Skip to content

Repository files navigation

DeepSeek for Copilot

Use DeepSeek V4 models in Copilot Chat.

Setup

  1. Install the extension.
  2. Ctrl+Shift+PDeepSeek: Set API Key → enter your key from platform.deepseek.com.
  3. Open Copilot Chat, pick a DeepSeek model. Chat.

Model Picker

Each model has a gear icon with two dropdowns:

Setting Options Default
Thinking Effort None, High, Max High
Temperature Balanced (1.0), Precise (0.2), Creative (1.3), Max (1.5) Balanced

No global settings needed, everything lives in the picker.

Inline Completion

DeepSeek FIM suggestions as you type. Off by default.

Ctrl+Shift+PDeepSeek: Toggle Inline Completion.

Setting Options Default
deepseek.inlineCompletion.enabled on, off off
deepseek.inlineCompletion.model deepseek-v4-flash, deepseek-v4-pro flash
deepseek.inlineCompletion.debounceMs idle time before a request 300
deepseek.inlineCompletion.maxTokens tokens per suggestion 128
deepseek.inlineCompletion.maxLines lines per suggestion 10

Suggestions are cut where they leave the block they started in, using a tree-sitter parse rather than a guess. Grammars ship for:

typescript typescriptreact javascript javascriptreact python go rust java csharp cpp c cuda-cpp ruby php shellscript powershell css ini properties

Other languages still get suggestions, just no block trimming, bounded only by maxLines. Suggestions are also re-indented to match the editor's tabs/spaces.

Turn off Copilot's own inline suggestions first (github.copilot.enable), otherwise both providers compete for the same ghost text.

If suggestions look untrimmed, check DeepSeek: Show Logs for a tree-sitter warning.

Commands

Command What it does
DeepSeek: Set API Key Store API key
DeepSeek: Clear API Key Remove stored key
DeepSeek: Set Temperature Pick a temperature preset or custom value
DeepSeek: Toggle Inline Completion Turn FIM suggestions on or off
DeepSeek: Open Settings Jump to the DeepSeek settings
DeepSeek: Show Logs Open output channel (token counts, cache hit rate)

Builds

Releases attach two files:

File Use
copilot-vscode-deepseek-<version>.vsix Install this one.
...-nes.vsix Enables proposed APIs (inlineCompletionsAdditions). Sideload only, never published to the Marketplace.

The -nes build also needs this in ~/.vscode/argv.json, which survives restarts:

{ "enable-proposed-api": ["DenizhanDaklr.copilot-vscode-deepseek"] }

Build locally with npm run package and npm run package:proposed.

Requirements

  • VS Code 1.125+
  • GitHub Copilot subscription
  • DeepSeek API key

License

MIT

About

Use DeepSeek models (deepseek-chat, deepseek-reasoner) in VSCode Copilot Chat

Resources

Stars

26 stars

Watchers

2 watching

Forks

Releases

Packages

Used by

Contributors

Languages