DeepSeek for Copilot

Use DeepSeek models in GitHub Copilot Chat, plus optional inline code completions.

Also from the same developer: Lodestar walks you through how code works, explains anything you hover, and reviews your changes on any model in VS Code, including this extension's. 3 free runs included, then $5 one-time.

Run DeepSeek: Set API Key and paste a key from platform.deepseek.com, or add it under Manage Models in the Copilot Chat model picker.
Pick a DeepSeek model in the model picker.
Set reasoning effort and temperature right in the picker.

Models, context windows and effort levels come live from DeepSeek's API, with models.dev filling any gaps, so new models show up without an update. Brand-new models appear right away with safe default limits until models.dev lists them. Copilot's context window indicator works as usual.

Model picker

Option	Values
Reasoning Effort	Live per model, currently Off, Low, High and Max. Starts at DeepSeek's own default.
Temperature	Balanced (1.0), Precise (0.2), Creative (1.3), Max (1.5) or Custom

Inline completion

Off by default. Run DeepSeek: Toggle Inline Completion, and turn off Copilot's own inline suggestions (github.copilot.enable) so the two don't compete.

Suggestions come from DeepSeek Flash. A tree-sitter parse trims them at the end of the block they started in, or after one statement when they start new ones. Grammars ship for TypeScript, JavaScript, Python, Go, Rust, Java, C#, C/C++, Ruby, PHP, shell, PowerShell, CSS and INI; other languages are only capped by maxLines.

Settings

Setting	Default	Notes
`deepseek.baseUrl`	`https://api.deepseek.com`	API base URL
`deepseek.temperature`	`1`	Used when the picker's temperature is Custom
`deepseek.apiKey`	empty	Plain-text fallback; Set API Key stores the key securely instead
`deepseek.inlineCompletion.enabled`	`false`	Inline completion on or off
`deepseek.inlineCompletion.debounceMs`	`300`	Idle time before a request is sent
`deepseek.inlineCompletion.maxTokens`	`128`	Tokens per suggestion, up to 4096
`deepseek.inlineCompletion.maxLines`	`10`	Lines per suggestion

Commands

All under DeepSeek: in the Command Palette: Set API Key, Clear API Key, Set Temperature, Toggle Inline Completion, Open Settings, and Show Logs (token counts and cache hit rate).

Builds

Install from the Marketplace or a release .vsix. Releases also attach a -nes.vsix with proposed inline-completion APIs for sideloading; it needs { "enable-proposed-api": ["DenizhanDaklr.copilot-vscode-deepseek"] } in ~/.vscode/argv.json. Build locally with npm run package or npm run package:proposed.

Requires VS Code 1.125+ and GitHub Copilot Chat. MIT license.

DeepSeek for Copilot

Denizhan Dakılır

DeepSeek for Copilot

Model picker

Inline completion

Settings

Commands

Builds