Skip to content

Trae

ByteDance

ByteDance 推出的 AI 原生 IDE——国内首款——深度集成 Doubao、DeepSeek 等 LLM,支持自然语言交互和多模态协作。

Official Site
01

概述

Trae 是 ByteDance 于 2025 年 1 月发布的 AI 原生集成开发环境,也是国内首款 AI 原生 IDE。它深度集成 Doubao、DeepSeek 等 LLM,通过自然语言交互、智能代码生成和多模态协作重塑开发范式。截至 2025 年底注册用户已突破 600 万,在中国开发者中获得了极高的认可度。2026 年 Q2 它进一步完善了 SOLO 模式和设计转代码功能,并全面兼容 MCP 协议 v1.1。

核心特性:AI 原生 IDE、多模型内置、SOLO 自主模式、设计转代码、MCP 工具市场。

02

安装

Trae 以独立安装包形式发布——从官网下载对应操作系统的安装包。国内用户可访问 trae.cn;海外用户可访问 trae.ai。提供 Windows 和 macOS(Intel 与 Apple Silicon 均支持)安装包。

bash
# Download links
# China site: https://www.trae.cn/
# International site: https://www.trae.ai/

# Windows: download the .exe installer and double-click to run
# macOS:   download the .dmg and drag it into Applications
# Launch Trae and sign in to start using it

Trae 基于 VS Code 内核构建,因此可以直接导入你现有的 VS Code 配置和扩展。

03

基本用法

启动 Trae 后,在 Chat 面板用自然语言描述需求,它会自动生成并编辑代码。在 SOLO 模式下它可以自主处理从需求到实现的完整流程。内置的 Builder 模式支持上传设计文件直接生成前端代码。你也可以接入 Claude API 等第三方模型以获得更强能力。

bash
# Enter a request in the Chat panel (example)
"Create a paginated user list page using React + Tailwind"

# Enable SOLO autonomous mode
# Switch to SOLO in the sidebar, describe the full requirement, and let it plan and build autonomously

# Connect third-party models (configure API keys in settings)
# Settings → Models → Add Claude / OpenAI API key

Trae 国内版内置 Doubao/DeepSeek 模型并提供慷慨的免费额度;国际版接入 Claude、GPT 等模型。

04

技巧

Trae 的 MCP 市场支持一键安装、零配置激活各类 MCP 工具,是构建 AI Agent 的利器。使用 #filename 引用具体文件、@function 定位代码片段,让 AI 更精准地理解上下文。设计转代码功能对前端开发者尤其友好——上传 Figma 文件或图片即可获得可运行的代码。

bash
# Reference files and symbols in a conversation
"Refactor the @login function in #src/utils/auth.ts and add error retry"

# One-click install of MCP tools
# Sidebar → MCP marketplace → search and install (e.g. database, browser, filesystem tools)

# Design to code
# Drag in a design image → Builder mode → auto-generate the corresponding component

提示:对于复杂项目,先让 AI 阅读项目结构(@project 根目录)再下达指令。

05

Configuration

Trae is configured through its visual Settings panel. The China edition ships with Doubao/DeepSeek models and a free quota, while the international edition connects to Claude, GPT, and others via API keys added under Settings -> Models. SOLO mode, the MCP marketplace, and design-to-code are toggled in the sidebar. Because Trae is built on the VS Code core, your VS Code settings.json and keybindings carry over.

bash
# Connect third-party models
# Settings -> Models -> Add API key (Anthropic / OpenAI / etc.)

# Enable SOLO autonomous mode
# Sidebar -> switch to SOLO -> describe the full requirement

# One-click MCP tools
# Sidebar -> MCP marketplace -> search and install
#   (database, browser, filesystem tools, zero config)

# Reference files/symbols in chat
"Refactor the @login function in #src/utils/auth.ts"

# Edit settings.json (VS Code-style, Cmd/Ctrl+,)
{
  "trae.solo.enabled": true,
  "trae.model.default": "doubao-pro"
}

The China edition's free Doubao/DeepSeek quota is generous; switch to Claude/GPT via API keys for heavier reasoning tasks.

06

FAQ

Common questions cover free quotas, third-party model support, data privacy, MCP usage, and design-to-code. Trae is free to download; the China edition includes a free model quota while the international edition uses BYOK (bring your own key) for Claude/GPT. MCP tools install with one click from the marketplace.

bash
Q: Is Trae free?
A: The IDE is free to download. The China edition includes a free
   Doubao/DeepSeek quota; the international edition uses your own API keys.

Q: Can I use Claude or GPT?
A: Yes—add the provider's API key under Settings -> Models.

Q: Does my code get sent to third parties?
A: When you use a cloud model, requests go to that provider's API under
   their policies. Check each provider's privacy terms.

Q: How do I install MCP tools?
A: Open the MCP marketplace in the sidebar and install with one click;
   no manual config is needed.

Q: How does design-to-code work?
A: Drag a Figma file or image into Builder mode and Trae generates
   runnable frontend code from the design.

For complex projects, let the AI read the project structure first before issuing instructions.

Ready to try Trae?

Visit the official site for the latest version and full documentation.

Visit Trae