我将我的氛围编码技术栈的成本优化到了 $0。
自从 vibe 编码问世以来，我一直在尝试构建各种产品。我的一些产品面向消费者，而另一些则是昂贵软件的内部克隆。然而，从一开始我就知道一件重要的事情——vibe 技术栈的成本很高。
我最初尝试了很多工具——Bolt、v0、Replit、Lovable 等，其中 Replit 给我的结果最好（是的，我可能因为选择的应用而有偏见）。但我经常每月支付 25 到 200 美元的费用。其他费用，比如 API、模型等，使得每月账单超过 300 美元。与雇佣开发者相比，这样的成本是否划算？是的。但是否物有所值？不。
因此，经过几个月的优化，我将整个技术栈调整为内部使用时要么免费（或成本极低），要么在面向消费者的产品中保持较低的成本。
以下是目前整个技术栈的情况：
- IDE - 谷歌的 AntiGravity（100% 免费，如果使用学生证可以获得更高的访问权限）
- AI 文档 - SuperDocs（100% 免费且开源）
- 数据库 - Supabase（Nano 计划免费，足够满足基本需求）
- 认证 - Stack Auth（免费至 1 万用户）
- LLM（AI 模型） - OpenRouter 或 Gemini 通过 AI Studio 进行测试，以及 Unsloth AI 提供的定制调优模型用于生产（你可以在 Google Colab Notebook 中使用 Unsloth 进行模型微调）
- 版本维护/分发 - Github/Gitlab（两者均完全免费且开源）
- 更快的部署 - Vercel（免费套餐足够供爱好者使用）
- 分析 - PostHog、Microsoft Clarity 和 Google Analytics（这三者都是免费的，适用于不同的跟踪，我建议同时使用它们）
这就是开发者们的清单！我知道我可能遗漏了一些内容。如果有，请随时问我或在评论中列出。如果你有任何具体问题，也可以问我。
查看原文
Since vibe coding came into existence, I have been experimenting with building products a lot. Some of my products were consumer facing and some.. well, internal clones of expensive software. However, since beginning, I knew one big thing - the vibe stack was expensive.<p>I initially tried a lot of tools - Bolt, v0, Replit, Lovable, etc. out of which Replit game me the best results (yes, I can be biased due to my selection of applications). But I often paid anywhere from $25-$200/mo. Other costs like API, models, etc. made monthly bills upward of $300/mo. Was it cost effective when compared to hiring a developer? Yes. Was it value for money? NO.<p>So, over the months, I optimised by complete stack to be either free (or minimal cost) for internal use or stay at a much lean cost for consumer-facing products.<p>Here's how the whole stack looks today -<p>IDE - Google's AntiGravity (100% free + higher access if you use student ID)<br>
AI Documentation - SuperDocs (100% free & open source)<br>
Database - Supabase (Nano plan free, enough for basic needs)
Authentication - Stack Auth (Free upto 10K users)
LLM (AI Model) - OpenRouter or Gemini via AI Studio for testing and a custom tuned model by Unsloth AI for production. (You can fine-tune models using Unsloth literal in a Google Colab Notebook)
Version Maintenance/Distribution - Github/Gitlab (both totally free and open source)
Faster Deployment - Vercel (Free Tier Enough for Hobbyists)
Analytics - PostHog, Microsoft Clarity & Google Analytics (All 3 are free and independent for different tracking, I recommend using all of them)
That's the list devs! I know I might have missed something. If yes, just ask me up or list them up in the comments. If you have any questions related to something specific, ask me up as well.