Privacy Policy

1. Summary

Ventrin is a tool that helps your team use AI assistants without leaking confidential content. It runs in two places: as a Chrome extension on your team's computers and as an API gateway that internal apps can call.

• The extension scans prompts in your browser before they reach ChatGPT, Claude, Gemini or Microsoft Copilot.
• The gateway scans API requests before forwarding them to the same providers using your organisation's stored credentials.
• We log a record of each decision (allowed, sanitised, blocked) so admins can audit usage. By default the prompt body is masked; full encrypted bodies are only stored if a workspace admin opts in.
• We do not train AI models on your content. We do not sell, share or transfer your data to third parties for marketing.

2. Who we are

Ventrin is operated by the team at ventrin.com. For the purposes of UK GDPR, Ventrin is a data processor when handling your team's prompts and a data controller for the account and billing data of the workspace administrator.

Privacy queries: hello@ventrin.app. Security disclosures: security@ventrin.app. See security.txt.

3. What we collect

From the workspace admin (account data)

• Email address (used for sign-in).
• Workspace name, vertical (legal / healthcare / generic), and chosen plan.
• Authentication metadata from Firebase Auth: hashed password, ID-token expiry, last sign-in time.

From end users (the team using the extension)

• Email address used to redeem a workspace invite.
• One row per evaluated prompt, containing: timestamp, the destination tool, the action Ventrin took, the policy categories that fired, a risk score, and a masked preview of the prompt. The masked preview keeps the first three characters of each word and replaces the rest with bullets.
• The browser extension keeps a cached Firebase ID token in chrome.storage.local so it can call our backend without rehydrating Firebase Auth on every prompt. This is isolated to the extension and not visible to web pages.

Optional, off by default

• Full body logging for the API gateway. When a workspace admin enables this, the original and sanitised prompt bodies are stored AES-256-GCM encrypted. Only admins can reveal them, and every reveal writes a separate audit row.
• Provider API keys (OpenAI / Anthropic / Gemini) the workspace uses, stored encrypted with the same scheme.

4. Why we collect it

Our lawful bases under UK GDPR Article 6:

• Performance of a contract for everything needed to deliver the service: account data, decision logs, audit trail, authentication.
• Legitimate interest for the masked preview of each prompt, which gives admins enough context to review usage without exposing the underlying content.
• Consent for full-body logging. This is opt-in by the workspace admin and can be turned off at any time.

5. What the extension does on your device

Run a deterministic regex layer on the outgoing prompt locally:

• Looks for credentials, API keys, emails, phones, postcodes, IBANs, NHS numbers and other regex-detectable identifiers.
• Looks for terms in the workspace's policy packs (e.g. "matter", "patient", "client", "DOB").
• Looks for any names the workspace admin added under Protections → Tracked Names.

That detection runs in the page's content script before the prompt leaves the browser. The result (allow / warn / sanitise / block) is decided locally for blocks of credentials and locally-detected hard patterns; for borderline cases we call the Ventrin server for a second opinion. We do not call any third party for the local detection step.

The extension does not collect browsing history. It only activates on ChatGPT, Claude, Gemini, Copilot, and our own ventrin.com domains (the latter to handshake with the dashboard at sign-in).

6. What the API gateway sees

When your application calls https://www.ventrin.com/v1/chat/completions:

1. We hold the request body in memory while the policy engine runs (typically tens of milliseconds).
2. We log the metadata + a masked preview to gatewayLogs.
3. If the workspace has full-body logging enabled, we encrypt and store the original and sanitised bodies. Otherwise they are dropped from memory once the response is returned.
4. We forward the request to your chosen provider using your workspace's encrypted provider credential.
5. The provider's response passes back through us to your application. We do not store the response body.

7. Who we share it with

Three categories:

• Sub-processors we use to deliver the service: Google Cloud (hosting, Cloud Functions, Firestore, Firebase Auth, all in the EU europe-west1 region). Google Fonts (CSS for the dashboard typography).
• The AI provider you choose (OpenAI, Anthropic, Google Gemini). The forwarded prompt — already sanitised if the decision was "sanitised" — leaves our perimeter and reaches the provider you configured.
• No-one else. We do not sell your data. We do not use it for marketing or advertising. We do not train any AI model on your content.

8. Retention

During early access, decision-log retention is unbounded. We are working on configurable retention (14 / 30 / 90 / 180 / 365 days) before paid plans launch. You can request manual purge at any time via the controls below.

When a workspace is deleted, all of its data — events, gateway logs, policies, tracked names, encrypted bodies, and stored provider keys — is removed within 30 days.

9. Your rights

Under UK GDPR you have the right to:

• Access the personal data we hold about you (subject access request).
• Have it corrected if it's wrong.
• Have it erased ("right to be forgotten"). We provide two admin-only callables — purgeEventsForUser and purgeGatewayLogsForUser — that delete a user's events and gateway logs within the workspace.
• Restrict or object to processing.
• Port your data to another service (export from the dashboard).
• Lodge a complaint with the UK Information Commissioner's Office (ICO) — ico.org.uk — if you believe we've handled your data improperly.

Email hello@ventrin.app with requests. We aim to respond within 30 days.

10. Security

• All transport is HTTPS only with HSTS preload.
• Stored prompts and provider credentials are encrypted at rest with AES-256-GCM. The encryption key lives in a server-side secret, not in the database.
• Workspaces are tenant-isolated at the database row level. Firestore security rules enforce that no client can read another workspace's data.
• API key plaintext is shown once at creation and stored only as a SHA-256 hash. Rotation is supported with a 24-hour grace.
• Reveal of a stored prompt requires step-up authentication (re-sign-in within the last 5 minutes) and writes an audit row.
• Per-IP rate limits and failed-auth bans are applied to the gateway endpoint.

Our most recent third-party security review and the residual risks we acknowledge are summarised internally; reach out to security@ventrin.app if your procurement process needs them.

11. International transfers

Ventrin's compute and storage run in europe-west1 (Google Cloud, Belgium). When you choose an AI provider, your prompt is forwarded to that provider's endpoint, which may process it outside the UK or EEA. The provider's own privacy policy applies to that processing. We do not transfer data to third parties beyond the provider you configured.

12. Changes to this policy

We will revise this policy as the product evolves. The version number and date at the top of the page indicate the last change. Material changes will be highlighted on the dashboard at next sign-in.

13. Contact

Privacy questions: hello@ventrin.app.
Security disclosures: security@ventrin.app.
ICO complaints: ico.org.uk.