How Locke protects your data

Sonomos is the company. Locke is its product: a desktop app that masks sensitive data before it reaches AI models, so you can use AI on confidential work.

When you send a message, Locke checks it for sensitive information and, when protection is needed, swaps it for one-way placeholders, so raw sensitive data never reaches external models. Everything runs on your device. Reversible tokenization is planned shortly after Locke 1.0.

Not your content. Detection and masking happen on your device, and your prompts, files and detections never reach Sonomos servers. Locke does need a Sonomos account: it signs in, periodically confirms your licence (account ID and plan tier only), and syncs your settings (which AI services are monitored, which categories are on) across devices. We do not sell data. The Locke Privacy Policy has the full list.

Locke shows a three-state indicator. Green: nothing detected. That is a detection result, not a guarantee the text is safe to send. Amber: possibly sensitive, so take a look. Red: sensitive data found, so fix it before sending.

Yes. Choose which types of information Locke detects: personal identifiers, client names, financial data, health information or custom terms for your field. Each rule can be switched on or off, or tuned for sensitivity.

Yes. Detection and automatic masking toggle separately. Pause detection for non-sensitive tasks, or turn off automatic protection while you review flagged content. The status indicator always shows what's active.

Personal identifiers (SSNs, dates of birth), contact details (emails, phone numbers), financial data (credit card numbers, bank accounts), health information, and client and customer names. You can add custom patterns for your industry.

Locke sits between you and external AI systems. It checks input before data leaves your environment, so you can keep using your AI tools and control what they receive.

Yes. Locke is designed for people who can't afford a mistake: lawyers, investigators, consultants, healthcare professionals and others in regulated industries, where exposing data creates liability.

Three hard rules: local-first processing, minimal exposure and visible system state. If a behavior can't be shown on screen, it doesn't ship.

It is designed not to. Detection runs locally, and only when content is on its way to an AI tool, not continuously. We are still validating performance across hardware and will publish numbers once we have them.

Locke is not a HIPAA-covered entity or a business associate. No software is "HIPAA compliant" on its own: that status attaches to your organisation and how it works, not to a tool. Detection and masking run on your device, so Sonomos does not receive PHI. Locke can help reduce the PHI that reaches an AI tool, but detection is probabilistic and can miss things. Whether your use satisfies HIPAA is for your compliance team to decide. Any BAA with the AI provider is between you and them.

Detection and masking run entirely on your device, so your content is never sent anywhere for analysis. The app must be online to sign in and to confirm your licence periodically, and the AI platforms you use (ChatGPT, Claude and others) need the internet anyway. Locke can also screen on-device model servers such as Ollama once the client points at its local proxy.

Yes. The Reports screen generates reports on your device, each with a report ID and a verification digest so you can check a copy against the original. Teams includes full compliance and technical reports, plus CSV or PDF exports of your detection and masking history. Free includes one basic Privacy summary a month. Reports cover protection metadata, never prompt content, and nothing is sent to Sonomos.

When you send it, Locke flags it with a color-coded indicator, amber for medium risk and red for high, and explains why. With automatic masking on, the data is masked before it reaches the AI, and the indicator updates to the current risk level.

Locke is an Early Access desktop app for x64 Windows and Linux, with macOS coming soon. With the Locke Extension it screens AI websites in your browser. On its own it screens command-line agents, scripts and SDK calls, including GitHub Copilot.

On your device, Locke’s router tells AI-service traffic from other traffic, including eligible services outside its built-in catalog. Recognizing a service does not mean every request is screened. In the browser, the Locke Extension screens AI websites by holding the request inside the page. A proxy on your own machine screens command-line tools and API traffic: Claude Code, gemini-cli, GitHub Copilot, curl and the official SDKs. Local model servers such as Ollama and LM Studio join them once the client points at that proxy. Locke also recognizes and screens desktop wrapper apps such as ChatGPT desktop, Claude desktop and Microsoft Copilot desktop. Coverage depends on the app, platform, request type and settings. Detection and masking happen 100% on your device.

Now, in Early Access for x64 Windows and Linux. macOS is coming soon. Download Locke, sign in, and join the Early Access waitlist inside the app.

Exactly two. Free gives one person the complete protection engine and one basic Privacy summary report a month. Teams is for teams, companies and organizations that need insight into, and control over, how their members use AI. It costs $14.99 per seat a month for the first 5 seats, then $10.99 per seat a month, and adds unlimited full compliance and technical reports, policies admins set for Members, report requests and admin control over who is on the team. A third plan, Enterprise, is coming soon: everything in Teams, plus custom solutions, hands-on support and a Locke rollout planned around your organization.

Locke never keeps your prompts or the sensitive content it finds, so there is none of that to transfer. It does keep a local, metadata-only record of detections on your own device (type, count, timestamp and match length), which you can delete at any time.