A privacy-first operating signal for healthy AI automation.
Measure AI Work is a consent-led Chrome extension, Windows companion, and live dashboard that measures whether AI is integrated into meaningful work at an appropriate level. Its red–yellow–green signal rewards useful automation with sustained work and human review. More AI activity does not increase the score indefinitely.
Status: functional developer release. Real local measurements and the first consented local-agent receipt lane are working. Chrome Web Store distribution, signed packaging, independent security review, and hosted multi-agent gateways remain pre-production work.
On August 3, 2026, we ran an end-to-end local adoption audit on Windows:
- Loaded the Manifest V3 extension unpacked in Chrome.
- Explicitly enabled tracking from the extension popup.
- Installed the optional native-messaging companion for the approved extension ID.
- Performed an AI-to-work cycle across supported host categories.
- Opened the hosted dashboard and observed it transition from
WAITING FOR EXTENSIONwith demo data toLIVE DEVICE CONNECTEDwith the local live score. - Observed subsequent score and traffic-light movement during continued use.
This proves the local extension → category counters → score engine → injected bar → dashboard message path. The developer SDK also proves an explicitly authorized agent can contribute content-free activity receipts. It does not yet prove production-scale accuracy, universal agent coverage, or enterprise-grade tamper resistance.
The extension includes an isolated audit mode for installation verification:
- Reload the unpacked extension after pulling the latest source.
- Open the toolbar traffic-light icon.
- Choose Run two-minute live audit.
- Use a supported AI tool briefly, then switch to GitHub, Office, Figma, Notion, Linear, Slack, or another supported work surface.
- Watch the dashboard report
LIVE AUDIT RUNNING, thenLIVE AUDIT COMPLETE.
Audit data uses the real collector and a compressed consistency target. It is stored separately from the normal daily score and can be exited without contaminating personal history.
The personal edition starts off. It stores daily category totals in chrome.storage.local and sends no raw activity to a Measure AI Work server.
Collected:
- Supported service or application category
- Active seconds
- Idle/active state
- AI-to-work transitions
- Human-review proxy intervals
- Timestamp buckets and local score inputs
Never collected by the personal edition:
- Prompts or AI responses
- Page, post, document, email, or message content
- Form fields or keystrokes
- Clipboard contents
- Screenshots
- Window titles or filenames
- Credentials or authentication tokens
- Contact lists or private identifiers
The Windows companion returns only foreground process name, idle state, and timestamp to the single extension ID authorized in its native-host manifest. It does not inspect the active document.
Active supported tab ─┐
├─> local category counters ─> explainable score ─> traffic light + dashboard
Foreground app ──────┘
Consented local Agent SDK (developer preview) ─> signed minimal receipts ─> local aggregation
Provider webhooks / OpenTelemetry (planned) ─> approved gateway ─> aggregate-only reporting
The extension measures foreground browser and desktop activity. A participating background agent can now report content-free category, duration, and human-review metadata through the local developer SDK in agent-sdk. It cannot truthfully see an unrelated agent that has not integrated and received consent.
The proposed cross-ecosystem design is documented in the Agent Event Contract. Its core principle is broad visibility through broad consented integration—not invisible surveillance.
| Component | Weight | Meaning |
|---|---|---|
| Automation balance | 55% | AI activity relative to measured work, with an upper-use penalty |
| Workflow consistency | 20% | Sustained meaningful work rather than isolated prompting |
| Human review | 25% | Return from AI activity into a supported work/review surface |
- Red: low responsible adoption or insufficient evidence.
- Yellow: developing integration.
- Green: healthy automation balance with meaningful work and review.
This is an operating indicator, not an employee-performance grade or a scientific diagnosis. Calibration, bias testing, and organization-specific validation are required before employment or governance decisions.
- Open
chrome://extensionsand enable Developer mode. - Choose Load unpacked and select this repository's
extensiondirectory. - Open the traffic-light toolbar icon and choose Turn on tracking.
- Optional: copy the extension ID and run the Windows companion installer:
cd "C:\path\to\measure-ai-work\companion"
.\install-host.ps1 -ExtensionId YOUR_32_CHARACTER_EXTENSION_ID- Reload the extension and open the dashboard.
No administrator permission is required. uninstall-host.ps1 removes the native-host registration.
Before production distribution, we require:
- Property and adversarial tests for scoring and event integrity
- Signed Windows packaging and reproducible builds
- Chrome Web Store permission review
- Consent grant, revocation, retention, and deletion tests
- Cross-tenant replay and correlation defenses
- Independent privacy and security review
- Published calibration methodology and limitations
Requires Node.js 22.13 or later.
pnpm install
pnpm exec vinext dev
pnpm exec vinext buildRepository areas:
extension/— Manifest V3 collector, score engine, toolbar icon, popup, and injected barcompanion/— optional Windows native-messaging hostapp/— product, installation, privacy, and live dashboard surfacesdocs/— evidence and integration contractsdns/— custom-domain records
Copyright © 2026 ThePolka.Cloud contributors.
AI Productivity Ledger™, Measure AI Work™, and their branding are trademarks of ThePolka.Cloud.
The source code is licensed under the Apache License 2.0. See NOTICE for attribution and trademark information.