The short version

gbrain.io is operated by The Overton Window Company, Inc., based in San Francisco, California. We host GBrain, the open-source knowledge layer for AI agents. Each brain you create lives on dedicated compute we run only for your brains. Not a shared database. Not a machine that runs anyone else’s brains. That compute belongs to your workspace until you delete it.

We look at your team’s content only to keep the service working: automated checks for failures, debugging and support. We never use it for marketing, we never sell it, and we don’t train AI on it. Your team’s notes, and the database behind them, stay on your brain’s compute. Chat history and the files you drop into chat are kept in our own database and file storage so the web app can show them; Move your GBrain out of a workspace lists what lives where. The AI provider that processes prompts on your behalf sees the prompt for that request only, under that provider’s published policy, and a workspace can limit its AI calls to providers it approves, such as one in the EU or its own. We don’t pipe your content anywhere it doesn’t have to go. Subprocessors lists every company we send it to.

Like every hosted service, we have operator-level access to the infrastructure we run. Security covers what that means in practice and where the limits are.

1. Scope and definitions

“Customer” means the organization (workspace) that has agreed to the Terms of service.

“Customer data” means anything written to your brains (by you, by your workspace’s members, or by an AI client you’ve connected) and anything your brains generate in response, along with the configuration you provide and audit metadata about activity in those brains.

“Personal data” means information about identifiable individuals. Some customer data may be personal data (an email message addressed to an individual, for example). Where we process personal data on your behalf, we’re acting as a processor and you’re acting as a controller in the sense those terms are used in GDPR, CCPA, and similar frameworks.

This page is the controller-facing description of how we handle personal data. Our Data Processing Addendum is part of the Terms and applies automatically; a countersigned copy is available on request at support@gbrain.io.

2. Data we collect

We collect three categories of data:

• Account data. The email address you sign in with (asserted by Google when you use Google sign-in, or verified by an email magic link when you use that path), the workspace name and URL identifier you choose, the role of each user in the workspace, and billing identifiers (Stripe customer id).

• Customer data. Anything you or an AI client you’ve connected writes to your brain (markdown notes, structured records, metadata) and anything your brain writes in response. Customer data lives on your brain’s compute as described in section 5, except chat history and files dropped into chat, which live in our control plane.

• Operational telemetry. Request logs (HTTP method, path, status, duration, user id, IP address and browser), structured app logs from our own code, error reports when something fails, and platform-level health metrics from our compute provider. We use this to debug, monitor health, and prevent abuse. We don’t try to read into request payloads.

• Where a visitor came from. When someone signs up from one of our public pages, the page they landed on, the site that sent them and any campaign tags in the link are carried in the sign-in link and saved on the new account, so we can count where customers come from. No cookie is involved.

Some teams put special categories of personal data into a workspace, such as health information or HR and legal files. The service applies no safeguards to that data beyond the ones on this page and in Security, and some optional features send data to companies without a transfer safeguard. A team with that kind of data should read A workspace for HR, legal or health data under GDPR before it starts.

Cookies

We set one cookie: _gbrain_key, which keeps you signed in. It is strictly necessary, encrypted so only our servers can read it, and lasts 120 days from your last visit. We set no advertising, analytics or tracking cookies, on our public pages or in the app. Plausible, which counts page views on our public pages, uses no cookies. The app keeps a few display preferences, such as whether to show the model’s reasoning, in your browser’s own storage; the app reads them to draw the page, and we don’t store them.

3. How we use data

• To operate the service. Authenticating sessions, routing requests to your brains, sending the prompts a brain constructs to the AI provider, returning the response, billing your subscription. Keeping it working includes an automated check in which an AI model reviews recent conversation snippets, with secrets removed, to catch failures; what it finds goes to our team.
• To support you. Responding to questions, debugging tickets you open, communicating about your account.
• To improve the service. Aggregate, non-identifying patterns (which features are used, where errors cluster, how long brain operations take). We do not use customer content for marketing, and we do not feed customer data into model training.
• To meet legal obligations. Tax records, security incident reports, court orders we can’t reject.

We don’t sell, rent, or syndicate customer data or personal data to anyone.

Google user data — Limited Use

GBrain’s use and transfer of information received from Google APIs to any other app adheres to the Google API Services User Data Policy, including the Limited Use requirements.

When you connect a Google account to a brain, the brain accesses only the Gmail data needed to do what you ask it to — reading, searching, organizing, and sending mail on your behalf. We do not use this Gmail data for advertising, we do not sell or transfer it, and we do not use it to train or improve generalized AI/ML models. No human reads it except where you explicitly ask us to (for example, a support ticket you open), where it’s necessary for security, or where the law requires it. Access tokens are short-lived and minted per request; the long-lived refresh token is encrypted at rest, and revoking the grant in your Google account — or disconnecting the account in your workspace — destroys the credential we hold.

4. Operator access

We can see enough to run the service, and you should know what that means rather than infer it.

Health and performance. We monitor the platform continuously: request rates and latencies, error rates, queue depth, machine and storage health, and the volume and outcome of tool calls a brain makes. This is how we notice a brain is wedged before you do. It is metadata about the shape of the work, not the content of it.

Getting a brain working again. When a brain is broken, fixing it sometimes needs a person to look at its actual state: its configuration, its logs, the structure of what it has stored. We do this to restore service, and we do the least that answers the question.

Hands-on setup. We offer to set brains up personally, and that is exactly what it sounds like: with your agreement, we work inside your brain the way you would. If you accept that offer, you are inviting us in, and we will say so plainly at the time rather than treating your acceptance as blanket consent for later.

What we don’t do. We don’t read your brain’s contents out of curiosity, we don’t use them for advertising, we don’t sell or transfer them, and we don’t use them to train generalized models. Access is limited to people who need it.

The staff-access record. Each time someone at GBrain, or GBrain’s support AI, opens a workspace through GBrain’s own systems, it is recorded before it opens, and if it cannot be recorded it does not open. That covers a page about the workspace in our staff console, our support AI looking inside its machine, a shell or command opened in it, and a staff member making themselves an admin of its organization, which also emails the organization’s admins. The record says when, who, what and why, cannot be edited or deleted, and is on each workspace’s Data protection page for its admins. It does not cover our support AI’s reads of GBrain’s own records database, which span every customer and are recorded centrally; our background job queue; or direct access to our servers and database, which we keep for outages and note in the incident record.

If you would rather we never look, say so and we will note it on your account. The trade is real: some problems cannot be diagnosed from metrics alone, and the answer may be that we cannot fix it for you.

5. Where data lives

Customer data lives on the dedicated compute we provision for each brain. Each brain has its own compute and its own encrypted storage volume, sized to the brain’s plan tier. No two brains share storage. Storage is encrypted at rest by the platform. Our servers, and every brain’s compute, are in Fly.io’s Chicago region in the United States.

Chat history and files dropped into chat live in our control plane: messages in the managed database described below, files in our object storage, which keeps copies near where a file is read and so may hold a copy outside the US. Move your GBrain out of a workspace lists which parts of a workspace live where.

Account data lives in our control plane in a managed database, encrypted at rest, in the same country. OAuth tokens for connected sources get a second layer of encryption at the application layer (AES-GCM) before they’re written.

Operational telemetry lives in our compute provider’s logging and metrics systems, plus our own log aggregation. Retention is in section 7.

AI processing. When a brain calls an AI model, the call passes through our servers in Chicago, which add the key and forward it to the provider. Where the provider processes it is listed on Subprocessors: Anthropic routes calls to the US, Europe, Asia or Australia, and OpenAI’s endpoint is global. An organization’s admins can add its own AI endpoint, for example an EU-hosted model or one on their own servers, and set a workspace’s AI routing to Only approved on the workspace’s Data protection page. Then every AI call from that workspace goes only to the destinations an admin approved, and we refuse any other before it leaves our servers, the optional search and lookup tools included. The same page shows where each of the workspace’s recent AI calls was processed. AI routing does not cover Telegram, iMessage or Slack, which reach their own services directly, or the accounts people connect; each of those is turned off on its own page.

If you require data residency outside the United States, talk to us at support@gbrain.io. Storing a workspace’s brains in the EU is a known need we’ll size based on customer demand; chat history and account data would stay in the US.

6. Subprocessors

We use subprocessors to deliver the service: the companies that host it, bill for it, send its email, run its AI models and power its optional tools. Subprocessors lists every one, with what it does, the data it receives, which workspaces it touches, where it processes data and the safeguard for transfers from Europe.

Applications, accounts and AI clients a workspace connects itself, such as Slack, GitHub, a person’s own Google or Microsoft account, its own AI endpoint, or an application from the directory, are chosen and directed by the customer. They aren’t our subprocessors, and the customer’s own terms with them govern what they do.

Adding or changing a subprocessor for material processing of customer personal data triggers email notice to workspace admins at least 30 days before the change takes effect. For emergency replacements (a subprocessor going out of business, for example) we’ll notify as soon as practical.

If a customer’s policy requires named subprocessor approval, that’s negotiable in a separate contract.

7. Data retention and deletion

* Not yet in force. These are the periods we will hold ourselves to. For most of them the code that applies them already runs, counting what it would remove and removing nothing; the deletion of lapsed workspaces and of accounts never used is still being built. Until we switch a period on, the data in its row is kept. We will switch each one on no earlier than 30 days after we email workspace admins about it (section 14), and update this page when we do.

Deleting a brain is a one-way door. The compute is destroyed, the storage volume is destroyed, its chat history is deleted, any credentials we held for the brain’s connected sources are destroyed with it, and its snapshots age out as above. Files dropped into its chats are kept with the workspace until the workspace is deleted. We don’t keep a separate “in case” archive. Revoking the connected source’s access at the provider itself (cutting Google’s OAuth grant, for example) is something you can do at any time from that provider’s account-settings page; we don’t do it automatically.

Deleting a workspace removes all of its brains and all account data tied to it, subject to the billing-records exception above.

8. International transfers

The service is operated from the United States. If your team uses the service from another country, your data is being transferred to the US for processing. By using the service, you authorize us to make those transfers as needed to operate it.

For personal data from the European Economic Area, the UK or Switzerland, our Data Processing Addendum incorporates the EU Standard Contractual Clauses (Module Two, or Module Three where the customer is itself a processor), with the UK and Swiss addenda. Our subprocessors receive data under their own Clauses or the EU-US Data Privacy Framework, as Subprocessors shows for each; two with neither run by default (Firecrawl for fetching web pages, and ZeroEntropy for memory search in older brains), and setting a workspace’s AI routing to Only approved stops both; the rest with neither are off unless a workspace turns them on, apart from Terminalwire, which serves our command-line installer to whoever downloads it.

We have assessed these transfers against US law, following the European Data Protection Board’s Recommendations 01/2020. Our conclusion: ordinary business data can be transferred under the Clauses with the safeguards on this page and in Security. Special-category data, and HR, legal or health work, carry more risk; for those we recommend AI routing set to Only approved with an EU model or your own, and the optional features without a safeguard left off (the checklist). A customer’s data protection officer can ask us for the full assessment at support@gbrain.io.

9. Your rights, and your workspace’s

Your own account. Signed in, you can do these yourself, from your account:

• Download your data. Data export (/users/<your id>/data-export) builds a copy of what GBrain holds about you as a zip of JSON files: your profile, memberships, sign-ins and sign-in identities, SSH keys, applications and connected accounts (never their keys or tokens), your activity, the chat messages you wrote, the files you uploaded, what you bought, and the AI usage of workspaces you pay for. The link works only for you, signed in, and expires after 7 days. Files you dropped into chats are listed with a link to each; the files stay in their workspaces.
• Change your email. Change email sends a link to the new address. The change happens when you confirm it from that link; then the old address is told, and your other sessions are signed out.
• Delete your account. Delete account lists what goes and what stays before you confirm. It deletes your sign-ins, keys and notifications, the organizations only you are in with their workspaces, and your Stripe customer record, and removes your name, email and picture from what stays. It refuses, and says what to do first, while you pay for or are the only admin of an organization other people use, or have a paid subscription that still renews. Billing records stay for tax, and what you wrote in workspaces other people use stays with those workspaces, which belong to their organization.

A workspace’s data. The organization a workspace belongs to is the controller of what is in it, so its admins answer for it. They can:

• Delete a brain (and the data on it) at any time.
• Delete the workspace.
• Download a copy of what GBrain holds about the workspace (its members, its chats and every message in them, a list of its files, and its AI usage) from its Data protection page, and move its memory out as Move your GBrain out of a workspace describes.

Where individuals have rights under GDPR, CCPA, or similar laws (access, rectification, erasure, portability, restriction, objection) over data that appears in a workspace, the workspace’s admins are the appropriate channel. We support them in fulfilling those requests.

If you’re an individual and you can’t reach the workspace admin, write to support@gbrain.io with enough detail to identify your request and we’ll route it.

If you’re in the EU, the UK or Switzerland

For the account, billing and security data we hold as a controller, our legal bases are performing our contract with you, our legitimate interests in securing, monitoring and improving the service and in counting where our customers come from, and legal obligations such as keeping tax records. We send product news by email to people in Europe only once they’ve said yes, and every such email has a one-click unsubscribe. Besides the self-serve tools above, you can ask us at support@gbrain.io to restrict or object to the use of your data, or for anything the tools don’t cover. You also have the right to complain to your local data protection authority. Our representative in the EU and the UK is named in section 15.

10. Security

The architectural details (brain isolation, encryption, access controls, vulnerability handling, incident response) live in Security. Two facts a procurement reviewer should know without digging:

• The service is not end-to-end encrypted. Your brain reads your team’s content in cleartext so an AI model can summarize it. That’s intrinsic to the product. Anyone offering “end-to-end encrypted” hosted summarization is misleading you.
• We do not hold SOC 2, ISO 27001, or comparable third-party certifications. The Security page describes the practices in place. Certifications follow when customers require them.

11. Breach notification

If we become aware of a personal data breach affecting customer data, we’ll notify affected workspace admins by email without undue delay, and in any case within 72 hours of becoming aware of it, with the details we have at the time and updates as the investigation proceeds. That is the commitment in our Data Processing Addendum section 8.

Where law (GDPR, CCPA, state breach notification laws) prescribes a specific notification window or content, we’ll follow it.

12. Legal requests

If we receive a subpoena, court order, or other legal demand for your data, we will:

• Notify you, unless we are legally prohibited from doing so.
• Push back on requests that look overbroad.
• Produce only what we actually have. We don’t retain customer email or document inboxes; we have account records, workspace metadata, audit logs, chat history and files dropped into chat, and the customer data on the relevant brain’s storage at the time of the request.

Where law allows, we will summarize legal requests received and our response in a public transparency note, with identifying details redacted.

13. Children’s data

The service is not intended for individuals under 18. We don’t knowingly collect data from children. If you discover a child has signed up, write to support@gbrain.io and we’ll delete the account.

14. Changes to this page

We maintain a public change history for this page: every published version stays readable at its own address, such as the version of 30 October 2026. Material changes (new categories of data we collect, changes to the subprocessor list, changes to retention periods) get email notice to workspace admins at least 30 days before they take effect, except for changes required by law or to prevent immediate harm. The Updated date at the top moves with each change.

15. Contact

• Privacy questions and requests: support@gbrain.io
• Legal notices: support@gbrain.io
• Everything else: reply to any email from us.

Our representative in the EU and the UK. Under Article 27 of the GDPR, and of the UK GDPR, we have appointed Data Protection Representative Limited, trading as DataRep, as our representative: 77 Camden Street Lower, Dublin, D02 XE80, Ireland. You and data protection authorities can contact them instead of us about anything on this page.

TODO: DataRep’s contact form or email address for data subjects goes here once their onboarding pack arrives. Until then, write to support@gbrain.io or to DataRep at the postal address above.

Data Protection Officer. We have not appointed one, because the GDPR does not require one for our processing: we don’t monitor people systematically on a large scale or process special categories of data as a core activity. Privacy questions go to support@gbrain.io.