Continue.dev

• 1. overview (490 chars)

  Continue.dev is a completely open source AI programming assistant that supports VS Code and JetBrains IDE and is regarded as the best open source alternative to GitHub Copilot. The project has more than 32,700 stars on GitHub and is currently one of the most popular open source AI programming tools among developers. Its core advantages are that it is completely free, has a high degree of freedom in model selection, supports local deployment, and has data security and controllability. It is especially suitable for teams and JetBrains users with high code security requirements.

• 2. Product background (713 chars)

  Continue.dev is developed by the Continuous AI team under the Apache 2.0 open source license, and the code is hosted on GitHub under the continuedev organization. The project has been continuously active since its launch, with more than 21,000 submissions and 822 released versions. The community is extremely active. The technology stack is mainly TypeScript (accounting for 84.4%), and also includes Kotlin, Python, Rust and other languages. The overall architecture is modern and modular.   As an "AI native development platform", Continue.dev not only provides basic code completion functions, but also supports the creation and use of custom AI code assistants. Its positioning is to allow developers to freely choose models, freely configure workflows, and get rid of the subscription binding and model lock-in of commercial products.

• 3. Product functions and usage experience (1611 chars)

  The core functions of Continue.dev can be summarized into three major modules: Tab completion, Chat dialogue and Agent agent.   Tab completion is the most frequently used feature on a daily basis, providing context-aware inline code suggestions in real time during coding. Compared with the completion of GitHub Copilot, the completion experience of Continue.dev is slightly inferior, but the advantage is that the model can be freely selected, and developers can configure models with different delays and costs according to budget and needs. A common combination is to use Qwen3.5-Flash or GLM-4.7-Flash for tab completion, which has extremely low latency and controllable cost.   The Chat conversation feature is one of the highlights of Continue.dev. The sidebar dialogue supports rich contextual reference syntax. Developers can reference specific files through "@Filename", analyze the entire directory through "@Folder", intelligently search for related codes through "@Codebase", and even analyze code changes through "@Git Diff" and analyze terminal output through "@Terminal". This contextual referencing system makes conversations more precise, allowing developers to focus questions directly on relevant snippets of code without having to paste code repeatedly into conversations.   The agent function is relatively conservative and can complete multi-step tasks independently, but each step requires confirmation by the developer. This design is more secure than aggressive agent tools such as Cline and is suitable for cautious developers.   In addition to basic functions, Continue.dev also provides auxiliary capabilities such as code specification checking, performance issue reminders, code explanations, and document generation. Its intelligent context system can understand the project code style and give suggestions that are more in line with the team's habits.

• 4. Pricing and business model (847 chars)

  Continue.dev adopts a completely open source and free strategy and does not charge any subscription fees. The only cost to the developer is the cost of the API calls, which depends entirely on the model chosen and frequency of usage.   Taking the OfoxAI API gateway as an example, the input price of Claude Opus 4.6 is US$5 per million tokens and the output is US$25 per million tokens; the input price of Claude Sonnet 4.6 is US$3 per million tokens and the output is US$15 per million tokens; DeepSeek V3.2 has a lower price and is suitable for developers with limited budgets. OfoxAI currently provides 81 models to choose from, and developers can flexibly match them according to the type of tasks.   Compared to GitHub Copilot’s $10 per month Pro version and $39 per month Pro+ version, Continue.dev’s zero-subscription model is very friendly to light users. Heavy users can even use it completely free if they configure a local model (such as DeepSeek-Coder 7B) and only need to bear the hardware cost of the GPU server.

• 5. User reviews (1337 chars)

  Judging from community feedback, user reviews of Continue.dev are generally positive.   Positive reviews mainly focus on several aspects. Open source transparency is the most frequently mentioned advantage. Developers can view the source code, modify functions by themselves, and ensure code security, which is especially important for enterprise users. The degree of model freedom is also well received. It supports multiple models such as OpenAI GPT, Google Gemini, DeepSeek, etc., and developers can flexibly choose according to their budget and needs. Local deployment capability is another highlight. With Ollama, open source models can be run locally without data being uploaded at all, meeting the needs of privacy-sensitive teams. JetBrains support is also a necessity for JetBrains users. VS Code tools such as Cursor and Windsurf do not support JetBrains. Continue.dev is almost the only reliable choice.   Negative feedback mainly focuses on the usage threshold and experience details. Local deployment requires certain technical capabilities and configuration, and also requires GPU server support to obtain better performance, which is not friendly enough to individual developers. The Tab completion experience is not as smooth as Cursor, and the completion accuracy and context understanding are slightly inferior. The LLM hallucination problem is also a common problem in all AI programming tools, and Continue.dev is no exception. It may fabricate non-existent code bases or invent non-existent syntax, which requires developers to verify themselves.

• 6. Industry evaluation (1034 chars)

  Based on industry press and community discussions, Continue.dev is widely viewed as the best open source alternative to GitHub Copilot.   In a comparative evaluation in 2026, the OfoxAI blog listed Continue.dev as one of the seven mainstream AI programming tools, commenting on its "zero subscription fee, high model freedom, and JetBrains support as its unique advantages." The article points out that Continue.dev is the preferred solution for JetBrains users, developers with a zero-cost budget, and teams seeking not to be locked in by vendors.   The in-depth evaluation article of the CSDN blog believes that Continue.dev is not a simple code prompt tool, but an intelligent assistant that can truly understand the project and code style. The article particularly recommends the "Continue.dev + Ollama" combination, believing that it takes into account both security and efficiency.   In discussions on Zhihu and Jiji communities, developers generally believe that Continue.dev’s customizability is its biggest advantage and is suitable for teams with high code security requirements. However, some users pointed out that Cursor is still a better choice for developers who pursue the ultimate completion experience.

• 7. Controversies and Risks (995 chars)

  The main risks with Continue.dev are usage thresholds and LLM illusion issues.   Local deployment requires configuring Ollama, downloading models, and setting up remote access, which requires certain technical capabilities. If the team does not have a GPU server, the inference speed of local deployment will be very slow, affecting the user experience. For individual developers or small teams, this threshold may discourage some users.   The LLM illusion is a common problem with all AI programming tools, and Continue.dev is no exception. Models may make up code bases that don't exist, invent syntax that doesn't exist, and give advice that seems reasonable but is actually wrong. Developers need to have certain code review capabilities and cannot blindly trust the output of AI.   Another potential risk is the complexity of model selection. Continue.dev supports a variety of models, but the capabilities and costs of different models vary greatly, and developers need to spend time learning and testing to find the configuration that best suits them. For developers who don’t want to do the fuss, the out-of-the-box experience of GitHub Copilot or Cursor may be more suitable.

• 8. Applicable groups and usage suggestions (1249 chars)

  Continue.dev is suitable for the following types of developers: the main editor is a developer of JetBrains, and Continue.dev is almost the only open source AI programming assistant that supports JetBrains; teams with high requirements for code security, open source and transparent, supports local deployment, and fully controllable data; individual developers or small teams with limited budgets, zero subscription fees, and controllable model costs; developers who pursue not being locked by manufacturers can freely switch models and API gateways.   Continue.dev is not suitable for the following types of developers: developers who pursue the ultimate tab completion experience, the completion feel and accuracy of Cursor are higher; developers who do not want to fiddle with configuration, the out-of-box experience of GitHub Copilot or Cursor is more friendly; developers who do not have a GPU server and have an extremely low budget, the local deployment experience will be poor.   In terms of usage recommendations, for daily chat and code explanation, it is recommended to configure Claude Sonnet 4.6, as the code understanding is accurate and fast. For complex refactoring and multi-file context processing, Claude Opus 4.6 is recommended. For tab completion, it is recommended to configure Qwen3.5-Flash or GLM-4.7-Flash, which has low latency and low cost. If you want to be completely free, you can configure GLM-4.7-Flash, which is completely free on OfoxAI.

• 9. Summarize (432 chars)

  Continue.dev is an open source AI programming assistant with clear positioning and outstanding advantages. Its core value is that it is open source and free, has free models, supports local deployment, and has data security and controllability. It is especially suitable for JetBrains users and teams with high code security requirements. For developers who don’t want to be tied to a commercial product subscription and pursue flexible configuration, Continue.dev is the best alternative to GitHub Copilot.

No in-depth report yet

• 

  JFisherSr

  —

  绝对的开源良心之作，不接受反驳。

---

• 

  greencat185

  —

  用 Ollama 跑本地模型的话，大家推荐哪个模型做代码补全？Qwen2.5-Coder 还是 DeepSeek Coder？我 8GB 显存跑不动太大的。

---

• 

  Russell_AndersonX88

  —

  数据隐私这块，Continue 是目前唯一的真答案。

---

• 

  CCarter_2024

  —

  手贱升级了一次配置，结果崩了，翻了一遍文档才找到问题。这工具对新手真不友好，出了问题自己排查还挺费时间的。

---

• 

  Edward.LewisK

  —

  建议新手一开始只配一个 Cloud 模型（Claude Sonnet 或 GPT-4o），等用熟了再加本地补全。一口吃成胖子容易踩坑。

---

• 

  EthanSanders_Pro806

  —

  大家觉得 Continue.dev 的 PR agent 和 GitHub Actions 自建代码检查比，哪个更香？感觉 Continue 的优势是自然语言写规则不用学 DSL。

---

• 

  AMendoza_X

  —

  在医疗科技公司搞开发，HIPAA 合规要求所有病人数据处理不能出内部网络。Continue + Ollama 本地跑 Llama 3.3 是唯一能通过安全审查的 AI 编程方案。零代码外传，零月费。

---

• 

  黄艳琪

  —

  新版本什么时候发布？社区 fork 有推荐的吗？

---

• 

  smallleopard172

  —

  Continue.dev 的 CI agent 是真创新，其他工具都没有。我们 CI 流程里跑了几个检查规则：安全漏洞扫描、性能 warning 检测、代码风格一致性。agent 并行跑在 PR 上，几分钟出结果，比以前人工 review 快多了。

---

• 

  KRamirez

  —

  API 直接走自己的 key，没有被中间商赚差价的感觉真好。

---

• 

  MiningMoleCook

  —

  工作流里我配了三个模型：Claude Opus 做架构决策、Sonnet 写日常代码、DeepSeek 做高容量 code review。这种分层路由的思路在 Cursor 和 Copilot 上都不可能实现。一个月 API 费用加起来也就 30-40 美元。

---

• 

  Joe_Bailey_2024

  —

  Cursor 收购了 Continue 团队后，这个项目现在基本就是社区维护了。虽然代码是 Apache 2.0 开源可以 fork，但感觉未来的更新会越来越慢。新用户入坑有风险。

---

• 

  TheSusanWallace_pro

  —

  在 Cursor 和 Continue 之间纠结了很久。一个需要换编辑器，一个需要自己配置。最后还是选了 Continue，因为我用 IntelliJ 习惯了不想换。

---

• 

  ETr_od

  —

  之前一直用 GitHub Copilot，换成 Continue + Ollama 跑 Qwen2.5-Coder 后，补全延迟从云端变成本地，响应快多了。虽然准确率还是比 Copilot 差一点，但胜在免费。

---

• 

  Philip.Rogers_2024624

  —

  Mac M2 跑 Ollama 7B 模型会不会发热很严重？

---

• 

  EmmaMartinez5205

  —

  如果想远程用公司内网的模型，可以用 SSH 转发或者在公司搭一个 OpenAI 兼容的代理端点。Continue 支持任何兼容 OpenAI API 的端点，一行 config 切换。

---

• 

  Nathan_Walker520

  —

  用 Continue.dev 的 Agent 模式做过一次跨三个文件的重构，从分析依赖到修改代码再到跑测试，一口气完成。虽然没有 Cursor Composer 那么丝滑，但省了我至少半小时的手动操作。

---

• 

  杨明

  —

  补全是真的慢，尤其是用云端模型做 autocomplete。延迟 1.4 秒才出建议，这时候我字都打完了，建议才弹出来，完全失去意义。只能用小模型做本机补全才行。

---

• 

  EllieGreen

  —

  JetBrains 用户终于有像样的 AI 辅助了，泪目。

---

• 

  Benjamin_HallX266

  —

  看到 Continue 的 Hub 服务要关停了，官方把重心放到了 CI agent 上。IDE 插件还会免费维护但不再主推。有点遗憾，但 CI agent 确实是个差异化方向。

---

• 

  JUand

  —

  我们把团队代码规范写成了 .continue/checks/ 下的 markdown 文件，每次 PR 自动用 Continue agent 检查。谁又在代码里留了 console.log 立马就被抓到，还直接给出修复建议的 diff，太爽了。

---

• 

  AngelaBakerQ

  —

  想问问各位，Continue 的 PR agent 检查你们的代码规则都是怎么写的？我们团队刚起步，想抄抄作业。

---

• 

  Charles_Ramirez_88

  —

  config.json 的配置太繁琐了，35 行的 JSON 要配 endpoint 地址、API key、模型名、上下文窗口……新手看了直接头皮发麻。模板是有，但还是得自己搞明白参数含义。

---

• 

  Jennifer.GonzalesK

  —

  配置完了之后是真的香，值得花那半小时。

---

• 

  Frank_Hall_66

  —

  问一下，JetBrains 上的 Continue 和 VS Code 版功能一样齐全吗？有没有什么阉割？

---

• 

  Jessica_Castillo520

  —

  Agent 模式偶尔会翻车，尤其是多步任务做到一半卡住了，而且不像 Cursor 那样能看到推理链条。出问题的时候 debug 起来很痛苦，只能手动重来。

---

• 

  Heather_Ramos_2024

  —

  创建自定义 slash command 是 Continue 被低估的功能。我在 /review 里配了安全审查的 prompt，/test 配了单元测试生成，从此团队有了统一的 AI 工作规范。

---

• 

  Diana.Bennett_99057

  —

  在 JetBrains 全家桶上试了 Continue.dev，IntelliJ 和 GoLand 都能用，配置好 Claude API 后侧边栏直接对话，补全走的本地 Qwen2.5-Coder 7B。对于我这种日常写 Java 和 Go 的，体验比预期好不少。

---

• 

  Tober

  —

  我是 PyCharm 用户，之前看到 Cursor 那些 AI 功能只能干瞪眼。Continue 装上 JetBrains 插件后，Chat、Edit、补全都用上了，体验意外的流畅。终于不用因为 AI 工具换编辑器了。

---

• 

  Nicholas_Williams_Pro

  —

  有没有人试过 Continue + Cline 一起用的？我用 Continue 做日常补全和代码查询，Cline 做自主 implementation，两个互补发现效率确实很高。

No reviews yet