Client Side AI
Ossum shifts its AI tools away from cloud to the client side.
Synapify and RIDDL are undergoing a change to make AI tooling easier to use and produce better results faster. Previously, Claude-based AI tooling for RIDDL generation was cloud-based, which introduced latency and other issues. The AI industry is trending towards AI inference on the client-side which results in faster and more efficient AI tooling. Correspondingly, RIDDL generation is making that shift as well to:
- reduce round trip latency (generation makes many round trips)
- improve security by not transmitting models and code off the client machine
- improve the overall user experience
- increase productivity and efficiency
Soon, a new tool, riddlg, will be delivered and installable for free. The MCP server, documentation generation, and API generation from RIDDL models will also be completely free. Using some of its advanced features (code generation) will required a subscription.
riddlg will be bundled with Synapify or you can download and install it yourself with homebrew if you don’t want to use Synapify. It can run in various modes, too:
- server - provides an HTTP/REST API for all functionality. This is how Synapify talk to it.
- cli - invoke commands to generate RIDDL or other artifacts
- mcp - use it as an MCP stdio server as a companion to your AI (e.g. Claude Code)
The ETA for riddlg is roughly mid-July, 2026.