
I built a bridge that lets Claude and Copilot natively refactor FileMaker scripts inside VS Code. I’m demoing it live on Tuesday.
Hey everyone,
We all know the massive wall we hit when trying to use modern AI tools with FileMaker. If you paste native FM XML into Claude or ChatGPT, you get hit with a massive token tax (it’s about 5.6x larger than standard code), and the AI usually hallucinates because it struggles to parse the proprietary tags. The native Script Workspace just wasn't built for the agentic era.
So, I bypassed it.
I built ai2fm—a bidirectional compiler that translates FileMaker XML into a deterministic, linter-enforced textual language (.fmscript). This lets you copy a script out of FileMaker, paste it into VS Code, and let autonomous AI agents actually read, refactor, and audit your business logic accurately.
Richard Carlton invited me onto FMTraining.tv this Tuesday to do a live "Vibe Coding" demonstration.
I’ll be taking legacy scripts, dropping them into VS Code, and deploying a triad of AI agents to instantly refactor the logic, write the documentation, and audit it for safety. I’ll also break down the zero-retention XSLT architecture we use to ensure your proprietary schema stays completely private.
If you are tired of the XML token tax and want to see how to actually use LLMs for professional-grade FM engineering, come hang out.
When: Tuesday, 05/26/2026 at 1PM Pacific Time Where: https://www.twitch.tv/fmtraining
Happy to answer any technical questions about the compiler architecture in the comments!