A practical guide to running a private, on-device AI coding assistant using two approaches: VS Code + Ollama + Continue extension, and Cursor with local model support. Covers installing Ollama, pulling DeepSeek-Coder-V2 (16B or lite), configuring the Continue extension with context providers, and a side-by-side quality comparison of local 7B/16B models versus GPT-4o. Key findings: local models handle boilerplate, known framework patterns, and single-file completions well, but fall short on multi-file reasoning and novel API usage. The VS Code + Ollama + Continue stack is fully free and guarantees code never leaves the machine, while Cursor offers more polish but still routes prompts through its servers even in privacy mode.
Table of contents
Table of ContentsWhy Go Local? The Privacy and Cost Case for On-Device AIThe Two Paths: Cursor vs VS Code + Ollama + ContinueSetting Up Ollama with DeepSeek-Coder-V2Integrating Continue into VS CodeThe Cursor Alternative: Built-In AI with a Privacy ModeQuality Comparison: Local 7B/16B Models vs GPT-4oImplementation Checklist and RecommendationsFinal ThoughtsSort: