Ollama

Local chat and auto-complete against models you already pulled. Fully offline.

NBI talks to a local Ollama server for chat and inline completion. No external API calls; all inference stays on your machine.

What you get

  • Local chat. Any Ollama chat model with tool-calling support drives the chat sidebar. Llama 3.x, Qwen, Mistral, and others work out of the box.
  • Local inline completion. A separate allowlist of completion-tuned models powers tab-complete. Smaller, faster models like qwen2.5-coder:7b are the sweet spot.
  • Zero egress. Useful for air-gapped environments, sensitive data, or regulated industries.

Setup

Install Ollama and pull a chat-capable model:

curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3:latest
ollama pull qwen2.5-coder:7b

Then pick Ollama as the chat provider in NBI Settings.

Configuration

{
  "chat_model": { "provider": "ollama", "model": "llama3:latest" },
  "inline_completion_model": { "provider": "ollama", "model": "qwen2.5-coder:7b" }
}

NBI talks to the default Ollama endpoint at http://localhost:11434. Set OLLAMA_HOST if you’ve moved it.

Install NBI