I’ve been building Runsight — a YAML-first workflow engine for AI agents.

The idea is simple: agent workflows should be as controllable and reviewable as the rest of your codebase.

You design workflows visually, but everything gets written as YAML straight to your filesystem. From there it behaves like real engineering artifacts — versioned in Git, reviewed in PRs, and easy to reason about.

Production reality is messy, so Runsight is built for it:

• Git-native workflows — no hidden state in databases, just YAML in your repo

• Cost visibility per run — understand agent spend before it hits your invoice

• Runtime control — pause a running workflow, change the prompt, and resume instantly

No redeployments. No black boxes. No “hope it works at 2 AM” engineering.

It’s for teams running agents in production who want the same discipline they already have for software: code review, version control, and operational control when things go wrong.

Open source. Self-hosted.