Continue - 安装与使用指南 | AI 智能体工具

概述

Continue 是一款完全开源的 AI 编码插件，同时支持 VS Code 和整个 JetBrains 系列（IntelliJ、PyCharm、WebStorm 等）。其核心理念是"模型自由"——你可以接入 OpenAI、Anthropic、本地 Ollama 模型，甚至自托管的开源模型，构建完全私有可控的 AI 编码助手。它支持代码补全、聊天、代码编辑、@ 上下文引用等完整能力，是企业内网和数据敏感场景的首选。

核心特性：完全开源、模型自由、VS Code + JetBrains 双支持、可本地部署。

安装

Continue 从 IDE 扩展市场安装。VS Code 和 JetBrains 的安装步骤略有不同；安装后需要在配置文件中设置模型。

bash

# === VS Code install ===
1. Open VS Code → Extensions panel (Cmd/Ctrl+Shift+X)
2. Search for "Continue" → install
# Or via command line:
code --install-extension Continue.continue

# === JetBrains install ===
1. Open the IDE → Settings → Plugins → Marketplace
2. Search for "Continue" → Install → restart

# === Configure models ===
# Config file location: ~/.continue/config.json (the .continue folder in your home directory)

安装后侧边栏会出现 Continue 图标；首次打开会引导你配置第一个模型。

基本用法

Continue 通过 config.json 配置。下面是一个接入 Claude 和本地 Ollama 模型的示例。配置完成后，用 Cmd/Ctrl+L 打开 Chat，Cmd/Ctrl+I 进行内联编辑，Tab 接受补全。

bash

# ~/.continue/config.json example
{
  "models": [
    {
      "title": "Claude Sonnet",
      "provider": "anthropic",
      "model": "claude-sonnet-4-5",
      "apiKey": "your ANTHROPIC_API_KEY"
    },
    {
      "title": "Local Llama",
      "provider": "ollama",
      "model": "llama3"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Local Llama",
    "provider": "ollama",
    "model": "deepseek-coder"
  }
}

# Common shortcuts
Cmd/Ctrl + L    # Open Chat
Cmd/Ctrl + I    # Inline-edit the selected code
Tab             # Accept completion

本地模型需要先安装 Ollama（ollama.com）并拉取模型：ollama pull deepseek-coder

技巧

Continue 的最大价值在于数据隐私和成本控制：用本地 Ollama 模型做补全（免费、离线、代码不外泄），用云端 LLM 做复杂对话。@codebase 让 AI 搜索整个代码库；@docs 引用官方文档；自定义斜杠命令可将常用工作流固化。企业可自托管模型代理统一管理。

bash

# Context references
"@codebase Where is the entry point of this project?"
"@docs Latest usage of React useState"
"@terminal How do I fix the error just now?"

# Custom slash commands (~/.continue/config.json)
"customCommands": [{
  "name": "test",
  "description": "Write tests for the selected code",
  "prompt": "Write unit tests for the following code: {{{input}}}"
}]

# Use a local model for completion + a cloud model for chat
# tabAutocompleteModel uses ollama, models uses claude

提示：本地补全模型推荐 deepseek-coder 或 starcoder2——体积小、速度快、质量好。

Configuration

Continue is configured through ~/.continue/config.json. The models array defines chat models (provider, model, apiKey), and tabAutocompleteModel sets the completion model. Supported providers include OpenAI, Anthropic, Ollama, and self-hosted endpoints. Custom slash commands codify reusable workflows. Local Ollama models give free, offline completion while a cloud LLM handles complex chat.

bash

# ~/.continue/config.json
{
  "models": [
    {
      "title": "Claude Sonnet",
      "provider": "anthropic",
      "model": "claude-sonnet-4-5",
      "apiKey": "your ANTHROPIC_API_KEY"
    },
    {
      "title": "Local Llama",
      "provider": "ollama",
      "model": "llama3"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Local Coder",
    "provider": "ollama",
    "model": "deepseek-coder"
  },
  "customCommands": [{
    "name": "test",
    "description": "Write tests for the selected code",
    "prompt": "Write unit tests for: {{{input}}}"
  }]
}

# Local model (free, offline completion)
ollama pull deepseek-coder

Use a local Ollama model for completion (free, offline, no code leakage) and a cloud LLM for complex chat.

FAQ

Common questions cover data privacy, local models, provider support, cost, and @ context references. With a local Ollama model no code leaves your machine. Continue supports OpenAI, Anthropic, Google, Ollama, and self-hosted endpoints. @codebase searches the whole project and @docs references official docs.

bash

Q: Is my code sent anywhere?
A: Only if you configure a cloud provider. With a local Ollama model,
   completion runs entirely on your machine—no code leaves.

Q: Which providers are supported?
A: OpenAI, Anthropic, Google, Ollama, LM Studio, and any OpenAI-compatible
   endpoint. Configure them in ~/.continue/config.json.

Q: How do I cut costs?
A: Use a local Ollama model for completion (free) and a cloud model only
   for complex chat tasks.

Q: What do @ references do?
A: @codebase searches the whole project, @docs references official docs,
   @file adds a specific file, @terminal references terminal output.

Q: Does it work in JetBrains IDEs?
A: Yes—Continue supports VS Code and the entire JetBrains suite
   (IntelliJ, PyCharm, WebStorm, etc.).

For local completion models, deepseek-coder or starcoder2 are recommended—small, fast, and good quality.

Ready to try Continue?

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

Visit Continue

All AI guides