AUTO-OPS: AI DevOps Agent

An AI DevOps agent that catches crashes, spins up Daytona sandboxes, tests Claude-generated fixes, and ships CodeRabbit-reviewed PRs with ElevenLabs voice summaries.

Comment 1

Inspiration

We wanted to build something that eliminates the “debug fatigue” developers face — when your code breaks and you spend hours tracing logs, fixing one bug just to introduce another. With AI tools evolving fast, we saw an opportunity to let an autonomous agent not only understand code errors but actually fix them safely in real time. That vision became AUTO-OPS, an AI DevOps agent that acts like your tireless coding teammate.

What it does

AUTO-OPS automatically detects runtime errors from any Python project, analyzes the code, generates a patch, tests it safely in a Daytona sandbox, and opens a pull request — all without human intervention.
It logs traces in Galileo, generates code and reasoning using Anthropic’s Claude, tests fixes inside Daytona environments, stores everything in Tigris, gets CodeRabbit reviews automatically, and even speaks a summary of what it fixed using ElevenLabs.

In short, it’s an autonomous software engineer that handles your crashes end-to-end.

How we built it

  • Frontend: Next.js + Tailwind dashboard showing incoming errors and a live “agent console” timeline.
  • Backend: Node/TypeScript API routes handling error intake, Claude prompts, Daytona sandbox orchestration, and GitHub PR creation.
  • Python client: A lightweight wrapper that catches exceptions and POSTs traceback + source code to AUTO-OPS.
  • APIs integrated:
    • 🧠 Anthropic Claude → analyzes stack traces & generates code patches
    • 🧰 Daytona → spins up isolated test environments and runs patched code
    • 🐇 CodeRabbit → automatically reviews created PRs
    • 📈 Galileo → logs every crash trace and agent decision
    • 💾 Tigris Data → stores incidents, logs, and patches for long-term memory
    • 🗣️ ElevenLabs → turns the AI’s summary into a spoken “fix report”

Everything runs through a coordinated agent loop managed in Node, with visible status updates on the dashboard.

Challenges we ran into

  • Coordinating six separate APIs under one asynchronous workflow.
  • Keeping Claude’s reasoning “safe” so it never writes destructive patches.
  • Managing Daytona environment creation and teardown efficiently under time limits.
  • Getting real-time error feeds to sync smoothly between Python and the dashboard.
  • Handling API rate limits and secret management without breaking demo reliability.

Accomplishments that we're proud of

  • Built a fully autonomous debugging pipeline that touches every partner API.
  • AUTO-OPS can realistically detect, fix, test, and commit a live Python error.
  • The project runs end-to-end in minutes with clear transparency via the console UI.
  • Created a voice-enabled demo where the AI literally talks you through the fix.
  • Clean architecture that other developers can extend to JavaScript or Go next.

What we learned

  • How to combine multiple AI systems — Claude, CodeRabbit, ElevenLabs — into a cohesive workflow that doesn’t collapse under API latency.
  • The importance of sandboxed testing before applying AI-generated code changes.
  • That strong observability (Galileo + Tigris) is essential when you give autonomy to code-writing agents.
  • Building fast is easy; building safe and explainable AI systems is the real challenge.

What's next for AUTO-OPS: AI DevOps Agent

  • Introduce real-time VS Code integration so the agent appears as an in-editor assistant.
  • Build team dashboards for organizations to monitor agent performance across repos.
  • Integrate LLM fine-tuning using historical Tigris logs to make AUTO-OPS learn your project’s coding style.
  • Expand sandbox orchestration to CI/CD pipelines for true self-healing infrastructure.

AUTO-OPS isn’t just a hackathon project — it’s a prototype for the future of hands-free software maintenance.

Built With

Share this project:

Updates