展示HN:Toolbase – 通过示例而非指令构建可靠的AI团队成员

1作者: David1238大约 17 小时前原帖
大家好,我们是David和Ethan,Toolbase(gettoolbase.com)的联合创始人。 Toolbase是一个AI代理和工作流构建工具,帮助您快速创建生产级的AI自动化——电池已包含。 ------- Toolbase的开发流程: 1. 定义一个粗略目标。 2. 连接任何API或MCP服务器(我们有成千上万的选择,或者您可以自带)。 3. 教AI什么是有效的输入和输出示例(这些示例后来将成为您的单元测试!)。 4. 让Toolbase生成完美的提示、代码、工作流或代理。 5. 将您的项目部署为API、MCP服务器或聊天界面! 编码可选,分享鼓励。 注意:如果您熟悉Cursor/Windsurf以及像Mastra这样的框架(Toolbase所基于的),您已经知道如何使用Toolbase。您保留了编码的灵活性,但可以避免样板代码和基础设施任务(集成、验证、上下文映射、测试等),除非您明确选择去做这些。 ------- 演示 这个简单的代理通过Tavily搜索验证我们CRM(Pipedrive)中的公司账单地址。如果地址不匹配,它会通过电子邮件请求人工选择正确的地址(请以2倍速观看): [观看演示](https://www.loom.com/share/540c61b2c5634996b088ebbb16989cf0?sid=8d076bd5-c50a-4887-b05a-713e3c3005c8) 演示中显示的输出电子邮件:[查看示例](https://gettoolbase.com/assets/demo-screenshot.png) 生成代码和更确定的工作流遵循类似的流程。 您可以在gettoolbase.com上尝试互动预览,无需注册。尝试点击出现在大蓝色文本旁边的工作流窗格右下角的播放按钮。提交查询后,点击每个节点可以查看其运行结果。全屏体验(右上角的图标)和展开节点(工作流步骤内的相同图标)可以让您以我们在演示中相同的方式训练该提示。第一个节点(“用户意图提取器”)故意模糊,以便您可以自己进行训练! ------- 为什么要另建一个代理/工作流构建工具? 我们开始构建Toolbase是因为对现有框架的失望,尤其是在生产使用方面,以及缺乏对MLOps工件(如提示、黄金数据、工作流和评估)的IDE支持。由于代理系统的大部分代码可以使用这些工件动态生成和验证,因此它们往往比代码本身更为重要。 与其他开发者交流后,我们也意识到,手动编码工作流、通过试错实验提示以及设置基础设施/集成都花费了远比应有的时间。像Cursor和Windsurf这样的工具有所帮助,但从AI生成的代码中提取意义的速度很慢。聊天机器人在有色聊天窗口中快速生成晦涩的代码,这种极端的方式演示效果很好,但根本无法维护(抱歉,氛围编码)。因此,我们选择了一种折中的方式:一个AI辅助的可视化构建工具,具有完整的代码回退功能。 ------- 您怎么看? 我们期待来自HN社区的任何反馈、想法或问题。 请在评论中告诉我们您的看法! - David & Ethan
查看原文
Hey HN, we’re David and Ethan, co-founders of Toolbase (gettoolbase.com).<p>Toolbase is an AI agent and workflow builder that helps you quickly create production-grade AI automations — batteries included.<p>------- The dev cycle in Toolbase: 1. Define a rough goal. 2. Connect any API or MCP server (we have thousands, or bring your own). 3. Teach the AI what valid input and output examples look like (these later become your unit tests!). 4. Let Toolbase generate the perfect prompt, code, workflow, or agent. 5. Deploy your project as an API, MCP server, or chat interface!<p>Coding optional. Sharing encouraged.<p>Note: If you’re familiar with Cursor&#x2F;Windsurf and the core concepts behind frameworks like Mastra (which Toolbase runs on), you already know how to use Toolbase. You retain the flexibility of coding but avoid boilerplate and plumbing tasks (integration, validation, context mapping, testing, etc.) unless you explicitly choose to do them.<p>------- Demo<p>This simple agent validates company billing addresses in our CRM (Pipedrive) by researching them through Tavily search. If there’s an address mismatch, it asks a human to pick the right one via email (watch on 2x speed):<p><a href="https:&#x2F;&#x2F;www.loom.com&#x2F;share&#x2F;540c61b2c5634996b088ebbb16989cf0?sid=8d076bd5-c50a-4887-b05a-713e3c3005c8" rel="nofollow">https:&#x2F;&#x2F;www.loom.com&#x2F;share&#x2F;540c61b2c5634996b088ebbb16989cf0?...</a><p>Output email for the example shown in the demo: <a href="https:&#x2F;&#x2F;gettoolbase.com&#x2F;assets&#x2F;demo-screenshot.png" rel="nofollow">https:&#x2F;&#x2F;gettoolbase.com&#x2F;assets&#x2F;demo-screenshot.png</a><p>Producing code and more deterministic workflows follows a similar process.<p>You can try an interactive preview at gettoolbase.com without registering. Try clicking on the play button at the bottom right of workflow pane that appears next to the big blue text. Clicking each node after you submit your query shows you its results for the run. Full-screening the experience (icon on the top right) and expanding a node (same icon within a step of the workflow) lets you train that prompt in the same way we did in the demo. The first node (“User Intent Extractor”) is purposefully vague so you can train it yourself!<p>------- Why another agent&#x2F;workflow builder?<p>We started building Toolbase out of frustration with existing frameworks, especially for production use, and the lack of IDE support for MLOps artifacts, such as prompts, golden data, workflows, and evaluations. Since much of the code for agentic systems can be dynamically generated and validated using these artifacts, they often become even more important than the code itself.<p>After speaking with other builders, we also realized that manually coding workflows, experimenting with prompts through trial-and-error, and setting up infrastructure&#x2F;integrations all took far more time than they should. Tools like Cursor and Windsurf help, but extracting meaning from AI-generated code is slow. Chatbots whipping up arcane code potions in tinted chat windows, which is the other end of the spectrum, demos really well but isn’t maintainable at all (sorry, vibe-coding). So we went with something in the middle: an AI-assiststed visual builder with full code fallback.<p>------- What do you think?<p>We’re excited for any feedback, thoughts, or questions from the HN community.<p>Let us know what you think in the comments!<p>- David &amp; Ethan