Add persistence, hierarchical globals, and TF-IDF cosine RAG

[swaekaa]

This PR improves the MUMPS RAG terminal while keeping it lightweight:

• Adds persistent storage for globals using JSON
• Replaces flat keys with hierarchical MUMPS-style storage
• Upgrades RAG scoring to TF-IDF + cosine similarity (no external deps)

Closes #39

[netlify]

✅ Deploy Preview for webxos ready!

Name	Link
🔨 Latest commit	ec2ea3d
🔍 Latest deploy log	https://app.netlify.com/projects/webxos/deploys/695f859c04352b00083a7416
😎 Deploy Preview	https://deploy-preview-40--webxos.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[webxos]

Mumps rag has session only storage for security by design. It's made to run on skupt libraries in webxos.netlify.app/injector (my own custom terminal which ive sanitized as much as possible to prevent privacy concerns). Not all libraries are supported by skulpt (very small amount), my only concern is that it will not work in my terminal. I also was sort of avoiding persistent storage for this idea - more of just a concept/for fun. But it would be cool to expand this type of idea into a back end (that was the intention one day) so that the RAG db can be used across back end sdks integrated with pytorch. Still a concept. I can't view the py for deployment but I appreciate the attempt. Good ideas for sure. Feel free to use the .py for any personal projects or if you want to try to build it into your own projects its open sourced!

[swaekaa]

Mumps rag has session only storage for security by design. It's made to run on skupt libraries in webxos.netlify.app/injector (my own custom terminal which ive sanitized as much as possible to prevent privacy concerns). Not all libraries are supported by skulpt (very small amount), my only concern is that it will not work in my terminal. I also was sort of avoiding persistent storage for this idea - more of just a concept/for fun. But it would be cool to expand this type of idea into a back end (that was the intention one day) so that the RAG db can be used across back end sdks integrated with pytorch. Still a concept. I can't view the py for deployment but I appreciate the attempt. Good ideas for sure. Feel free to use the .py for any personal projects or if you want to try to build it into your own projects its open sourced!

Thanks for the explanation, that makes sense. I didn’t realize it was meant to run in a Skulpt-based sandbox, and I understand why session only storage is important. My changes were more exploratory from a backend perspective. Really appreciate the feedback and the opensource project.
it’s a very cool concept.