Hi everyone, ![]()
I wanted to share an update on a methodology and community tool I’ve been maintaining to get the most out of our instances: n8n-as-code (a CLI/VS Code extension that enables bidirectional GitOps sync for n8n).
We recently crossed a symbolic milestone of 100 stars on GitHub
, but more importantly, the tool has matured a lot recently to solve a major pain point: working with AI coding agents.
If you’ve ever tried to ask an LLM to generate or edit a raw n8n JSON workflow, you know it often hallucinates parameters or breaks the structure.
To fix this, we pushed some major updates under the hood:
-
Transition to TypeScript: Local workflows are now saved, synced, and merged as .workflow.tsfiles. It provides better readability, type safety, and it’s the perfect format for AI coding agents (like Claude, Copilot, or Cursor) to manipulate workflows reliably. -
Rationalized Git-like Sync: We revamped the core engine to use a standard Git-like flow (list,fetch,pull,push,resolve). It makes version control and conflict resolution between your local code and your remote n8n instance much more predictable.
But honestly, the best part of this journey has been seeing this community come alive. When you build an open-source tool, you usually code alone in your corner. Lately, the repo has been getting highly relevant Issues, great architectural discussions, and the very first Pull Requests from other developers here.
Going from a solo side-project to merging external contributions is an amazing feeling. So, a massive thank you to anyone from this forum who took the time to test it, report a bug, or write some code. ![]()
If you want to set up a clean GitOps workflow for your n8n setup, or give your AI coding agents some “n8n superpowers,” the repo is right here:
https://github.com/EtienneLescot/n8n-as-code
I’d love to keep this momentum going. Feedback, ideas, and PRs are always welcome.
Curious to hear your thoughts: what techniques are you guys currently using to version control your instances, and has anyone else been experimenting with LLMs to build workflows?
Cheers!