GreenLens AI
Measure the full environmental cost of your enterprise AI.
GreenLens AI gives leadership teams visibility into what their organization's AI deployment actually costs: in carbon, in data center cooling water, and in unused licences. It connects to the admin-level usage data your AI providers already expose, runs it through a multi-agent analysis pipeline, and turns the results into an executive dashboard with concrete, board-ready recommendations and an exportable ESG report.
Live app: https://greenlens-ai-pi.vercel.app
GreenLens measures what the organization deployed, not what individuals said. It only reads aggregate admin and usage reports; no prompts, messages, or per-user content are collected.
Features
- Integrations — connect OpenAI (organization usage API), Microsoft 365 (Graph reports for Copilot) and Google Workspace (Gemini licence data) in a few minutes, with no code changes.
- Carbon & water footprint — estimates AI-related emissions and cooling-water use from token volume, model class, and per-model energy factors.
- Model efficiency — classifies workloads and flags tasks running on heavier models than they need.
- Licence intelligence — utilization, unused seats, and renewal risk across enterprise AI licences.
- Usage analytics — anomaly detection (z-score), trend regression with 30-day projections, and hourly usage patterns.
- Strategic recommendations — decisions with effort estimates, projected savings, and the regulatory and financial incentives they support (CSRD, tax incentives, grants).
- Benchmarks & trends — month-over-month scores so improvement is visible over time.
- ESG export — download a PDF sustainability report for stakeholders.
How it works
- A company signs up (GitHub, Google, or Microsoft sign-in via Supabase Auth) and completes onboarding.
- It connects one or more providers through OAuth or an admin API key.
- Triggering an analysis starts a resumable, step-by-step pipeline of agents:
- Usage Analyst — normalizes provider usage and runs the statistical analysis.
- Carbon & Water Accountant — converts usage into emissions and water figures.
- License Intelligence — evaluates seat utilization and renewal exposure.
- Strategic Translator — turns findings into executive recommendations (Backboard with a persistent thread, falling back to Gemini, then to a deterministic generator).
- Synthesis — assembles the final report.
- Results are stored in Supabase and rendered across the dashboard: overview, footprint, models, licences, decisions, incentives, benchmark, and ESG pages.
Tech stack
| Layer | Technology |
|---|
| Framework | Next.js 16 (App Router), React 19, TypeScript |
| Styling & charts | Tailwind CSS 4, Recharts, Lucide icons |
| Auth & database | Supabase (Postgres, Auth, Row Level Security) |
| AI | Backboard (agent threads), Google Gemini |
| Data sources | OpenAI Usage API, Microsoft Graph, Google Workspace Admin SDK |
| Reporting | @react-pdf/renderer |
| Validation & tests | Zod, Node test runner via tsx |
| Hosting | Vercel |
Getting started
Prerequisites
- Node.js 20+ and npm
- A Supabase project
- API credentials for the providers you want to connect (see below)
Setup
git clone https://github.com/zhao0524/greenlens-ai.git
cd greenlens-ai
npm install
cp .env.example .env.local # fill in the values below
npm run dev # http://localhost:3000
Database
In the Supabase SQL editor, run DATABASE_FULL_SETUP.sql. It creates all tables, seeds the model energy and incentives libraries, and applies RLS policies, and is safe to re-run. Incremental changes live in supabase-migrations/. reset-all-tenant-data.sql and reset-test-users.sql wipe data for a clean demo state.
Environment variables
| Variable | Purpose |
|---|
NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_ANON_KEY | Supabase project (browser-safe) |
SUPABASE_SERVICE_ROLE_KEY | Server-side Supabase access for the analysis pipeline (server only) |
BACKBOARD_API_KEY | Backboard agent threads for the Strategic Translator |
GEMINI_API_KEY | Gemini fallback for recommendations and the ESG report narrative |
GITHUB_CLIENT_ID, GITHUB_CLIENT_SECRET | GitHub sign-in |
MICROSOFT_CLIENT_ID, MICROSOFT_CLIENT_SECRET, MICROSOFT_TENANT_ID | Microsoft 365 / Graph integration |
GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET | Google Workspace integration |
NEXT_PUBLIC_SITE_URL | Base URL used for OAuth callbacks |
ENABLE_GOOGLE_USAGE_SPIKE | Optional Google usage-spike analysis (false by default) |
Scripts
| Command | Purpose |
|---|
npm run dev | Start the development server |
npm run build | Production build |
npm start | Serve the production build |
npm run lint | ESLint |
npm test | Unit tests (tests/*.test.ts) |
Project structure
src/
app/ Landing page, auth, onboarding, dashboard pages, API routes
api/integrations/ Microsoft/Google OAuth and OpenAI key connection
api/pipeline/ Trigger and poll analysis jobs
api/reports/ Report data and PDF export
components/ Dashboard, onboarding, and landing UI
lib/
agents/ Orchestrator and the five pipeline agents
analysis/ Statistical analysis, job state, polling
calculations/ Carbon and water models
integrations/ Provider clients and token handling
reports/ Report loading and availability
supabase/ Browser, server, and admin clients
analysis/pipeline.py Reference Python implementation of the statistical analysis
tests/ Unit tests
supabase-migrations/ Incremental SQL migrations
The statistical analysis runs in TypeScript (src/lib/analysis/run-stat-analysis.ts) for Vercel compatibility. src/analysis/pipeline.py and requirements.txt are the original Python reference implementation.
Figures are estimates derived from provider usage data and published energy factors. They're meant to guide decisions, not replace audited emissions reporting.
