, and how to turn it off for all your projects or for one. This corrects section 2, which said the mirror was off until you turned it on for a project.

Also in version 2026-11-04: sections 2 and 3 now name the switch in Settings > Privacy in the desktop app that turns off each read from your device. Section 2 also says that computer use starts at Ask in new installs, so the assistant asks before it captures your screen, that the password scan runs at startup only when the last scan finished more than 12 hours before, that turning off Share usage counts stops the relay sending the records under Relay and tool logs, and that Anonymize my data is on by default in new installs of the app. This corrects section 2, which said screenshots were off unless you turned computer use on, that the password scan ran at every startup and that no setting turned off the relay and tool logs, and section 3, which said each read was a separate setting.

Also in version 2026-11-04: section 2 no longer lists an Apple ID email you give us to test the iPhone app, and section 5 no longer says that Apple receives it.