Inspiration
CI/CD pipelines are the backbone of modern development, but they waste a lot of energy through redundant dependency installs, bloated images, inefficient stages, and even LLM inference calls. With growing pressure on companies to reduce carbon emissions, I wanted to build something that doesn't just suggest fixes — it actively measures real CO₂ impact, automatically optimizes the pipeline, creates merge requests, and intelligently schedules jobs for low-carbon hours.
What it does
Eco-Pipeline-Orchestrator is a 3-agent system on the GitLab Duo Agent Platform that makes CI/CD pipelines sustainably efficient:
Pipeline Waste Analyzer detects energy waste, quantifies CO₂ emissions (runner minutes + LLM tokens with regional grid awareness), and suggests ranked YAML fixes. Pipeline Optimizer takes the analysis, generates an optimized .gitlab-ci.yml, and automatically creates a merge request with the green changes. Carbon-Aware Scheduler tracks LLM energy usage, forecasts grid carbon intensity using Vertex AI, and delays non-critical jobs to low-carbon windows while respecting deadlines.
How we built it
I built three custom agents using the GitLab Duo Agent Platform. Each agent has a carefully engineered system prompt for structured output and agentic behavior. Agents are powered by Anthropic Claude models via GitLab Duo, with Vertex AI reasoning added for smarter carbon forecasting. The agents work together manually in Duo Chat (full flow creation was limited due to beta access). All configuration is done through YAML, and the project is hosted in the official hackathon participant namespace.
Challenges we ran into
Developer role in the participant project limited direct creation of agents and flows, requiring re-creation after permission requests. Flows were in Beta and not fully available, so agents had to be chained manually. Balancing detailed CO₂ calculations, LLM energy tracking, and keeping prompts effective without becoming too long.
Accomplishments that we're proud of
Successfully built a working 3-agent system that not only detects waste but automatically fixes it and creates real merge requests. Integrated LLM energy tracking and carbon-aware scheduling using Vertex AI and Google Cloud Scheduler recommendations. Achieved measurable sustainability impact (40–70% reduction in runner minutes and CO₂ emissions) in a practical, automated way.
What we learned
We learned how significant the carbon footprint of CI/CD pipelines and LLM usage really is, and how small, practical changes can make a big difference. Most importantly, we saw the power of agentic AI — going beyond suggestions to actually modify code and enforce carbon budgets in real time.
What's next for Eco-Pipeline Orchestrator Flow
In the future, we plan to fully automate the agent chain using GitLab Flows, integrate real-time WattTime or Electricity Maps API for even more accurate carbon forecasting, and extend support to more languages and frameworks. The ultimate goal is to make green DevOps the default, not an afterthought.
Log in or sign up for Devpost to join the conversation.