Model setup

Vicode works with the model you choose. Run one locally with Ollama, use hosted Ollama, or plug in an OpenAI API key.

Pick whatever fits your hardware, your privacy needs, and the task.

Local Ollama

1. Install Ollama for Windows.
2. In Vicode, open provider settings.
3. Start or connect the local Ollama runtime.
4. Pull a model from settings, or run ollama pull <model> and refresh.

Hosted Ollama

1. Add your Ollama API key in provider settings.
2. Refresh to load hosted models.
3. Choose the model for the current task.

Hosted models are not offline-only. The selected prompt and context are sent to the chosen hosted provider.

OpenAI fallback

1. Open provider settings.
2. Add your OpenAI API key.
3. Use OpenAI when your local hardware or model is not cutting it for a task.

Never paste API keys into bug reports, screenshots, or shared docs.

Choosing a model

• The in-app Models catalog browses the Ollama library and Hugging Face GGUF variants, with recommended models and per-variant size and context details.
• For coding-style tasks, give the model enough context when your hardware allows.
• Ollama labels such as tools, thinking, vision, code, and cloud are useful hints, not guarantees that a model will handle every task in Vicode.
• Smaller local models are often enough for answering questions from your notes and code.
• Heavier reasoning and larger coding tasks usually need bigger local models, hosted Ollama, or OpenAI.