Continue is the leading open-source AI coding assistant for VS Code. It works with any LLM and is completely free. By pairing it with DeepSeek V3, you get a GitHub Copilot-like experience at a fraction of the cost.
Prerequisites
- Install VS Code.
- Get an API Key: Sign up at DeepSeek Platform.
- Install Config: Search for "Continue" in the VS Code Marketplace and install it.
Core Configuration: config.json
Continue relies on config.json for all settings.
- Open Config: Press
Cmd + Shift + P(Mac) orCtrl + Shift + P(Win) in VS Code, typeContinue: Open config.json, and press Enter. - Update 'models': Add the following configuration block to the
modelsarray.
Recommended (DeepSeek Official)
{
"models": [
{
"title": "DeepSeek V3",
"provider": "deepseek",
"model": "deepseek-chat",
"apiKey": "sk-your-deepseek-key"
},
{
"title": "DeepSeek R1 (Reasoning)",
"provider": "deepseek",
"model": "deepseek-reasoner",
"apiKey": "sk-your-deepseek-key"
}
],
"tabAutocompleteModel": {
"title": "DeepSeek V3",
"provider": "deepseek",
"model": "deepseek-chat",
"apiKey": "sk-your-deepseek-key"
}
}Note: If
provider: "deepseek"causes errors, your Continue version might be outdated. Update the extension, or useprovider: "openai"with"apiBase": "https://api.deepseek.com".
Using SiliconFlow
For faster speeds in specific regions or to use free tiers:
{
"title": "DeepSeek V3 (SiliconFlow)",
"provider": "openai",
"model": "deepseek-ai/DeepSeek-V3",
"apiKey": "sk-your-siliconflow-key",
"apiBase": "https://api.siliconflow.cn/v1"
}Verify Setup
- Chat: Select "DeepSeek V3" in the Continue sidebar (left) and ask "Hello DeepSeek".
- Autocomplete: Type code in the editor and wait for grey text suggestions. Press
Tabto accept. - Edit: Highlight code, press
Cmd + I, and type instructions (e.g., "Add comments").
FAQ
Q: Autocomplete isn't working?
A: Check the Continue status icon in the bottom right. If it's red, there's an error. Ensure tabAutocompleteModel is configured correctly.
Q: Error with provider: "deepseek"?
A: Recent versions of Continue support DeepSeek natively. If you face issues, fallback to the "openai" provider method.

