Last year, my co-founder and I were building Lagom — an AI text humanizer.
We made the mistake of pushing simultaneously. Maneesh on the frontend, me on the backend. What followed was a cascade — merge conflicts, Node version mismatches, Vercel deployment breaking, GitHub Actions failing across the board.
Two hours. For something that should have taken minutes.
The frustrating part wasn't the errors. It was that every tool told us exactly what broke. None of them fixed anything. We were still the repair loop.
That moment stuck with us.
So we built Prash
Prash is an autonomous CI/CD repair agent for GitHub Actions.
Your workflow fails → Prash gets notified instantly
Downloads and reads the full logs
AI diagnoses the root cause
Opens a pull request with the exact fix
Re-runs CI on the fix branch
Marks it Verified ✓ only when all checks pass
If the first fix fails, Prash re-diagnoses with the failed output as additional context and tries again.
What we learned building it
Trust tiers matter more than raw accuracy.
Not every fix should be auto-applied. We built three tiers — auto-apply for high confidence fixes, review-recommended for anything uncertain, manual-required for complex logic changes.
The verification loop is the actual moat.
Anyone can prompt an LLM to suggest a fix. Watching the fix run, checking the result, and iterating that's the hard part.
Most CI failures fall into 5-6 recurring patterns.
Node version mismatches, missing env vars, dependency drift, flaky tests. The taxonomy is smaller than you'd think.
Logs over 80KB degrade diagnosis quality.
We built smart truncation to handle this without losing the important context.
Where we are
Launched yesterday April 30th, 2026. Actively onboarding first testers. Early, doesn't catch everything yet, learning fast.
If you've ever lost hours to a CI failure that turned out to be a one-line fix, we built this for you.
Free for all. 2 minutes to connect.
prashbydrufiy.vercel.app
Top comments (0)