Agentic CICD

Landing page
Chat Page
Repo and pipeline created by the agent

Inspiration

Over the past year, I’ve been deeply immersed in the world of AI agents — building, experimenting, and learning how they can assist with real-world tasks. After graduating this May and starting my journey as a DevOps intern, I found myself constantly thinking: “What if AI agents could handle most of the DevOps workload?”
Driven by that curiosity and passion, I decided to build a platform where DevOps automation is not just scripted — it's intelligent. And that's how Agentic CICD was born: a step toward fully autonomous DevOps pipelines powered by agentic systems.

What it does

Agentic CICD is a modular, multi-agent platform designed to automate and manage CI/CD workflows using AI agents. Each stage of the DevOps lifecycle — from code validation, testing, deployment, to monitoring — is handled by a specialized agent. These agents can:

Analyze pull requests and suggest improvements
Monitor pipelines and fix common CI/CD issues
Integrate with tools like GitLab, Jenkins, and Prometheus
Perform reasoning over system states and logs
Adapt workflows dynamically based on project context
All with the help of GitLab MCP Server

How we built it

Tech Stack:
- 🧠 LLM Backbone: Gemini + LangGraph for agent orchestration
- 🛠️ Tooling: GitLab MCP Server, Jenkins, Prometheus, Ansible
- 🧱 Architecture: Multi-Agent System (MCP) with each tool exposed as a separate GitLab MCP server
- 🔌 Integration: LangChainJS + Tavily for enhanced search and context-aware decisions

Challenges we ran into

Designing a scalable agent orchestration system without losing performance
Maintaining context across asynchronous events in CI/CD pipelines
API rate limiting and authentication with GitLab and Jenkins
Streaming SSE-based communication between the React agent and MCP servers
Debugging tool interop across independently running agents

Accomplishments that we're proud of

Built a fully functional agentic CI/CD platform in a hackathon timeline
Successfully integrated multiple DevOps tools under a single agentic interface
Created a modular system where agents can be plugged/unplugged without rewriting orchestration logic
Developed a live demo showcasing autonomous pipeline diagnosis and resolution