Skip to main content
    SonomosSupport Center
    Main Site

    The app, screen by screen

    Controls

    Everything that decides what Locke protects lives here, on three tabs: where it watches, what it looks for, and how cautious it is about deciding.

    Available toAll plans

    The three Controls tabs
    TabAnswersChange it when
    Protected surfacesWhich AI apps and websites does Locke watch?You start using a new AI tool, or want to stop covering one.
    Detection data typesWhat counts as sensitive?Locke is missing something specific to your work, or flagging something you don't care about.
    Scanning floorHow sure must Locke be before it masks?Too much is slipping through, or too much is being masked unnecessarily.

    Protected surfaces

    A surface is one place your data can go — a website you use in a browser, or a desktop app. This list manages AI-service screening. When desktop capture is active, discovery can also decrypt and inspect initial requests to eligible unknown HTTPS destinations that reach the local proxy, including services later classified as non-AI. AI-service controls apply after classification; changing them does not by itself disable initial discovery inspection. See the Locke Privacy Policy for discovery limits and bypasses. Check the list matches what you actually use.

    Websites on the left, desktop apps on the right, and anything newly spotted at the top.

    1. Detected, not yet managed. AI tools Locke has noticed you using but isn't covering yet. + Protect moves one into the right list, switched on.
    2. AI websites — sites in a browser, like chatgpt.com.
    3. AI apps — desktop applications and the API endpoints they talk to, like api.openai.com.
    4. Each row has a toggle (watch it or don't) and Remove (drop it from the list entirely).

    The four things you can do here

    Actions on the Protected surfaces tab
    ActionEffect
    Toggle a surface offTurns screening off for that surface but keeps it in the list, so you can switch it back on later. Initial desktop discovery inspection is separate, as described above.
    Remove a surfaceDrops it from the list. If you use the tool again, Locke will re-discover it and offer it back.
    Add a surface by handType a domain or endpoint into the box at the bottom of either column and press Add. For internal AI tools and self-hosted models that auto-discovery won't find.
    Protect a discovered surfaceOne click on + Protect. The fastest way to keep coverage current.
    Walkthrough

    Protecting a new AI tool you've started using

    Step 1 of 4

    1You've been using Perplexity for a week. Locke noticed, and it's sitting at the top of the Controls page under 'Detected, not yet managed' — seen 3 days ago, not covered yet.

    Read all 4 steps as text
    1. You've been using Perplexity for a week. Locke noticed, and it's sitting at the top of the Controls page under 'Detected, not yet managed' — seen 3 days ago, not covered yet.
    2. Click + Protect on the perplexity.ai row.
    3. It moves into AI websites, switched on. The discovered card disappears because there's nothing left unmanaged. Locke starts checking Perplexity immediately — no restart, no confirmation step.
    4. For an internal tool Locke can't discover on its own, type its domain into the box at the bottom of the column and press Add. Same result.

    Detection data types

    The catalog of what Locke recognizes, grouped into categories — personal details, government IDs, financial data, health data, credentials, crypto, legal identifiers, device identifiers. Each type has a toggle and a risk level.

    Categories of sensitive data, each type individually switchable.

    1. The filter box narrows the whole catalog as you type. With 78 types, it's faster than scrolling.
    2. Risk pills — High, Mid, Low — are Locke's assessment of how damaging that kind of value is if it leaks.
    3. Region-specific types are off by default. If you work with UK, Canadian, or Indian identifiers, switch them on.
    4. Custom rules at the bottom — for the identifiers no catalog could know about.

    Most types are on out of the box. The ones that aren't are regional — an NHS number, a UK National Insurance number, a Canadian SIN — because switching on every country's formats at once produces false alarms for people who never encounter them. The full catalog is listed in What Locke detects.

    Custom rules

    No catalog can know your matter codes, your internal project codenames, or the client shorthand your team uses. Custom rules are how you teach Locke those.

    Parts of a custom rule
    FieldWhat it holds
    RuleA name you'll recognize later, like 'Matter codes'.
    MatchWhat to look for — either a pattern like MTR-\d{5}, or a plain list of terms.
    Actionmask replaces the value like any other detection; flag records it in your history without changing what gets sent.

    Use + Add rule to create one, Edit to change it, and Reset to default to discard your rules and return to the shipped set. Before trusting a new rule, test it in the Sandbox — that's exactly what the Sandbox exists for.

    Scanning floor

    Two settings: how much doubt Locke tolerates, and which kinds of input it checks.

    One dial for caution, four toggles for what gets checked.

    1. The sensitivity dial. Balanced is the default and the right answer for most people.
    2. The explanation under the dial changes with your choice — read it before you move it.
    3. Where Locke looks. Text is always on. Files and voice are opt-in. Images are not screened in this release.
    The three scanning floor levels
    LevelBehaviorGood forTrade-off
    RelaxedMasks only clear, high-confidence matches.Low-noise workflows where an occasional miss costs less than an interruption.Fewest false alarms. Borderline cases get through.
    BalancedMasks clear and likely matches. The default.Almost everyone.Occasional false alarm, very few misses.
    StrictMasks anything that looks even possibly sensitive.Regulated work where a miss costs more than a false alarm.Catches the most. Expect more false alarms.

    Where Locke looks

    Input types Locke can check
    InputCoversDefault
    TextTyped and pasted text, form fields.On
    ImagesScreenshots and uploaded photos. Not screened in this release (no text reading, no NSFW filtering): sent as they are unless blocked. We plan to add this in a future release.Sent unscreened
    FilesPDF, DOCX, CSV and more.Off
    VoiceDictation and voice input.Off

    Changes on this tab need Save changes at the bottom. The other two tabs apply immediately.

    If some controls are locked

    Available toTeams Member

    On a Teams plan, your admin sets a floor: required surfaces, required data types, and a minimum sensitivity. Those controls appear here but can't be turned down — you can always protect more, never less; see the rule that explains Teams for why. A banner at the top of the page points you at My Team › Policies for the team-wide rules, and For team members covers what's yours versus theirs.