DeepSeek for Copilot Chat (Community Fork)

English | 简体中文

Pick DeepSeek V4 from the Copilot Chat model picker — and keep everything else Copilot already gives you.

Unofficial community fork. Published by fallenstarowo from FallenstarOwO/deepseek-for-copilot. Not affiliated with or endorsed by the original author of Vizards/deepseek-v4-for-copilot; the MIT license and original copyright notices are retained. Both extensions register the same deepseek model vendor, so uninstall the official extension before installing this one.

Love DeepSeek's price-performance but don't want to give up GitHub Copilot's agent mode, tool calling, and polished UI? This extension adds DeepSeek V4.1 Flash and other DeepSeek models to the Copilot Chat model selector — with native vision or Vision Proxy, thinking mode, and your own API key.

Why this extension?

• Don't replace Copilot — power it up. No new sidebar, no new chat UI to learn. Just a new model in the picker you already use.
• Agent mode, tool calling, instructions, MCP, skills — all of it still works. Copilot's entire stack, now running on DeepSeek.
• Two ways to work with images. V4.1 Flash receives image attachments natively. V4 Pro keeps its existing text context while a configurable Vision Proxy turns images into descriptions.
• BYOK, pay DeepSeek directly. Your API key, your bill, your rate limits. Stored in the OS keychain, never on disk.

Features

DeepSeek models in the model picker

The picker includes V4.1 Flash and V4 Pro, with long context, tool calling, and configurable thinking effort.

Native Vision and Vision Proxy

Choose the image path that fits the conversation:

• DeepSeek V4.1 Flash handles image attachments natively, without Vision Proxy.
• The DeepSeek V4 Pro entry uses Vision Proxy: an image-capable model first describes each attachment, then the main DeepSeek model receives the description with the conversation. Auto mode selects V4.1 Flash when available, while an explicitly configured VS Code model or API endpoint remains supported.

Context caching requires a full match with a cached prefix.

Thinking Mode and Effort Control

Supports DeepSeek's thinking mode and reasoning_content. none disables thinking mode; low, high (default), and max set the thinking effort.

Inherits Every Copilot Capability

Because this plugs into Copilot's native provider API, you get the full stack for free:

• Agent mode — autonomous multi-step tasks
• Tool calling — file edits, terminal, workspace search, Git, tests
• Instructions & skills — all your .instructions.md, AGENTS.md, and skills just work
• Prompt caching stats — DeepSeek's cache hit rate logged in the output channel so you can see the savings

Web Search and Fetch

Two built-in tools plug Copilot Chat into the official DeepSeek Search and Fetch APIs:

• deepseek_web_search — ranked web results with title, URL, date, site name, summary, and more. Inline reference: #deepseekSearch.
• deepseek_web_fetch — fetches one page and returns clean Markdown. Inline reference: #deepseekFetch.

In agent mode the model calls them automatically when a task needs fresh web information. You can also add them via the chat tools picker, or mention #deepseekSearch / #deepseekFetch explicitly. Both APIs bill per successful request — see the official docs and the pricing page for current rates.

Notes:

• Each tool can be disabled independently in Settings (search for webTools; keys: webTools.search.enabled / webTools.fetch.enabled). Disabled tools are hidden from Copilot Chat and cannot be called.
• When deepseek-copilot.baseUrl is customized, the target service must implement POST /v1/tools/search and POST /v1/tools/fetch.
• These standalone tools work with every DeepSeek model in this extension. The Responses API's built-in web_search tool is only available on deepseek-v4-pro and requires switching the whole chat protocol to /responses, which this extension does not do.
• The tools always use your configured DeepSeek API key, even when another chat model is selected.

Secure by Default

API key lives in VS Code's SecretStorage (OS keychain on macOS / Windows / Linux). Never in settings.json, never in your Git history.

Zero Runtime Dependencies

Pure VS Code API + Node.js built-ins. No Python, no Docker, no local proxy server to babysit.

Getting Started

Prerequisites

• VS Code 1.116 or later. This extension relies on non-public Copilot Chat APIs that may break on newer VS Code versions — report an issue if you hit one.
• GitHub Copilot subscription (Free / Pro / Enterprise — the free tier works)
• DeepSeek API key from platform.deepseek.com, or a compatible provider token when using a custom deepseek-copilot.baseUrl

Installation

Install from the registry used by your editor:

1. Microsoft VS Code — install from VS Code Marketplace.
2. Editors that use Open VSX — install from Open VSX.

Usage

1. Run DeepSeek: Set API Key from the Command Palette (Cmd+Shift+P)
2. Paste your key or compatible provider token (official DeepSeek keys usually start with sk-)
3. Open Copilot Chat, click the model picker, and choose DeepSeek V4.1 Flash or DeepSeek V4 Pro
4. That's it — chat away

Models

Both entries support thinking mode, tool calling, and 1M token context. See the official Models & Pricing documentation for billing details.

Settings

Thinking Effort is configured from Copilot Chat's model picker for each DeepSeek model.

Example settings.json override for compatible API proxies:

{
  "deepseek-copilot.modelIdOverrides": {
    "deepseek-flash": "your-v4.1-flash-model-id",
    "deepseek-v4-pro": "your-pro-model-id"
  }
}

Known Issues

DeepSeek models are missing from Cloud sessions in the Agents window

The Agents window composer offers several session kinds. Cloud sessions (Copilot → Cloud) run in a GitHub-hosted sandbox and resolve their model list server-side, so the only entry is Auto — third-party BYOK models cannot be used there, and no extension setting changes that.

Pick the local Copilot session type instead: the model picker there lists deepseek/deepseek-flash and deepseek/deepseek-v4-pro alongside your other BYOK providers. If they are missing from the local session picker too, enable chat.agentHost.byokModels.enabled and fully restart VS Code, since the agent host caches the provider list per window.

{
  "chat.agentHost.byokModels.enabled": true
}

The extension additionally declares "capabilities": { "agentsWindow": { "supported": true } } so it may run inside the Agents window itself. On VS Code 1.143 / 1.144 that declaration is only honored when extensions.experimental.enableAgentsWindowCapability is enabled; use extensions.supportAgentsWindow if you want the extension's own commands available in that window. Neither setting affects whether the models appear — model availability comes from the BYOK bridge, not from loading the extension there.

Note on activation order: the extension registers its provider synchronously during activation and refreshes the model list after Copilot Chat finishes activating, so a slow Copilot startup does not leave the picker empty.

Models disappear from the Agents window once every VS Code window is closed

The agent host does not run this extension's provider by itself. With chat.agentHost.byokModels.enabled, it wires in BYOK providers from a running VS Code extension host. Close the last VS Code window and that extension host shuts down, so the DeepSeek models disappear from the Agents window picker and in-flight requests fail.

Workaround: keep at least one VS Code window open while working in the Agents window. Leaving it minimized is enough — the process has to exist, it does not have to be visible. Alternatively, run the same agent task from the Chat view of that window. This is a VS Code-side limitation that the extension cannot work around on its own.

Compared to alternatives

Acknowledgements

This project is a community fork of Vizards/deepseek-v4-for-copilot by Vizards. The provider implementation, the vision pipeline, the i18n scaffolding, and most of the architecture come from that project — this fork would not exist without it. The upstream work is MIT licensed, and its copyright notice is preserved in LICENSE and shipped with every build.

Thanks also to the upstream contributors and everyone who reported issues or sent pull requests there; the features in this fork build directly on that groundwork.

This fork adds DeepSeek web search and web fetch tools, and publishes independently so those changes stay available while they are under review upstream.

License

MIT