LM Studio

LM Studio is a popular local LLM GUI, which supports downloading, quantizing, and running various open-source models locally. Cherry Studio can serve as a front end to connect to LM Studio's local service, providing a better chat experience while preserving local privacy.

Prerequisites

1. from Official website Download and install the client
2. Download at least one model in LM Studio (Llama 3.x 8B or the Qwen series is recommended to try first)
3. Open the top of LM Studio Server tab, click Start Server(default port 1234)

Configure in Cherry Studio

• After opening Settings → Model Services, find LM Studio Provider to enter the details page
• API Address By default http://localhost:1234, if you've changed the LM Studio port, please update it accordingly
• API key Can be left blank (local inference does not require authentication), or fill it in after enabling authentication in LM Studio
• Click Get model list, Cherry Studio will automatically fetch the models loaded in LM Studio

Model list empty? LM Studio only exposes already Load into memory models; those not loaded will not appear in the list. Go back to LM Studio, load the model first, then come back and "Get model list".

Recommended usage

Scenario	Recommendation
Privacy-sensitive conversations	Choose a small model (under 8B) to run locally, completely offline
Apple Silicon (M-series chips)	LM Studio uses the MLX backend, with significantly higher efficiency than llama.cpp
Embedding model	LM Studio can also load embedding models for knowledge bases

Differences from Ollama

	LM Studio	Ollama
Form	GUI + Server	Command line / background service
Model management	GUI browse/download	ollama pull
API	OpenAI-compatible	OpenAI-compatible
Suitable for	users who prefer graphical interaction	users who prefer command line / Docker deployment

Both can connect to Cherry Studio; just choose according to your preference.

Common questions

• Cherry Studio can't connect: check whether Server in LM Studio has been started (green dot status)
• Responses are extremely slow: the model is too large / VRAM is insufficient; switch to a smaller model or a more aggressive quantization (e.g. Q4 → Q3)
• Garbled text / truncated output: the context length exceeds the model's limit; increase it in LM Studio n_ctx

---

Get help and submit feedback

If you encounter any questions, bugs, or have suggestions for feature improvements during configuration or use, please refer to Feedback and Suggestions for the official channels provided.