CI Healer
CI Healer eliminates one of the most soul-crushing parts of software development: sitting in front of a failing build log, decoding cryptic errors, and hunting down the exact line that broke the pipeline. It's an AI-powered DevOps agent that watches your GitHub repositories, detects CI failures the moment they happen, clones the repository, runs static analysis, executes failing tests, and uses LangGraph-orchestrated LLM agents to generate a precise, minimal patch. It then opens an automated pull request with the fix — creating a truly self-healing CI/CD loop.
The Challenge
CI/CD failures are expensive interruptions. A broken pipeline blocks your entire team — deployments halt, context switching spikes, and engineers burn hours decoding cryptic build logs that could be fixed in minutes with the right context. In fast-moving teams, this isn't an edge case; it's a daily tax on engineering velocity.
The Solution
CI Healer monitors your repositories continuously. The moment a pipeline fails, it triggers a multi-agent workflow: one agent reads and classifies the error, another runs the failing tests locally to reproduce the issue, and a third generates a targeted code patch using an LLM with full codebase context. The patch is committed to a new branch and a pull request is opened automatically — often before the engineer who broke the build has even refreshed their GitHub tab.
Screenshots
Key Features
Failure Classification & Root Cause Analysis
Parses build logs, identifies the failure category (dependency conflict, test failure, configuration error, type mismatch), and routes to the appropriate repair strategy — not a generic prompt.
LangGraph Multi-Agent Orchestration
A directed agent graph where each node has a single responsibility: clone, analyse, patch, or submit. Conditional edges enable automatic retry if the first patch attempt doesn't resolve the issue.
Automated GitHub PR Workflow
Creates a descriptively named branch, commits the patch with a structured commit message, and opens a pull request with a detailed description explaining what failed and what was changed.
Configurable Repair Iterations
For complex failures, the agent loops up to a configurable maximum, running tests after each patch and refining the fix until the pipeline goes green or the iteration limit is reached.
Real-Time Decision Stream Dashboard
A live monitoring UI streams every agent decision — which tool was called, what the analysis concluded, what patch was generated — giving engineers full transparency without requiring them to dig through logs.
Impact & Results
Automated the full CI failure-to-fix cycle, eliminating manual log analysis for common failure patterns
Demonstrated production-viable agentic DevOps using LangGraph's stateful, multi-actor graph architecture
Built a reference architecture for self-healing pipelines that scales from solo projects to team repositories
Enjoyed this project?
There's plenty more where that came from.