# A fully offline coding assistant in VS Code

- URL: https://spark.tlec.top/en/plays/local-coding-assistant/
- Date: 2026-10-28
- category: coding
- difficulty: intermediate
- vram: 16GB recommended
- source: https://docs.continue.dev
- models: qwen2.5-coder
- tools: Ollama, Continue, VS Code
- platforms: mac, nvidia
- tags: offline, privacy

> Connect the Continue extension to a local Ollama model so completions and chat never leave your machine — great when company code can't go to the cloud.

> Sample entry.

## Approach

- Use a larger model (e.g. 14B) for chat and a small one (1.5B–3B) for inline completion to balance quality and speed
- In Continue's config set the provider to `ollama` and pick separate chat and autocomplete models

## Notes

Smaller completion models mean lower latency. If VRAM is tight, keep completion snappy and let chat be slower.