LM Studio
LM Studio is a popular local LLM GUI that supports downloading, quantizing, and running inference for various open-source models locally. BLACK HOLE can connect to LM Studio's local service as a frontend, providing a better conversational experience while maintaining local privacy.
Prerequisites
- Download and install the client from the LM Studio official website
- Download at least one model in LM Studio (recommended to try Llama 3.x 8B or Qwen series first)
- Open the Server tab at the top of LM Studio and click Start Server (default port
1234)
Configure in BLACK HOLE
- Open
Settings → Model Provider, find the LM Studio Provider, and enter the details page - The API Address defaults to
http://localhost:1234. If you have changed the LM Studio port, update this accordingly - The API Key can be left empty (local inference requires no authentication), or filled in if you enable authentication in LM Studio
- Click Fetch Model List, and BLACK HOLE will automatically retrieve the models loaded in LM Studio
Recommended Usage
| Scenario | Recommendation |
|---|---|
| Privacy-sensitive conversations | Use small models (under 8B) locally for fully offline operation |
| Apple Silicon (M-series chips) | Use the MLX backend in LM Studio for significantly higher efficiency than llama.cpp |
| Embedding models | 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 |
| Best For | Users who prefer graphical interaction | Users who prefer command line / Docker deployment |
Both can be integrated with BLACK HOLE Studio. Choose based on your personal preference.
FAQ
- BLACK HOLE cannot connect: Ensure the Server in LM Studio is started (green dot status)
- Response is very slow: The model is too large or there is insufficient VRAM. Switch to a smaller model or a lower quantization (e.g., Q4 → Q3)
- Garbled text / truncated output: The context length exceeds the model's limit. Increase
n_ctxin LM Studio
Get Help and Submit Feedback
If you have any questions, bugs, or feature improvement suggestions during configuration or usage, please refer to the official channels provided in Feedback and Suggestions.