I am trying to configure Jade with LM studio to use a local model with my 64GB of VRAM and save on cloud tokens. I tried different models, enabling CORS, etc. If the URL for the server is incorrect, XOJO crashes. With the correct URL, Jade gives the following message:
“Error occurred while receiving response from server Check that the local server is running.”
The server is running. I can even see the accurate prompt output in the LM studio console? Why is it not getting back to XOJO?
Since others have already DMed me, the key is for LM Studio, local only. This machine is behind firewalls and uses local-only auth. The key being visible doesn’t matter at all, and might just help the user with their local LLM setup.
I also encountered the unhandled exceptions that the OP reported, assuming they’re the same ones, and I reported it to Xojo here.
FWW I gave up on local LMs. On my MacBook Pro M4 Pro 24GB it is very slow. But I only used Qwen 2.5 Coder 14B Instruct. Never tried others. Are there better and faster LMs?
Qwen 3 Coder is fairly quick on my M2 Max 64GB, but it’s not very good. Other Qwen models are slower but more accurate. Gemma seemed to forget everything almost immediately.
I guess 24GB just isn’t enough for running coding LMs. Even analyzing a small Xojo project took hours and hours, and sometimes it would just spit out gibberish. That’s when I finally understood why the RAM buzz is so big.
I now mostly use Claude or the basic ChatGPT tier, which is pretty cheap, and I just copy and paste my code manually. I never have to worry about token usage. It’s not ideal, but it’s affordable.
The only thing I may worry about is that sharing files with them maybe risky. Not sure though - the paying tier states it does not share code.
It’s on my to-do list to use ChatGPT together with a free LLM but this needs thorough testing and - yes - speed measurements.
You are really missing out. Using ChatGPT directly on the code is a game changer. It needs a different approach to development. It’s more up front design work and you will need the Xojo MCP server. But I develop faster than ever with a way better quality.
