PagerLens

You're at a party when your app crashes. Instead of rushing to your laptop, PagerLens finds the bug, writes the fix, gets it reviewed, and sends you a ready-to-approve solution.

Comment

Sentry, Code Rabbit and Daytona on Smart Glasses!!

Link to demo video: https://drive.google.com/drive/folders/1_lUlLeyYF8TSXi5vnKvJ9ioWUWdQt4Nu?usp=sharing

Inspiration

Picture this: you're at your best friend's wedding, champagne in hand, when your phone buzzes. Production is down. You know what comes next. Apologize to the bride, find a quiet corner, pull out your laptop, and miss the bouquet toss for a bug that turns out to be a missing null check.

Every on-call engineer has been there.

Now imagine a different night. Your smart glasses light up: "Payment API error. AI analyzing." You keep dancing. A minute later: "Fix ready. Say 'deploy' to merge." You glance, say "deploy it," and catch the bouquet.

That's PagerLens. An AI teammate that fixes bugs and asks for approval through your smart glasses. No laptop. No leaving the party.

What it does

PagerLens is an automated incident response pipeline with a hands-free interface:

  1. Sentry detects a production error and sends an alert
  2. Smart glasses notify you with a summary of the incident, no phone needed
  3. Daytona spins up an isolated sandbox environment with the codebase
  4. Claude AI analyzes the stack trace, locates the bug, and generates a fix
  5. A GitHub PR is created automatically
  6. CodeRabbit reviews the code and provides feedback
  7. Claude iterates on the feedback and pushes a second commit
  8. Your smart glasses show the fix summary and ask for approval
  9. You say "deploy it" and the PR is merged, completely hands-free

From error to fix in under 3 minutes. No laptop required.

How we built it

  • Smart Glasses Integration: Real-time notifications and voice-controlled approval flow
  • Daytona SDK: Creates ephemeral sandboxes where Claude can safely read files, run commands, and execute tests without touching production
  • Sentry Webhooks: Trigger the pipeline when errors occur
  • Claude AI (Anthropic): Analyzes errors, navigates codebases, and generates patches using tool-use capabilities
  • CodeRabbit: Provides AI code review, and we built an iteration loop where Claude addresses feedback and pushes additional commits

The entire flow is automated but human-approved. Engineers review and approve through their smart glasses without touching a keyboard.

Challenges we ran into

  • Getting Daytona paths right: Claude was searching in /workspace/repo but Daytona clones to /home/daytona/repo. Debugging this required tracing through the entire execution flow.
  • CodeRabbit iteration loop: CodeRabbit posts "processing" comments before the actual review. We had to filter these out and build robust verdict detection to know when to iterate vs. when the review passed.
  • Branch management: Initially, we tried cloning directly to a feature branch, but Daytona needs a valid ref. We switched to cloning the default branch first, then creating the feature branch inside the sandbox.
  • State machine transitions: Our incident state machine didn't initially allow going back from PATCH_PROPOSED to TRIAGING for iterations. We had to update the valid transitions.
  • Hands-free UX: Designing an interface that delivers enough context through smart glasses without overwhelming the user. Condensing stack traces and diffs into glanceable summaries.

Accomplishments that we're proud of

  • Hands-free incident response: From notification to merge without touching a laptop
  • Two-commit PRs: Our PRs show the AI iteration process. First commit is the initial fix, second commit addresses CodeRabbit's feedback Safe execution: All code analysis and patching occur in isolated Daytona sandboxes and are reviewed by CodeRabbit.

What we learned

  • Building reliable AI agents requires extensive error handling and state management
  • The combination of AI code generation + AI code review creates a powerful feedback loop
  • Smart glasses change the incident response UX. You can stay present in the real world while handling production issues

What's next for PagerLens

  • Voice Q&A with Claude through smart glasses for complex incidents
  • Support for more complex multi-file fixes
  • Learning from past incidents to improve fix accuracy
  • Slack/Discord notifications with fix summaries

Built With

  • anthropic
  • bun
  • claude-ai
  • coderabbit
  • daytona
  • fastify
  • github-api
  • mongodb
  • sentry
Share this project:

Updates