Mendral

Mendral is an AI DevOps agent that autonomously diagnoses CI failures, fixes flaky tests, and opens fix PRs. We make CI green and fast, so engineers stop wasting time on broken builds and keep on shipping.

AI-generated code is increasing volume faster than traditional CI infrastructure can handle. CI was built for humans. The volume of code being shipped today requires a new layer of automation. Mendral is that layer.

We're two engineers who spent the last decade building developer infrastructure. Sam was Docker's first hire and VP of Engineering, scaling the team from 3 to 100+. Andrea authored Docker's first commit. We co-founded Dagger (YC W19) with Solomon Hykes, then started Mendral (YC W26) to solve the reliability problem we kept hitting across every team we worked with.

We have paying customers, real commercial traction, and moving fast.

Our mission is to make software delivery self-healing.

We build an AI agent that monitors PRs and CI runs in real time, diagnoses failures, fixes the root cause, and opens fix PRs automatically. We don't quarantine flaky tests. We fix them.

Our agent uses a multi-model architecture (Anthropic, OpenAI) where different models handle different cognitive tasks. Haiku for fast triage, Sonnet for analysis, Opus for complex reasoning and code generation. Picking the right model for the right job is a core part of our system design.

We write Go, deploy on Google Cloud Run, and integrate deeply with GitHub (via GitHub App) and Slack. Our customers onboard in under 5 minutes.

The kinds of problems we're solving:

1. How do you diagnose a CI failure from terabytes of logs without a human reading them?
2. How do you determine whether a test failure is flaky or a real regression?
3. How do you generate a correct fix PR with enough context to pass CI on the first try?
4. How do you build an agent that earns trust from senior engineers who are skeptical of AI?
5. How do you scale multi-agent orchestration across hundreds of concurrent CI runs?