Lemonade Local Models

[PhillLittlewood]

Every Lemonade Model I try fails in OpenMaic. I have tried Deepseek, Gemma4 E4B, Llama3 and Bonsai. But OpenMaic always throws a Bad request 400 Error. Can any adivise?

[cosarah]

Thanks for reporting. Lemonade is intended to work without an API key in OpenMAIC, so an empty key should normally be OK.

Could you check whether the Lemonade “Test connection” also returns 400? If the short test passes but course generation fails, this is likely caused by the model being loaded in Lemonade with too small a context window. If the short test also fails, please share the Lemonade server error body/log so we can identify which OpenAI-compatible request parameter is being rejected.

[PhillLittlewood]

Thanks for the reply but I have decided to use LMStudio instead Lemonade is far too slow. Just working to get OpenMaic to work with ComfyUI.

[richardchen874-sys]

This sounds more like a local OpenAI-compatible server compatibility issue than a model issue.

For Lemonade / LM Studio style local backends, I’d usually check:

• whether the short test request works
• the exact server-side 400 error body
• unsupported request parameters
• context window size used when loading the model
• whether streaming and chat-completions payloads are handled differently

Local backends can look OpenAI-compatible at the endpoint level, but still reject specific parameters or fail once the generated task needs a larger context window.