Inspiration
After writing code, developers lose hours to repetitive pipeline work — installing dependencies, running tests, waiting on security scans, deploying, monitoring. When something breaks at 2 AM, someone has to wake up and fix it. I wanted to build an agent that handles the entire post-code lifecycle autonomously.
What it does ShipPilot is an autonomous DevSecOps agent that takes your code from commit to production with zero human intervention across 9 stages:
- Install — dependency resolution with caching
- Test — automated test suite execution
- Security Scan — vulnerability scanning
- Build — artifact creation
- Deploy — zero-downtime deployment
- Smoke Test — post-deploy verification
- Monitor — health checks
- Auto-rollback — instant rollback on failure detection
- Notify — status alerts
Its AI brain (Nebius LLM with heuristic fallback) reasons about failures and decides: retry, rollback, or escalate.
How I built it
- Pipeline: GitLab CI/CD with 9 stages, verified GREEN on a GitLab Ultimate trial
- AI Brain: Nebius-hosted LLM for failure reasoning, with a deterministic heuristic fallback so the pipeline never stalls
- Duo-ready agents: Agent definitions in
.gitlab/duo/(pipeline-guardian, security-remediator, deploy-strategist) designed for the GitLab Duo Agent Platform - Stack: Node.js, GitLab CI/CD, Docker, Nebius AI
Challenges The biggest challenge was making the agent truly autonomous — handling the long tail of failure modes without human input. I solved this with a layered approach: the LLM reasons about novel failures, while the heuristic fallback handles known patterns deterministically.
What's next Deeper Duo Agent Platform integration, multi-cloud deployment targets, and predictive failure detection.
Log in or sign up for Devpost to join the conversation.