Announcing: riddlg

riddlg is Ossum’s new tool for generating RIDDL. Here are its main features:

• Validation - Parse and validate RIDDL files (same checks as riddlc)
• Documentation Generation - AsciiDoc or MkDocs sites from your model
• API Specification Generation - Smithy, gRPC, or OpenAPI
• AI RIDDL generation - Describe a system in plain language, get a validated RIDDL model; large systems can be built compositionally across multiple files
• Server Mode - Run as a localhost HTTP/WebSocket service (used by Synapify)
• MCP Mode - Run as an MCP stdio server for AI assistants
• Code Generation (Pro) - Quarkus project skeletons, with optional AI-filled method bodies that are compile-verified

Developing riddlg on an M2 Mac was brutal and took 4 weeks because of LoRA training. The GPU was running so hot that macOS experienced many thermal shutdowns (the computer goes to sleep). So, I put a shop fan on it:

That helped it run to completion with fewer thermal shutdowns. It now generates RIDDL reliably from natural language using a single command and a decent amount of patience (15 mins or so). That is still better than trying to hand-code it, and it will speed up considerably on more advanced hardware.

It can also generate API specifications and code templates, which riddlg or another competent AI coding agent can further expand into working code that executes the model.

riddlg has been validated against all 187 models in the riddl-models repo.

For further details, please see the documentation here.