AI/Prompt-Driven Blueprints
Description
Mostly a discussion point to start here, but with our Blueprint architecture, and now enabling previews of Blueprints without the need for a full NR-instance, could we open up a text box and let users ask an LLM to generate Node-RED flows for them as a starting point based on a given prompt, e.g. "Create me a CRUD API"
ChatGPT for example would struggle with Dashboard-based content given how new it is, we could look at training our own, but source flows would be a challenge
Which customers would this be available to
None
### Design
- [ ] https://github.com/FlowFuse/flowfuse/issues/3916
### Development
- [ ] https://github.com/FlowFuse/flowfuse/issues/3921
- [ ] https://github.com/FlowFuse/flowfuse/issues/3922
- [ ] https://github.com/FlowFuse/website/issues/2229
I've used Gemini (Google) with D2.0, and that worked out well for me so far.
Icebox-ing this as after experimentation, the LLM for flow generation (from a quick attempt) was not up to standards we'd need. Would require dedicated resource to train custom LLMs
With new AI work in progress, we should reevaluate the possibility of this. It would be a significant step forward in bringing AI into FlowFuse for the purpose of greatly accelerating time to value and delivering on our low-code value. Related to https://github.com/FlowFuse/node-red/issues/26
Just to emphasise the _why- on the reevaluation. LLMs have come a long way since we experimented a year ago, it's worth trying again, and trying a hybrid approach of ML/LLM generated suggestions.