RWould probably not be too hard to use an open-source model and have a prompt to generate code based mostly on the scripts in the hub. e.g of such models: https://github.com/sahil280114/codealpaca
GitHub
Contribute to sahil280114/codealpaca development by creating an account on GitHub.
RTHat could be a cool internship
TThe scripts is not an issue because we can copy paste it. Alpaca has a long way to be on the same level as gpt4. I think the potential is in training gpt4 on the openflow structure and asking it to build workflows.
RThat's fair, i think having a way to give it a key to multiple service and a few premade prompts might be enough. The open flow format is simple enough with enough exemples
RI'm just thinking that enterprise might enjoy a fully self hosted solution
TMy personal opinion is to skip this and leave it to the community to build while the team focus on the core features of windmill.
RAnd hugging face and data bricks already released competitive models
RAgree agree but we have space for an internship 
ROne possibility is to have a very efficient flow to fill the hub with many integrations
RLike, generate the script, review, test it, approved
Ti know integrations with other services is great for on-boarding and it's something you should focus on. But with deno and gpt4 i can build and test integrations easily. That also can be something the community can handle while the team improve the areas that i can't solve.
RI agree that it's a distraction for us but we have budget to explore the area and a smart young graduate could do a lot without disturbing the team
Tsounds great, let them check Langchain if you are interested in LLMs.
RI think 10 years ago i would have loved such internship
TI'm losing sleep just to keep up with the LLM updates and trying to integrate it with our apps.
RIt's truly the dawn of a new age
Tsome companies are integrating gpt4 or openai api as a node or a script where it help you call them and get response. While using api agents in langchain or openai new plugin structure, you can just talk to gpt4 and it will trigger all the needed api based on the prompts.
Tyou can check their plugin demo about the todo list or the one made by zapier.
RI think something like windmill is a better target for gpt-4 and that instead of centering the plugin system around chatgpt it should be the opposite, but openai has a lot of leverage and brilliance so of course they did it that way
RI both admire openai and actually could have joined them at the very beginning when i was working on rl4j and they were just beginning but hate how hypocrite they are currently
RBut dumb me wanted to finish his master first
Tlol, been there buddy. This is what i meant https://langchain.readthedocs.io/en/latest/modules/agents/agent_toolkits/openapi.html
TWindmill workflows as OpenAPI specs , where langchain will read them and trigger which one based on the prompt.
TFYI if you plan to get into LLMs expect things to change a lot since it's an early phase and a lot of libs are adopting new use-cases.
RBuilding a retriever for the hub would work with something like this: https://blog.langchain.dev/retrieval/
LangChain
TL;DR: We are adjusting our abstractions to make it easy for other retrieval methods besides the LangChain VectorDB object to be used in LangChain. This is done with the goals of (1) allowing retrievers constructed elsewhere to be used more easily in LangChain, (2) encouraging more experimentation with alternative