Hermes Analyst 体验升级：UI 大幅改进与嵌套编排器解析

✍ 0xJeff🕐 2026-06-17📦 5.3 KB 🟢 已读 𝕏 文章列表

作者分享了 Hermes Analyst Agent 近期的重大更新体验。重点介绍了 Hermes Desktop 端的发布解决了 WSL2 下的文件访问痛点，提供了类似 ChatGPT 的可视化和配置管理界面。文章深入解析了“嵌套编排器（Nested Orchestrator）”功能，通过编排层和子代理的并行协作，显著提升了复杂研究任务的输出质量和深度。此外，还提及了 Stripe 集成和 Telegram 富文本支持等新特性。

Hermes Agent Desktop 嵌套编排器工具评测 Nous Research UI/UX 研究能力

# Hermes Analyst 10x Better **作者**: 0xJeff **日期**: 2026-06-16T14:00:20.000Z **来源**: [https://x.com/0xJeff/status/2066883577141428563](https://x.com/0xJeff/status/2066883577141428563) --- ![Image](/doc/2530/image/img_821f5409a14d.jpg) Hermes Analyst just got 10x better — the UI/UX drastically improved + configuring the agent has never been easier + research capability is better thanks to nested orchestrator feature In the past 2 weeks, @NousResearch team has shipped a lot of cool stuff - Hermes Windows - Hermes Desktop - Agent profiles - Asynchronous sub-agent - Nested orchestrator - Native agentic commerce integration with @stripe - Rich text/outputs on Telegram And probably a whole lot more that I missed out. In this article, I’ll quickly go through what I’m most excited about on the changes + my experience of using Hermes after the last round of patch updates > tl;dr -> it’s 10x smoother, props to Nous team I’ve been using Hermes on WSL2 on Windows which hasn’t been ideal — accessing files on Linux was complicated, sometimes the agent would drop the file in places that were hard to access, sometimes it’d do things on Linux that wouldn’t surface on Windows. And so Hermes Windows + Hermes Desktop came at a right time and the migration was simple - “hermes backup” on Ubuntu - install Hermes Desktop by downloading .exe client on Nous site - and then “hermes import” Everything was ported over (except agentcash x402 wallet + Hindsight data which I had to manually migrate) but all in all pretty seamless migration. ## Hermes Desktop - Designed for Everyone ![Image](/doc/2530/image/img_babec7a70eea.jpg) Hermes Desktop is a desktop UI that’s designed to be a headquarter of using Hermes. ChatGPT/Claude-like interface with rich visuals, chain of thoughts, tables, the outputs all of which are shown in a paginated manner now (vs. just simple texts). This makes it a lot easier to go through the outputs than before. It’s also easier to check & configure skills, messaging apps, cron jobs, models, different agent profiles via Hermes Desktop instead of having to manually go to .env or config.yaml to adjust things. ![Image](/doc/2530/image/img_cc30cf7cd94d.jpg) Given the recent Fable 5 ban + OpenRouter Fusion (where a combination of open models can achieve Fable 5-level of intelligence), there’s a lot better R/R in tinkering with Hermes now. More higher quality workflows that can offload your tasks thanks to better harness/toolkits & models powering them. ## Hermes Research Capability just got 10x better If you’ve been following my Hermes Analyst series, you might remember the Deep Research Pipeline, a skill I created for Hermes to do a structured deep research better. The challenge with that was, the agent does the entire research step-by-step, waiting for one thing to finish before starting another. This is ok but when the pipeline expanded and many tools get wired into it, it became difficult for the agent to stick to all the tools. The result? Hermes skipped a few tools and the output degraded in quality. “Skill bundling” (i.e. let sub agents batch skills and execute them in parallel) does help ensure that the agent doesn’t skip a tool. BUT.... there’s a way to do this better with much better research output. Nous Research just released a feature called “Nested Orchestrator“ This is a feature where Hermes spawns - 1 Orchestrator sub-agent (or more) - The orchestrator then spawns 3 worker (leaf agents) to do complex work Depending on how complex the task is, more orchestrators can be run concurrently, managing more leaf agents. ![Image](/doc/2530/image/img_e04340c6c68c.jpg) The result = orchestrator cross-pollinate sources from leaf agents, synthesize the insights and the delta, resulting in better outputs. You could probably see the difference in the outputs below (the right is a lot richer and more detailed than the left) ![Image](/doc/2530/image/img_2a193972258c.jpg) ## What’s Next? I’m super excited to use Telegram to do quick query with Hermes now since rich text messages are enabled. Met many cool teams during SuperAI and got a list of cool repos/tools I need to catch up and experiment on. Will be testing them out this week and share the use cases with you guys in the next episode. For now, I’ll continue running structured research pipeline with the nested orchestrators to get up to speed and macro, geopolitics, equities, crypto, and more. If you want to learn more about Hermes + how to train him to become an ultimate analyst - feel free to check out last week's article Thanks for reading and see you in the next one! ## 相关链接 - [0xJeff](https://x.com/0xJeff) - [@0xJeff](https://x.com/0xJeff) - [23K](https://x.com/0xJeff/status/2066883577141428563/analytics) - [@NousResearch](https://x.com/@NousResearch) - [@stripe](https://x.com/@stripe) - [Upgrade to Premium](https://x.com/i/premium_sign_up) - [10:00 PM · Jun 16, 2026](https://x.com/0xJeff/status/2066883577141428563) - [23.6K Views](https://x.com/0xJeff/status/2066883577141428563/analytics) - [View quotes](https://x.com/0xJeff/status/2066883577141428563/quotes) --- *导出时间: 2026/6/17 07:59:54* --- ## 中文翻译 # Hermes 分析器性能提升 10 倍 **作者**: 0xJeff **日期**: 2026-06-16T14:00:20.000Z **来源**: [https://x.com/0xJeff/status/2066883577141428563](https://x.com/0xJeff/status/2066883577141428563) --- ![Image](/doc/2530/image/img_821f5409a14d.jpg) Hermes Analyst 刚刚实现了 10 倍的性能提升——UI/UX 大幅改进 + 配置代理从未如此简单 + 得益于嵌套编排器功能，研究能力也得到了增强。在过去两周里，@NousResearch 团队发布了很多很酷的功能： - Hermes Windows - Hermes Desktop - 代理配置文件 - 异步子代理 - 嵌套编排器 - 与 @stripe 的原生代理商务集成 - Telegram 上的富文本/输出可能还有更多我遗漏的内容。在本文中，我将快速浏览我对这些变化中最兴奋的部分 + 我在上一轮补丁更新后使用 Hermes 的体验 > tl;dr -> 体验流畅度提升了 10 倍，向 Nous 团队致敬我之前一直在 Windows 的 WSL2 上使用 Hermes，这并不理想——访问 Linux 上的文件很复杂，有时代理会把文件放在难以访问的地方，有时它在 Linux 上做的一些操作无法在 Windows 上显示。因此，Hermes Windows 和 Hermes Desktop 的出现恰逢其时，而且迁移过程很简单： - 在 Ubuntu 上运行 `hermes backup` - 通过 Nous 网站下载 .exe 客户端安装 Hermes Desktop - 然后运行 `hermes import` 所有内容都移植过来了（除了 agentcash x402 钱包 + Hindsight 数据，这是我必须手动迁移的），但总体来说迁移过程非常顺畅。 ## Hermes Desktop - 为每个人而设计 ![Image](/doc/2530/image/img_babec7a70eea.jpg) Hermes Desktop 是一个桌面 UI，旨在成为使用 Hermes 的大本营。它具有类似 ChatGPT/Claude 的界面，包含丰富的视觉效果、思维链、表格，所有输出现在都以分页的方式显示（而不仅仅是简单的文本）。这使得浏览输出比以前容易得多。通过 Hermes Desktop 来检查和配置技能、消息应用、定时任务、模型以及不同的代理配置文件也更容易，而不必手动去修改 .env 或 config.yaml 文件。 ![Image](/doc/2530/image/img_cc30cf7cd94d.jpg) 鉴于最近的 Fable 5 封禁事件 + OpenRouter Fusion（即开源模型的组合可以达到 Fable 5 级别的智能），现在折腾 Hermes 的投入产出比要高得多。得益于更好的利用工具/工具包以及支持它们的模型，可以卸载你任务的更高质量的工作流出现了。 ## Hermes 的研究能力刚刚提升了 10 倍如果你一直关注我的 Hermes Analyst 系列，你可能还记得“深度研究管道”，这是我为 Hermes 创建的一项技能，用于更好地进行结构化的深度研究。其中的挑战在于，代理会一步一步地完成整个研究过程，等待一件事完成后才开始另一件。这虽然还可以，但当管道扩展并且许多工具接入其中时，代理就很难坚持使用所有工具。结果就是？Hermes 跳过了一些工具，导致输出质量下降。 “技能捆绑”（即让子代理批量处理技能并并行执行）确实有助于确保代理不会跳过工具。但是……有一种更好的方法可以获得更好的研究输出。 Nous Research 刚刚发布了一项名为“嵌套编排器”的功能。这个功能让 Hermes 生成 - 1 个编排器子代理（或更多） - 然后编排器生成 3 个工作器（叶子代理）来做复杂的工作根据任务的复杂程度，可以并发运行更多的编排器，从而管理更多的叶子代理。 ![Image](/doc/2530/image/img_e04340c6c68c.jpg) 结果 = 编排器交叉引用来自叶子代理的资源，综合见解和差异，从而产生更好的输出。你可能会在下面的输出中看到差异（右侧比左侧更丰富、更详细） ![Image](/doc/2530/image/img_2a193972258c.jpg) ## 接下来是什么？既然启用了富文本消息功能，我现在非常兴奋能通过 Telegram 使用 Hermes 进行快速查询。在 SuperAI 期间遇到了很多很棒的团队，我收到了一系列很酷的仓库/工具清单，我需要去了解和实验。本周我将测试它们，并在下一期与你分享用例。现在，我将继续使用嵌套编排器运行结构化研究管道，以了解宏观、地缘政治、股票、加密货币等方面的动态。如果你想了解更多关于 Hermes 的信息 + 如何训练它成为终极分析师 - 欢迎查看上周的文章感谢阅读，下一期见！ ## 相关链接 - [0xJeff](https://x.com/0xJeff) - [@0xJeff](https://x.com/0xJeff) - [23K](https://x.com/0xJeff/status/2066883577141428563/analytics) - [@NousResearch](https://x.com/@NousResearch) - [@stripe](https://x.com/@stripe) - [升级到 Premium](https://x.com/i/premium_sign_up) - [10:00 PM · Jun 16, 2026](https://x.com/0xJeff/status/2066883577141428563) - [23.6K Views](https://x.com/0xJeff/status/2066883577141428563/analytics) - [View quotes](https://x.com/0xJeff/status/2066883577141428563/quotes) --- *导出时间: 2026/6/17 07:59:54*

🔗 相关文档