zmy15/deepseek-for-visualstudio

Overview

DeepSeek for Visual Studio brings AI chat, code editing, solution-aware tools, terminal execution, and multimodal understanding directly into the IDE. It is designed for Visual Studio 2022 (17.14 or later) and Visual Studio 2026. Beyond the official DeepSeek API, it also supports any OpenAI-compatible chat/completions endpoint configured with a custom Base URL.

The extension combines a native-grade WebView2 chat experience with five cooperating agents, reusable Skills, MCP tool servers, Ghost Text completion, and persistent project memory.

⭐If this project helps you, please consider giving it a Star — it helps more developers discover it.

Screenshots

Inline AI edit — Enter to apply / Esc to cancel

Project Highlights

Core Features

• Chat and reasoning: streaming responses, Deep Reasoning, Pro/Flash models, resumable streaming, and Prefix Cache.
• OpenAI-compatible endpoints: connect any chat/completions-compatible service through a custom Base URL.
• Code editing: inline AI edit, exact/multi replacement, apply_patch, file creation, diff preview, and Ghost Text completion.
• Project tools: solution exploration, file parsing, terminal commands, Git operations, and build verification.
• Skills system: reusable workflows defined in SKILL.md, triggered by /skillname or semantic matching.
• MCP integration: connect HTTP and stdio tool servers and classify their tools for agents.
• Multimodal AI: image understanding, window screenshots, page-by-page PDF reading, web image reading, and OCR.
• Search: Baidu Qianfan and DuckDuckGo with automatic fallback.
• Memory: persistent user, session, and repository notes injected into new conversations.
• Internationalization: automatic Chinese/English UI switching with custom translation overrides.

Requirements

Installation

Download the latest .vsix from Releases, close Visual Studio, and install it.

• Full package: x64 only, includes local OCR.
• No-Local-OCR package: x64 and ARM64; local OCR is removed and other features remain available.

To build from source, install the .NET Framework 4.7.2 SDK and the Visual Studio extension development workload:

git clone https://github.com/zmy15/DeepSeek-for-VisualStudio.git

Open .slnx in Visual Studio, build the solution, and press F5 to launch the experimental instance.

Quick Start

1. Get an API key from platform.deepseek.com/api_keys.
2. Open Tools > Options > DeepSeek Chat, enter the API key, and select a model.
3. Open View > Other Windows > DeepSeek Chat.
4. Ask a question, select code and press Ctrl+I for inline editing, or dispatch an agent command.

Recommended starting settings:

Agents and Skills

Use @ask, @plan, @edit, or @build to select an agent explicitly, or let Handoff route the task automatically. Skills are loaded from project-level directories, user-level directories, and built-in skills.

TODO

Contributing

Branches and workflow

• The default branch is master; day-to-day development happens on dev, so please do not push to master directly.
• Branch off the latest dev using a feature/xxx, fix/xxx, or docs/xxx name, then open a Pull Request against dev.
• dev is merged into master by the maintainers at release time only.

Before opening a Pull Request

1. Open an Issue first: every change should start with an Issue describing the background, expected behavior, and rough implementation approach. Wait for the maintainers to confirm the direction before you start coding. The PR description must link the Issue with Fixes #123 or Closes #123; PRs without a corresponding Issue will be asked to file one first.
2. Sync before you submit: rebase or merge onto the latest dev to avoid conflicts and meaningless merge commits.
3. Build cleanly: build the whole solution in the Release configuration and make sure you add no new warnings or errors.
4. Pass the tests: run dotnet test DeepSeek_v4_for_VisualStudio.Tests\DeepSeek_v4_for_VisualStudio.Tests.csproj --configuration Release. Add tests for new logic and do not lower existing coverage.
5. Respect the i18n constraint: user-facing text must go through localization resources; do not add hardcoded Chinese string literals in Services/ or View/.
6. Keep the change focused: one PR solves one problem — do not mix in unrelated formatting, renames, or dependency upgrades.

What a PR should contain

• Title: follow Conventional Commits, e.g. feat(chat): support session export or fix(edit): fix out-of-range patch matching.
• Description: explain the background and motivation, the concrete changes, the blast radius (whether settings, shortcuts, SKILL.md, or MCP behavior are affected), and how you verified it.
• Linked Issue: reference the Issue with Fixes #123 or Closes #123.
• UI changes: attach a screenshot or screen recording so reviewers can confirm the visual result.
• Compatibility notes: call out anything involving localization, the official DeepSeek API, compatible endpoints, or other platforms.
• Do not commit: API keys, tokens, internal paths, proprietary code, or build output such as bin/, obj/, and TestResults/.

CI and review

Pull Requests automatically trigger the Build & Test workflow (build, unit tests with an uploaded coverage report, and the hardcoded-CJK string check), and all checks must pass before merging. Address review feedback with follow-up commits rather than force-pushing over existing discussion.

Support

• Usage questions and discussions: Discussions
• Bug reports and feature requests: Issues

Contributors

Star History

License

MIT © 2026 zmy15