LLMCopilot

中文

This is a Visual Studio extension based on Ollama. Thanks to Ollama's localized deployment and powerful LLM (Large Language Model) support, you can get code suggestions and tips similar to GitHub Copilot, but all data is local and can even run offline.

If this extension is helpful to you, please give us a 5-star rating on the marketplace.

Version History

Version History

• Please note that the shortcut key to accept all predictions has been changed to Alt+Q to reduce misinterpretation of combinations like Ctrl+C during normal input. For more details, see the Settings section.

Features

• Based on Ollama's local large model, all data is stored locally and can be used even offline.
• Assists programming with large model capabilities:
  • Explains code
  • Auto-completion
  • Code error checking
  • Adds comments
  • Chats with the local large model
  • Generate Unit Test

Settings

• Supports customization to use different large models, allowing you to experiment:

  • Custom chat large model
  • Custom auto-completion large model

• Supports customization of the "Fill in the middle" token, compatible with more large models.

• Auto-completion switch, default is off

  • Press Alt+Q to accept all predictions
  • Press numbers 1-9 to accept the top N lines of predictions
  • Press ESC to reject predictions

• Note! The default large model is DeepSeek-Coder:6.7b. If your VRAM is less than 4G, it is recommended to use the DeepSeek-Coder model. For VRAM of 8G and above, it is recommended to use DeepSeek-Coder:6.7b.

• If VRAM is less than 16G, it is not recommended to use different large models for chatting and code completion simultaneously, as this will cause the Ollama server to frequently load different models, leading to slow response and affecting the experience.

• Supports setting the model response language to Chinese or English (depending on model support).

Screenshots

Acknowledgements

• Thanks to Ollama for providing an excellent large model server backend and easy-to-use API interface.
• Thanks to OllamaSharp for the client to access the Ollama API.
• Thanks to MdXaml for markdown rendering in the chat window.
• Thanks to privy for the prompt templates.

Many thanks to them for providing excellent tools.