← Atlas/Sheet 09/AI Agents · Green Software

GreenPipe

First GitLab Duo Agent implementing GSF carbon-aware CI/CD scheduling

Role

Sole engineer · Hackathon project

Timeline

Feb 2026 - March 2026

Year

2026

Category

AI Agents · Green Software

Interactive demo— simulated, runs in your browser

Survey notes

A GitLab Duo Agent that measures the carbon footprint of every CI/CD pipeline using the GSF Software Carbon Intensity (SCI) standard — ISO/IEC 21031:2024. It classifies pipeline urgency from commit messages via a fine-tuned INT8-quantized DistilBERT model, autonomously reschedules deferrable pipelines to low-carbon windows, profiles code for energy inefficiencies using Anthropic Claude, and gamifies sustainability with a contributor carbon leaderboard. GreenPipe applies the same optimisations it recommends: its own SCI score is ~0.00079 gCO₂e per pipeline analysis.

Notable terrain

  • 01SCI carbon measurement per pipeline using ISO/IEC 21031:2024 formula: SCI = ((E × I) + M) / R
  • 02DistilBERT NLP urgency classifier (INT8 quantized, 58% less energy than FP32)
  • 033 auto-deferral modes: recommend-only, approval-required, auto-execute
  • 04Multi-region carbon comparison across 5 cloud regions to find greenest runner
  • 05Anthropic Claude code profiler for energy inefficiency detection
  • 06Contributor carbon leaderboard and automated GitLab MR comment reports

Next sheet

Optimal Congestion Zone Analysis

→