HackerNews中文版

您好，我正在为我未来的 GH 服务开发一个入职流程。我认为如果用户在 AI 聊天的帮助下充实他们的入职回复，将有助于提高入职回答的质量。但是，AI 聊天没有集成到 Web 应用程序的任何其他部分，这意味着我必须专门为入职流程构建整个聊天界面——这似乎非常浪费时间。我考虑过，不如允许用户将代理/系统提示复制到他们选择的 LLM 聊天应用程序中，然后在代理确定其质量足够高时再完成入职流程。这时，我很惊讶地发现，没有用于 LLM 的 `mailto://` 类型的 URI 方案。或者至少我不知道有。我考虑过类似这样的方案： ``` llm://[最小模型生成]?subject=....&prompt=......&schema=[base64 编码的模式输出] ``` 例如： ``` # 完整链接示例在链接的 gist [0] 中 llm://>2024?subject=Yocto.is%20Onboarding%20Helper%20Agent&prompt=... ``` 第一部分也可以选择一个特定的模型例如： ``` # 特定模型 llm://gemini-2.5-pro?... # 特定提供商，最小模型 llm://>=gemini-2.5-pro?... # 特定提供商，大于模型 llm://>gemini-2.5-pro?... ``` 谢谢等等。 [0]: https://gist.github.com/smashah/f6192d7af114ca059b3a47b33ec1df18

查看原文

Hello,I'm developing an onboarding flow for my future OF for GH service, I think it would help with the quality of onboarding answers if users fleshed out their onboarding responses with the help of an AI chat. But AI chat is not integrated in any other part of the webapp which means I have to build out this whole chat interface just for onboarding - seems like a massive waste of time.I was thinking instead allowing the users to copy over the agent/system prompt to their LLM chat app of choice and then just complete the onboarding once the agent determines its of sufficient quality.At this point I was surprised to learn that there is no `mailto://` type URI scheme for llms. Or at least not that I know of.I was thinking something like this:``` llm://[min-model-generation]?subject=....&prompt=......&schema=[base64 encoded schema output] ```e.g``` # full link example is in the linked gist [0] llm://>2024?subject=Yocto.is%20Onboarding%20Helper%20Agent&prompt=... ```the first part can select a specific model as welle.g``` # Specific model llm://gemini-2.5-pro?... # Specific provider, min model llm://>=gemini-2.5-pro?... # Specific provider, gt model llm://>gemini-2.5-pro?... ```Thanksetc.[0]: https://gist.github.com/smashah/f6192d7af114ca059b3a47b33ec1df18

Ask HN: 有 AI 聊天应用实现了 LLM URI 方案吗？