Skip to main content
    Locke, by Sonomos · Coming Soon

    Privacy everywhere you use AI.

    Locke masks sensitive data before it leaves your machine — in every AI app, 100% locally. You're looking at the actual app below. Click around.

    Request early accessWindowsLinuxmacOS planned

    INTERACTIVE DEMO

    Seeded with fictional sample data — nothing you click leaves this page.

    LOCKEby Sonomos
    View as
    Live preview · Sample data
    Pause events are audit-logged for compliance.

    Dashboard

    At-a-glance protection status

    Welcome to Locke

    A 3-step tour of what makes Locke different — click through at your own pace.
    0 of 3 done
    1Watch a prompt get maskedRun a sample prompt in the Sandbox and see Detect → Mask happen live.
    2Choose what to protectOpen Controls: surfaces, 70+ data types, and your own custom rules.
    3See the proofReports turn protection into documents you can hand to an auditor.
    Dismissing hides this for future visits — saved in this browser only.
    Protected
    Local-only detection active · 5 surfaces seen this session · nothing leaves this device

    Compliance check

    Low risk
    No unmasked sensitive sends in the last 24 hours
    37
    Sensitive items caught
    Every appearance of sensitive information Locke has found in what you count.
    37
    Sensitive items hidden
    Of the items found, how many were masked before they could reach an AI.
    30
    Apps & sites protected
    AI apps and websites Locke watches for sensitive data before it's sent.
    Counting You typed · occurrences —
    37 sensitive occurrences masked before reaching an AI model — roughly 3 per session.

    Live monitoring feed

    6 detections this session.

    Sensitive Data Origins

    You typed
    Info you actually wrote into your message
    in your tally25% · 37
    Injected context
    Things your AI tool quietly attaches to every message — saved notes, memory, project files
    not counted30% · 45
    System prompt
    The AI tool's own built-in hidden instructions
    not counted19% · 28
    Tool output
    Results from files or commands the AI read, fed back in
    not counted15% · 22
    The AI's replies
    Text the AI generated
    not counted10% · 15
    Other
    Everything else
    not counted2% · 3
    Much of what gets sent to an AI isn't typed by you — it's added by the app, its tools, or the AI itself. Locke protects all of it the same way; you choose what's included in your tally under Settings → What counts as a detection.

    DASHBOARD

    Your protection, quantified.

    The dashboard answers one question the moment the app opens: am I protected, and what has Locke caught?

    • Protected / Paused status with a live count of the AI surfaces seen this session
    • Quick metrics — sensitive items caught, items hidden before they reached an AI, and apps & sites protected
    • A live monitoring feed of every catch, processed locally and never leaving your device
    • A plain-language compliance check: your current risk, and why

    CONTROLS

    You decide what counts as sensitive.

    Controls is where Locke stops being generic: pick the surfaces it watches, the data types that matter, and add your own.

    • Protected surfaces: Locke auto-discovers AI websites and apps as you use them — protect a new one with one click
    • Detection data types: 70+ built-in types across identity, financial, health, work data, and more — toggle what counts
    • Custom detection types: teach Locke your matter numbers, patient IDs, or project codenames — they mask like any built-in type
    • A scanning floor that sets how cautious detection should be, machine-wide

    REPORTS

    Proof, not promises.

    Reports turn day-to-day protection into documents you can hand to an auditor, a client, or your own IT team.

    • Compliance reports mapped to a privacy regime — Comprehensive privacy, HIPAA, EU/UK GDPR, or CCPA/CPRA
    • Technical reports for IT and security: key metrics, protection rate, data types, AI destinations, daily activity
    • A complete detection & masking history, plus a one-click combined export to PDF
    • Generated entirely on your device — reports cover protection, never prompt content

    SANDBOX

    Rehearse the send before you send.

    The sandbox is a safe test area: stage a prompt, watch detection light up every sensitive span, and see exactly what an AI would receive instead.

    • Watch Detect → Mask happen span by span — names, SSNs, keys, and your custom types
    • See the masked text the AI would get; Locke 1.0 uses one-way placeholders, with reversible tokenization planned shortly after 1.0
    • Flip custom rules on and off to test them before you rely on them
    • Nothing tested in the sandbox is ever sent anywhere

    TEAMS

    Whole-team protection. Zero-visibility admins.

    On the Teams plan, admins set the rules and see coverage — never content. Members get protection that's already configured.

    As the admin

    You manage the rules, never the data.

    • A dedicated My Team tab: seats, invites, and roles in one place
    • Shared policies synced to every seat — required surfaces, data types, and a scanning floor every member inherits
    • Team reports show protection coverage per member — including when protection was paused, and for how long

    As a member

    Protected out of the box.

    • Protected from the first login — the team policy arrives pre-configured
    • Org-managed rules are visibly locked; personal custom types stay yours
    • The same 100% local guarantee: admins manage rules, they never see data

    Accountability is built in: pause events are audit-logged, and both compliance and technical reports include when each member turned protection off and for how long — so "Locke was off" is never a mystery in an audit. Prompt content is never collected. Locke reports on protection, not on people.

    ONE APP, THREE VANTAGE POINTS

    Individual or Teams — the demo shows both.

    Individual

    One person, every AI app. Your rules, your local dashboard.

    Teams — Admin

    Manage the rules, never the data. Seats, policies, coverage.

    Teams — Member

    Protected out of the box, with admin-managed policies.

    Pricing is being finalized — see the plans or request early access to lock in launch pricing.

    WHY DESKTOP

    Your browser isn't the only place you use AI.

    Locke runs as a native app because that's the only place system-wide protection is possible. Under the hood, specialized agents detect, mask, and verify in real time — adapting to context instead of rigid rules.

    System-wide coverage

    Every AI app on your machine — ChatGPT Desktop, Claude Desktop, Ollama, your IDE — not just what happens in a browser tab.

    100% local, even offline

    All detection and masking happens on your device. Nothing is sent to us or any cloud — and it all works with zero connectivity.

    CLI, IDE & local LLMs

    Shields AI assistants in Cursor, Claude Code, and Copilot, terminal agents, and on-device models like Ollama and LM Studio.

    Injection & Unicode defense

    Blocks prompt-injection attacks and hidden Unicode tricks before they reach the model.

    File & clipboard scanning

    Opt-in inspection of documents and pasted content before they're sent to any AI tool.

    Native on Windows & Linux

    Desktop apps with hardware-level access for faster, more accurate detection — macOS to follow.

    QUESTIONS

    Locke FAQ

    Is the demo on this page the real app?

    It's a click-through replica of the Locke desktop app, built from the app's real interface and seeded entirely with fictional sample data — every name, team, and number is invented. You can switch between the Individual and Teams views, open every page, and run the sandbox; nothing you click is typed, stored, or sent anywhere. It shows the UI and workflow, not the product's internals.

    What can a Teams admin see?

    Admins manage seats and shared policy rules, and team reports show protection coverage — including when a member paused protection and for how long, since pause events are audit-logged. What members type, and what Locke detects for them, never leaves their own device. Locke reports on protection, not on people.

    When will Locke be available?

    We haven't announced a release date yet. Reach out through the contact page to request early access and we'll be in touch as soon as builds are available.

    What does Locke protect?

    Locke, the Sonomos desktop app, extends protection system-wide — covering native apps like ChatGPT Desktop and Claude Desktop, on-device models like Ollama, and AI assistants embedded in your IDE or terminal. Every AI tool on your machine, not just your browser.

    Is anything sent to the cloud?

    No. All detection and masking happens entirely on your device. Nothing is ever sent to us or to any third party.

    What pricing will Locke have?

    Pricing is still being finalized. Request early access and you'll lock in launch pricing — and we'll give you a heads-up the moment details are published.

    HOW IT COMPARES

    How Locke compares

    Evaluating Locke against your current approach? See how a local, zero-knowledge privacy layer stacks up next to the alternatives.

    vs Microsoft Purviewvs CASB / SSEvs self-hosted AIAll comparisons

    Want in early?

    We're talking to early users now. Request early access to Locke and lock in launch pricing before we publish it.

    Request early access