AgentOps — CI/CD Failure Detective

An AI agent that diagnoses GitLab CI/CD pipeline failures by fetching logs and using Gemini to identify root causes and suggest fixes.

Comment

Inspiration

Every developer knows the frustration of a failed CI/CD pipeline. You get a vague error notification, dig through logs manually, and waste 30 minutes figuring out what went wrong. I wanted to build an agent that does this automatically.

What it does

AgentOps is a CI/CD debugging agent that:

  1. Connects to your GitLab project via the GitLab API
  2. Automatically fetches failed pipeline logs
  3. Uses Gemini to analyze the root cause
  4. Suggests a specific fix with corrected code

How we built it

  • Flask API backend serving the agent
  • Google Gemini as the reasoning engine with function calling
  • GitLab REST API as the data source (pipelines, jobs, logs)
  • Three custom tools: list_pipelines, get_pipeline_jobs, get_job_log
  • Deployed on Render with environment-based secrets

Challenges we ran into

  • Getting Gemini function calling to work with multi-step tool use
  • GitLab API authentication and project path encoding
  • Deploying with environment variables securely on Render

Accomplishments that we're proud of

  • Agent correctly identifies root causes from real log output
  • Multi-step reasoning: fetches pipeline → jobs → logs → diagnosis
  • Live public deployment judges can test themselves

What we learned

  • How to build multi-step agents with Gemini function calling
  • GitLab API structure for CI/CD data
  • How to deploy Python agents as web services

What's next for AgentOps — CI/CD Failure Detective

  • Auto-create merge requests with the fix applied
  • Support for multiple GitLab projects
  • Slack/email notifications when pipelines fail
  • Add Arize observability for agent tracing

Built With

Share this project:

Updates