What Is Continue?Continue 是什么？
Continue is the leading open-source AI coding assistant that lives inside your editor: extensions for VS Code and JetBrains (plus the cn CLI for terminal work), with four modes — Agent, Chat, Edit, and Autocomplete — covering everything from inline suggestions to full multi-step refactors. Released under Apache 2.0, it collected roughly 36k+ GitHub stars as the go-to "bring your own model" alternative to GitHub Copilot.Continue 是住在编辑器里的开源 AI 编程助手头牌：VS Code 和 JetBrains 插件（再加终端用的 cn CLI），四种模式——Agent、Chat、Edit、Autocomplete，从行内补全到全流程重构全包。Apache 2.0 开源，靠「模型自己选」这条路线攒下约 3.6 万+ GitHub star，是 GitHub Copilot 最知名的开源替代。
Its core idea is configurability: a single config.yaml assigns models to roles — chat, edit, apply, autocomplete, embed, rerank — with custom prompts, rules, MCP tool connections, and context sources. You can point it at OpenAI, Anthropic, Mistral, local models via Ollama, or anything in between, and tune the assistant to your team's exact stack.它的核心思想是可配置：一个 config.yaml 把模型分派到不同角色——chat、edit、apply、autocomplete、embed、rerank，再加上自定义提示词、规则、MCP 工具连接和上下文源。可以接 OpenAI、Anthropic、Mistral、Ollama 本地模型，把助手调成完全贴合你团队技术栈的样子。
Then came the twist: in 2026 Continue was acquired by Cursor — the paid editor it once competed with. The continuedev/continue repository is now read-only for all users and no longer actively maintained. The team shipped a final 2.0.0 release (June 19, 2026) that removed anonymous telemetry and authentication and fixed bugs, and hopes the codebase "continues to serve as a foundation for others." Confirm the current state on continue.dev and GitHub before adopting.然后故事出现了转折：2026 年 Continue 被 Cursor 收购——那个它曾经要挑战的付费编辑器。continuedev/continue 仓库已对所有用户只读，不再积极维护。团队发布了最终版 2.0.0（2026-06-19），移除了匿名遥测和鉴权、修了 bug，希望代码库「继续成为他人的地基」。采用前请先到 continue.dev 和 GitHub 确认现状。价格以官网为准。
Continue's acquisition is the ironic ending the AI coding wars deserved: the most popular open-source "own your assistant" project became a subsidiary of the paid editor it was built to compete with. Don't read it as a failure — Continue proved that developers will do real configuration work to control their AI stack, and that idea now lives on everywhere (config.yaml for model roles was genuinely ahead of its time). The practical question is what to do if you depend on it. Because the code is Apache 2.0 and the final 2.0.0 release is clean (no telemetry, no auth), Continue is unusually fork-friendly for a frozen project — you can pin it and run it for years. But a read-only repo means zero upstream fixes, so treat it as infrastructure with an owner: either your team owns the fork, or you plan the migration to an actively maintained tool (Cline, Aider, or yes, Cursor). The worst option is doing neither and waking up to a broken extension after the next VS Code update.
Continue 的收购结局带着一丝讽刺，正好配得上这场 AI 编程大战：最受欢迎的「自己掌控助手」开源项目，成了它当年要挑战的付费编辑器的子公司。但这不算失败——Continue 证明了开发者愿意为掌控自己的 AI 栈做真正的配置工作，这个思想现在到处都是（按角色分派模型的 config.yaml 在当时是真的超前）。实际问题是：如果你已经依赖它，该怎么办？因为代码是 Apache 2.0、最终版 2.0.0 又很干净（无遥测、无鉴权），Continue 是冻结项目里极罕见的「好 fork 型」——锁住版本能跑好多年。但只读仓库意味着上游零修复，所以要把它当需要负责人的基础设施：要么你的团队拥有这个 fork，要么规划迁移到仍在维护的工具（Cline、Aider，或者没错，Cursor）。最糟的是两头不做——下次 VS Code 一更新，醒来发现插件挂了。
— 中国吧 AI Tools Hub Editorial Team
Who Should Use Continue?谁适合使用Continue？
✓ Good Fit For适合以下场景
- Privacy-sensitive teams that want local models (Ollama) or self-hosted model routing inside the IDE需要本地模型（Ollama）或自托管模型路由的隐私敏感团队
- Developers who want one config file to tune models, prompts, and MCP tools to their stack想用一个配置文件把模型、提示词、MCP 工具调到贴合技术栈的开发者
- Copilot refugees who want an open-source IDE assistant without vendor lock-in想摆脱 Copilot 绑定、要开源 IDE 助手的开发者
✕ Not Ideal For不适合以下场景
- Anyone who needs upstream support — the repo is read-only and issues/PRs go nowhere需要上游支持的人——仓库只读，issue/PR 没人处理
- Beginners who want plug-and-play — Continue requires configuring your own API keys想要开箱即用的新手——Continue 需要自己配 API key
Getting Started with ContinueContinue 快速开始
1. Install the extension from the VS Code Marketplace (or Open VSX), or the CLI via npm从 VS Code Marketplace（或 Open VSX）安装插件，或用 npm 装 CLI $ npm install -g @continuedev/cli # terminal assistant: cn2. Open config.yaml and assign models to roles (chat/edit/apply/autocomplete/embed/rerank)打开 config.yaml，给各角色（chat/edit/apply/autocomplete/embed/rerank）分派模型3. Try Agent mode for multi-step tasks, or cn -p "prompt" for one headless turn in CI多步骤任务用 Agent 模式；CI 里用 cn -p "prompt" 跑单轮无头模式Key Features核心功能
-
Inside Your Editor住在编辑器里 — Extensions for VS Code and all major JetBrains IDEs, plus the cn CLI for terminal work — no context-switching to a browser tool.VS Code 与主流 JetBrains IDE 全套插件，再加终端用的 cn CLI——不用切到浏览器工具。
-
Four Modes, One Config四种模式，一个配置 — Agent, Chat, Edit, and Autocomplete cover inline suggestions to full multi-step coding tasks, all wired through a single config.yaml.Agent、Chat、Edit、Autocomplete 四种模式，从行内补全到全流程编码任务，全走一个 config.yaml。
-
Any Model, Any Role模型任选，角色分派 — Route chat, edit, apply, autocomplete, embed, and rerank to different models — OpenAI, Anthropic, Mistral, or fully local via Ollama.把 chat、edit、apply、autocomplete、embed、rerank 分派给不同模型——OpenAI、Anthropic、Mistral，或 Ollama 全本地。
-
Privacy-First Design隐私优先设计 — Self-host under Apache 2.0, run models locally — code never has to leave your machine or your firewall.Apache 2.0 自托管、模型本地跑——代码不用出你的机器、不用出防火墙。
-
MCP & Custom ContextMCP 与自定义上下文 — Custom context providers, rules, and MCP tool connections let teams build an assistant tuned to their exact codebase and stack.自定义上下文源、规则和 MCP 工具连接，让团队调出完全贴合自家代码库和技术栈的助手。
Pros & Cons优缺点
✓ Good Fit For适合以下场景
- Completely free and open source (Apache 2.0) — ~36k+ stars of battle-tested community完全免费开源（Apache 2.0），3.6 万+ star 的社区打磨
- Best-in-class model configurability — different models for chat, edit, autocomplete, and more模型可配置性顶级——chat、edit、补全各用不同模型
- Strong privacy story: local models via Ollama, self-hosted, no forced cloud隐私叙事强：Ollama 本地模型、自托管，不绑云
- Final 2.0.0 is clean — no anonymous telemetry, no authentication — unusually fork-friendly最终版 2.0.0 很干净——无匿名遥测、无鉴权，罕见的「好 fork」
✕ Not Ideal For不适合以下场景
- No longer actively maintained — repo is read-only since the Cursor acquisition (2026)不再积极维护——2026 年被 Cursor 收购后仓库只读
- No upstream fixes: new models, editor API changes, and dependency updates are on you无上游修复：新模型、编辑器 API 变化、依赖更新都得自己扛
- Requires configuring your own API keys — not plug-and-play for beginners需要自己配 API key——新手开箱即用体验差
Use Cases应用场景
Popular ways people use Continue.人们使用Continue的常见场景：
⌨️ IDE Chat & Autocomplete编辑器内聊天与补全
The classic use: inline code suggestions and a chat sidebar that sees your codebase, without leaving VS Code or JetBrains.最经典的用法：行内代码补全 + 能看到整个代码库的聊天侧边栏，全程不离开 VS Code 或 JetBrains。
🤖 Multi-Step Agent Tasks多步骤 Agent 任务
Agent mode handles full refactors and feature work across files — autonomous enough for real tasks, still inside your editor.Agent 模式处理跨文件的全流程重构和功能开发——够自主做正事，还待在你的编辑器里。
🔒 Local-Model / Air-Gapped Teams本地模型/离线团队
Route every role to Ollama-hosted models: proprietary codebases get AI help with zero data leaving the building.所有角色都走 Ollama 本地模型：专有代码库也能用 AI，一行数据不出大楼。
🏭 CI-Integrated AI ChecksCI 集成的 AI 检查
Run cn -p "your prompt" headless in scripts and CI pipelines for automated code review and checks.在脚本和 CI 里用 cn -p "prompt" 无头跑，做自动化代码审查和检查。
Similar AI Tools相似 AI 工具
If Continue doesn't fit your needs, here are other popular AI tools you might consider:如果Continue不符合你的需求，可以看看这些热门 AI 工具：