OpenCursor / Local models
Local AI coding with Ollama and LM Studio
OpenCursor works with models that run on your own machine. Nothing leaves it, there is no bill, and it keeps working without internet.
Updated October 29, 2026
What works
- Ollama, LM Studio, or any server with an OpenAI compatible API.
- On this computer or another one on your network, such as a desktop with a strong GPU.
- Chat, the agent with tools, Cmd+K edits and Tab completion all run on local models.
- Fill in the middle coder models such as Qwen Coder or Codestral make Tab completion fast and accurate.
Set up with Ollama
- Install Ollama and pull a model, for example
ollama pull qwen2.5-coder:7bfor Tab and a larger model for chat. - Open OpenCursor, open the model menu and choose Local. Ollama is found on its usual address, http://localhost:11434.
- Pick the model for chat and, under Settings, a separate model for Tab if you like.
What to expect
Small local models are quick and good for completion and focused edits. Long agent tasks across many files need a large model and a strong machine, or a hosted model. You can mix them: Tab on a local coder model, chat on whatever you prefer.
Questions
Does OpenCursor send my code anywhere when I use a local model?
No. Requests go to the address you set, nothing else. The only network call OpenCursor makes on its own is a once a day anonymous install count, which you can turn off.
Which local model should I start with?
A Qwen Coder model in a size your machine runs comfortably is a good start for both chat and Tab.
More: Claude Code and Codex · Free Claude Code alternative · Free models · Cursor alternative · Compare · Void alternative · Roo Code and Continue