Guides · Updated 2026-10-17

Set up a free local coding assistant (Continue, VS Code, Cursor)

Use Ollama models for chat, edits and tab autocomplete in VS Code with Continue, plus the honest truth about local models in Cursor.

A local coding assistant keeps your code on your machine and costs nothing per request. It won’t match the best cloud models on hard tasks, but for explanations, small edits, tests and autocomplete it’s genuinely useful.

1. Pick your models

You want two: a chat/edit model and a small, fast autocomplete model.

ollama pull qwen2.5-coder:7b
ollama pull qwen2.5-coder:1.5b

2. Continue in VS Code or JetBrains

Continue is an open-source assistant extension with first-class Ollama support.

1. Install Continue from the VS Code Marketplace (or the JetBrains plugin).
2. Open its config (config.yaml) from the Continue panel’s settings and add:

name: Local
version: 0.0.1
schema: v1
models:
  - name: Qwen Coder 7B
    provider: ollama
    model: qwen2.5-coder:7b
    roles: [chat, edit, apply]
  - name: Qwen Coder 1.5B
    provider: ollama
    model: qwen2.5-coder:1.5b
    roles: [autocomplete]

3. Pick “Qwen Coder 7B” in the chat dropdown and start asking. Autocomplete suggestions appear as you type.

“Model requires more system memory” error? Continue may request a larger context than Ollama’s default. Lower it with defaultCompletionOptions: { contextLength: 4096 } on that model, or use a smaller model.

3. VS Code’s own chat

Ollama also publishes an official Ollama extension for VS Code Chat. Install it, open Chat, and choose a model from the Ollama section of the model picker. For local models, raise the context length in Ollama’s settings (Ollama recommends at least 64K for this) if you have the VRAM.

4. Cursor: possible, but not truly local

Cursor doesn’t connect to localhost models. Its requests are routed through Cursor’s servers, so the “Override OpenAI Base URL” setting needs a public HTTPS endpoint. The workaround is a tunnel (ngrok or Cloudflare Tunnel) to your Ollama server, with a dummy API key. That means:

• your prompts still pass through Cursor’s backend, so you lose the privacy benefit;
• your tunnel is exposed to the internet, so protect it.

If privacy is the goal, use Continue, the Ollama VS Code extension, or another tool that talks to Ollama directly (Zed, Cline and Roo Code all support Ollama).

Agentic coding

For agents that read files and run commands, use models trained for tool use: Devstral Small 2, Qwen3-Coder 30B or Qwen3.6 27B coding, and give them a large context (32K+), which in practice means 16-24 GB of VRAM.

Want a ready-made local app to build on? The Local-LLM App Starter kit wires up Ollama chat, RAG and a GPU fit checker.