DeepSeek FIM Autocomplete

A deliberately tiny VS Code extension that provides inline autocomplete suggestions with DeepSeek's Fill-In-Middle completion API.

Features

• TypeScript extension code.
• VS Code inline completions at the cursor.
• Status bar indicator that spins while DeepSeek requests are in flight.
• Prefix and suffix context sent to DeepSeek FIM.
• Simple JSON settings for API key, model, base URL, token limit, temperature, trigger mode, timeout, and context size.
• Unit tests for the core request/config/context logic.

Settings

Add these to your VS Code settings.json:

{
  "deepseekFimAutocomplete.apiKey": "YOUR_DEEPSEEK_API_KEY",
  "deepseekFimAutocomplete.model": "deepseek-v4-pro",
  "deepseekFimAutocomplete.baseUrl": "https://api.deepseek.com/beta",
  "deepseekFimAutocomplete.maxTokens": 128,
  "deepseekFimAutocomplete.temperature": 0.2,
  "deepseekFimAutocomplete.triggerMode": "automatic"
}

Set deepseekFimAutocomplete.triggerMode to manual if you only want requests when running DeepSeek FIM: Trigger Inline Completion. The default keybinding is Ctrl+Alt+Space (Cmd+Alt+Space on macOS).

Development

npm install
npm test

To run inside VS Code, open this folder and press F5 to start an Extension Development Host.

Package locally

npm run package
code --install-extension deepseek-fim-autocomplete-0.0.3.vsix

After installing, reload VS Code and set deepseekFimAutocomplete.apiKey in settings.json.

Publish to the VS Code Marketplace

Create .env from .env.example and fill in your marketplace token:

cp .env.example .env

Required keys:

VSCE_PERSONAL_ACCESS_TOKEN=your-vscode-marketplace-personal-access-token
VSCE_PUBLISHER=simon-liang

The publish script loads .env with dotenv/config. Then validate the publish configuration without publishing:

npm run publish:vsce:dry

Publish the current package version:

npm run publish:vsce

You can pass normal vsce publish arguments after --, for example:

npm run publish:vsce -- patch
npm run publish:vsce -- --pre-release