SECURE ENTERPRISE BROWSING
Management console
Policies, people and protection in one workspace.
Administrator access
Sign in with your enterprise identity, or use the bootstrap token configured on the server.
Applies immediately to the management-service export. Security audit collection remains active. Delivery resumes from its durable cursor when enabled.
POLICY POSTURE
Current enforcement
Based on the published policy. Configuration is not a device health attestation.
OPERATOR WORKFLOW
Configure, review, publish
- Choose a template or adjust your controls.
- Test data protection with representative examples.
- Review changes and publish a signed policy.
- Check device versions and security events.
ENTERPRISE SETUP
Start with a reviewed policy template
Templates tighten selected controls while keeping your existing rules and infrastructure. Preview changes, add them to your draft, then publish when ready.
Integration configuration
These checks describe policy configuration. They do not certify that an enrolled device can reach the service.
Your existing network and security services
Keep Windows-managed proxy and DNS for existing enterprise deployments, or mandate corporate resolvers and VPN interfaces. ICAP connects compatible antivirus and sandbox services before content is released.
TLS certificate validation stays enabled. Customer CA certificates and mutual TLS credentials belong in protected deployment configuration. Plaintext ICAP exposes submitted content; approve it only for a separately protected trusted transport. VPN adapter checks do not prove tunnel routing. Deferred verdicts and modified payloads are rejected. Publish the signed policy to apply changes.
CENTRAL CONFIGURATION
One policy. Every managed browser.
Devices refresh every 15 seconds. Changes and actor identity are recorded in the audit log.
IDENTITY AND FEATURE ACCESS
User classes and website access
Create profiles for any user class with your own names and controls. Employee, developer and high-protection templates are optional starting points. Assign profiles to verified enterprise subjects or explicit managed devices. Asset-owner labels are not authenticated identities. Website denials and category restrictions accumulate across matching profiles; an allowlist does not bypass a denial.
| Profile | Assignments | Developer tools | Action |
|---|
Create profile
Developer grants require central delegation; any applicable deny wins. No profile can override mandatory scanning or unlock centrally hidden settings.
Preview effective device policy
Central categories and developer delegation
Categories contain your approved domain lists. This is not an automatic external website reputation feed. Enabling delegation permits reviewed developer-profile grants; denying profiles still win.
Website categories
Use exact hosts or wildcard subdomains such as *.social.example. Changes stay in this local draft until explicit publication or independent approval.
Assignment visibility
| Device | Verified subject | Effective profiles |
|---|
CONTROLLED DEPLOYMENT
Groups and pilot rollouts
Apply additional restrictions to explicit device groups. Start with selected pilot devices, inspect their reported policy versions, then explicitly expand the cohort. Group rules can tighten the baseline; they cannot weaken it. No automatic promotion occurs.
| Group | Members | Stage | Actions |
|---|
Create a device group
Membership, restrictions and cohort changes are signed policy changes. Mandatory independent approval applies to these changes too.
Additional company bookmarks
These managed bookmarks are added for the selected rollout cohort. They do not replace baseline company bookmarks.
| Title | URL | Folder | Action |
|---|
Effective policy preview
Reported rollout state
| Device | Effective groups | Reported / target version | Version acknowledgment | Last contact |
|---|
Acknowledgment compares the device-reported version with its target. It is not security attestation or proof that every control is functioning. Last contact is shown separately.
INDEPENDENT CHANGE CONTROL
Policy approval queue
Loading approval requirements.
Prepared changes remain inactive until approved. A different administrator must review the exact payload and base version. Stale changes must be prepared again against the latest policy.
| Prepared | Author | Change | Base version | Status | Review |
|---|
CHANGE CONTROL
Policy history and recovery
Review persisted signed versions and compare changes. Restoring reapplies the selected security rules, permissions, bookmarks and network settings, group memberships and rollout cohorts. The original version stays preserved; restoration publishes a new version with a fresh expiry. Devices never receive a lower policy version.
| Version | Published | Actor | Review |
|---|
Managed browser settings
Hide settings or deny changes centrally. These permissions are enforced in the native browser process as well as the interface. Hiding a section denies its user actions; publishing a new signed policy applies changes to connected devices at refresh.
| Section | Hide and deny access | Lock user actions |
|---|---|---|
| Appearance | ||
| Bookmarks and tab groups | ||
| AI workspace | ||
| Password manager | ||
| Manual document scan | ||
| Local agent controls |
Appearance values become mandatory when appearance is hidden, locked, or settings access is disabled. Otherwise users may choose local preferences. Vault locking remains available to keep credentials safe. Mandatory security inspection continues independently of settings visibility. Publish signed policy above to apply.
COMPANY LIBRARY
Managed bookmarks
Publish intranet and approved service links to every browser. Company bookmarks are read-only for employees. Personal bookmarks remain separate; configure whether employees may add, edit or delete them.
| Title | HTTPS / HTTP URL (no query or fragment) | Folder | Actions |
|---|
Company changes are drafts until you publish the signed policy. Personal permissions also respect the centrally locked or hidden Library section.
Device personal bookmarks
Authorized administrators can review and remotely edit a device's synchronized personal bookmarks. Changes are applied when the browser next refreshes management state. Titles and URLs are never included in security audit events.
| Title | HTTPS / HTTP URL (no query or fragment) | Folder | Actions |
|---|
Saving compares the loaded revision so another user's or administrator's updates cannot be overwritten silently.
TEACH YOUR BROWSER
Define what must stay inside
Add phrases that must never leave, or register an exact file fingerprint. Reference files stay on your computer; only their SHA-256 fingerprint enters policy.
Local semantic reference protection
Describe a confidential category. Your approved local EmbeddingGemma 2 runtime creates pinned text reference vectors. Document extraction and local OCR text can be compared against these references. This workflow does not train a model or provide native video inspection.
Only normalized reference vectors and coverage features enter the policy draft. No document, extracted text or model fine tuning is retained. Complete extraction and chunk coverage are required; publish explicitly to approve.
Requires locally provisioned model weights, an exact model revision, matching dimensions and the document-v1 profile. Similarity is a detector score, not a confidence guarantee. Incomplete extraction or chunk coverage always blocks.
Enterprise security and local agents
Configure centrally signed DNS blocklists, YARA source rules, content phrases, restricted Sigma metadata detections and agent permissions. Unsupported Sigma syntax is rejected.
Disabled by default. Development support is limited to immutable agent-owned nonpersistent browser partitions bound to device, principal and credential generation. It additionally requires network:background in both baseline and individual grant, allowed destinations and current policy. Profiles may deny it and cannot expand permission. This does not enable human/default/shared/legacy workers or unattributed requests. Owner-registry migration and native verification are required; never infer ownership from the active tab.
Visual export is a separate opt-in capability for the active managed webpage viewport only. It does not authorize desktop, other windows or browser-interface capture. Required content inspection must succeed before pixels are returned. Grant it in both the central baseline and the individual identity; remove the scope or disable agents and publish to revoke access. Sensitive-read permission does not bypass visual inspection. Availability is limited to the implemented and tested export path.
Saving, listing and restoring sessions require separate capabilities. Saved sessions contain approved URLs and tab group metadata only: no passwords, form values or page state. Restoration rechecks current policy, destination permissions and DLP; it never restores revoked agent authority.
Vault status, masked listing and autofill are separate capabilities. A person must unlock the local vault; agents cannot unlock it or export passwords. Listing and filling are restricted to the current page and the credential's exact origin. Autofill also requires the central vault and autofill permissions. Grant only the capabilities an agent needs.
Individual agent identities
Individual mode requires an enrolled local identity matching an enabled, unexpired grant. Grant scopes and destinations further restrict the central baseline. Credentials stay on the device; no private keys or tokens are stored here. Revocation and generation changes take effect after the new signed policy reaches the browser.
Changes remain in this draft. Publish with the usual policy review. Disabling a grant revokes it; advancing its generation invalidates previously paired credentials and requires local pairing again. Empty capabilities permit no actions.
Advanced source configuration: YARA, Sigma and visual references
Structured controls above are authoritative for DNS, content and agent permissions. Preserve advanced rule sources when making other changes.
Agents require paired keys and authenticated post-quantum sessions; capabilities are further restricted by each token, destination rules and DLP. Disable paired local agents and publish to revoke control. Agent permissions never grant administrator privileges or unlock centrally hidden settings.
Publish signed policy above to apply. Downloads require configured local YARA or external ICAP. Every enabled scanner is mandatory; unavailable inspection blocks downloads. Scan limit is 20 MB. Sigma supports exact metadata selection only.
Governed document teaching
Submit a reference document to the private local inspector, then explicitly approve its fingerprint for publication. For incomplete extraction, explicitly approve available fingerprints with partial coverage; unextracted media is not fingerprinted. Only hashes, coverage counts and the reference label are retained. Full documents, extracted text and base64 are discarded after local processing and are never persisted by management.
Local password vault
Credentials stay on the browser device. Management stores permissions and lifecycle audit events only; passwords, usernames and credential identifiers are not uploaded.
Publish signed policy to apply these controls.
Visual impersonation protection
For example, register a trusted Google login screenshot and allow only https://accounts.google.com. A matching login on another origin can then be blocked even when its domain is absent from every blocklist.
Upload your own trusted PNG of a brand login page or cropped login form. Only a derived perceptual hash, dimensions and allowed origins enter the policy draft. Reference pixels are discarded; no automatic Google or other brand profile is supplied.
Add trusted references, enable the control above and publish the signed policy. Use multiple references for different themes, languages and login steps. False positives, viewport changes and visual evasion require validation. The browser checks rendered pixels; this is not pre-render response inspection.
Managed clipboard and ClickFix protection
Control supported browser copy, cut and paste actions centrally. High-risk shell command patterns are blocked without sending clipboard text to an AI provider. Clipboard text and credentials never enter security logs.
Publish signed policy to apply. Mandatory DLP remains active when supplemental clipboard controls are disabled. Unsupported formats and unavailable required inspection fail closed. Pattern detection covers selected execution chains; it does not prove text or commands safe.
DEVICE FLEET
Enrollment and access
Enable or disable supplemental manual document analysis per browser. Mandatory upload DLP and download YARA continue regardless of this setting.
| Device | Platform | Identity | Policy | Last seen | Access |
|---|
ADMINISTRATOR ACCESS
Active management sessions
Administrators can review current SSO sessions and immediately revoke a session. Identity-provider group membership controls whether that person can sign in again. Bootstrap access has no stored browser session.
Session inventory requires the administrator role.
| Identity | Role | Expires | Action |
|---|
SECURITY OPERATIONS
Incident triage
Review security event metadata, assign an investigation and record its outcome. Protected evidence is shown separately only with explicit, audited access. Keep investigation notes free of secrets and personal data.
| Created | Security event | Status | Severity | Assignee | Review |
|---|
Incident
Device-reported protected evidence
Investigation history
AUDIT LIFECYCLE
Database event retention
Loading deployment configuration.
Retention applies only to exported database events that are not referenced by incidents. Investigation history is retained. External SIEM, file exports and backups have separate retention policies. Deployment configuration controls this setting.
SECURITY MONITORING
Recent audit events
| Time | Event | Device | Metadata |
|---|
This view shows the latest 100 events. Prompts and document contents are excluded. Exports contain up to 1,000 events; cursor-based APIs and durable HTTPS delivery support continuous SIEM ingestion.
COMPANY EXPERIENCE
Company logo and start page
Roll out your company identity through signed policy. Name and accent above apply alongside this logo.
Logo pixels are re-encoded locally; SVG is not accepted. News is retrieved by management only from deployment-approved feed hosts. Ask your server operator to allow the exact host. Feed URLs cannot contain credentials or fragments. Use server-configured authentication; do not put secrets in URLs. Publish and review these changes with your normal policy workflow.
Download inspection
Document DLP is required by default. Disabling it permits incoming documents after every enabled malware scanner approves them; upload protection remains enforced. Without any antivirus scanner, complete document DLP is still mandatory. Office macros are not executed. Unreadable or unsupported macro extraction blocks when local macro scanning is enabled. ICAP remains mandatory whenever enabled.
AUDIT INTEGRITY
Verify security audit records
Run an explicit consistency and signature check. This check is separate from OpenTelemetry traces and SIEM delivery.
Not verified in this session.
PROTECTED SECURITY EVIDENCE
Protected evidence collection
Opt in through centrally signed policy. All collection controls are off by default. Ordinary security audit records and OpenTelemetry traces remain separate.
Download evidence contains the source URL without its query or fragment, filename, SHA-256, byte count, MIME type and verdict. It does not contain the file body or the OS save path. Passwords and recognizable authentication credentials are excluded. A deployment-protected evidence key is required before enabling recording.
Disabling recording prevents new device evidence writes and removes pending native evidence payloads; ordinary security events remain. Existing encrypted evidence retains its access and retention controls. Interaction capture records the website origin, session, action and bounded target metadata. Guarded clipboard and agent effects are marked completed; trusted foreground page clicks and activated form events are marked observed, which does not prove network transmission. Sensitive forms, page paths, values, queries and fragments are excluded. It excludes clipboard contents, entered values and control labels. Collection is bounded and limited to the active main-frame page; it is not a complete replay. Publish and review this draft through the normal policy workflow.
OBSERVABILITY
Browser OpenTelemetry
Publish centrally signed operational or detailed debug collection. Spans report code stages, timing and outcomes. Confidential security evidence needs its separate protected storage; passwords and authentication tokens are excluded.
HTTPS is required except for a numeric loopback collector. Queue and export limits remain enforced. Browser logs and traces use OTLP. Durable security events are separately delivered by the management service in ECS or CEF; its HTTPS destination, TLS certificates and credentials are configured by the server operator, not exposed in browser policy. Backend service traces use the server OpenTelemetry configuration. Endpoint credentials do not belong in policy URLs. Disable and publish to stop new collection.