Hermes Agentic OS is The Future
Decision Card
Effort: Afternoon project (~2–3 hours) — install Hermes Agent with the one-line installer from hermes-agent.nousresearch.com, grab the AionUi release for your OS from the AionUi GitHub releases, select Hermes as the primary agent, and run one test task (spreadsheet build or folder cleanup).
Honest take: The video is a surface-level demo that garbles both project names (“News Research” for Nous Research, “Ion”/“Gensec UI” for AionUi) and claims you can drive AionUi from WhatsApp, while AionUi’s own documentation lists Telegram, Lark, DingTalk and WeChat — not WhatsApp (WhatsApp is a Hermes-side channel). It also runs demos in “Yolo” auto-approve permission mode without a word about the security implications of an autonomous agent with file, code and browser access.
Concrete next steps:
- Read the Hermes Agent README to understand its memory/skills model before layering a GUI on it (~20 min)
- Install AionUi (github.com/iOfficeAI/AionUi, Apache-2.0), confirm it auto-detects your locally installed CLI agents, and run one bounded task in a scratch directory with review-mode permissions instead of Yolo (~1 hour)
- If remote access matters, test AionUi’s documented channels (Telegram / WebUI per the AionUi wiki) rather than the WhatsApp path the video implies (~30 min)
- Skip if you don’t already run (or want to run) a persistent self-hosted agent — AionUi adds a cowork GUI on top of Hermes; it doesn’t remove the underlying setup, API-key, and supervision burden.
TL;DR
The video shows how to combine Nous Research’s open-source Hermes Agent with AionUi, a free cross-platform desktop “cowork” app, to build a local “agentic OS” where multiple autonomous agents handle spreadsheets, file organization, report writing, browser tasks and small coding jobs in parallel. It is a setup walkthrough plus three cherry-picked demos (financial dashboard in Excel, desktop file cleanup, a small coin-collecting game), with links promised in the description.
Key Points
- Hermes Agent is an open-source (MIT) project by Nous Research designed to run 24/7 on your own infrastructure while building long-term memory, reusable skills, and a model of its user. 00:16
- The core pitch: turn Hermes into an “agentic OS” — an infrastructure layer managing multiple autonomous agents on multi-step tasks without constant oversight. 00:45
- The enabling piece is AionUi (“IonUI” in the video), a free open-source cowork platform where agents read/manage files, write and execute code, browse the web, and automate workflows visibly alongside the user. 01:46
- Prerequisites: Hermes Agent installed (via its GitHub repo), plus AionUi requirements of macOS 10.15+, 4 GB RAM, 500 MB storage. 02:55
- AionUi auto-detects locally installed AI agents; you pick Hermes as the primary agent powering the OS, and can also connect remote agents by URL. 03:59
- Task-specific “assistants” combine agents with custom rules and skills (e.g. a 3D-morph PPT assistant), and you can inject additional skills and MCP servers. 04:41
- Remote control is claimed via WhatsApp or Telegram from a phone, plus a WebUI — the WhatsApp part doesn’t match AionUi’s docs. 05:52
- Permission modes range from “Yolo” (full autonomy) to auto-edit with per-change review; the demo runs autonomously. 06:49
- Demo 1 builds a three-sheet financial dashboard (income statement, revenue breakdown, charts, conditional formatting) while a second agent simultaneously writes an EV-charging white paper — the multi-agent selling point. 07:09
- Demos 2 and 3: desktop thumbnail files reorganized numerically into a new folder, and a small coin-collecting game coded from a preconfigured skill. 09:07
Notable Quotes
“it’s an agent that can run 24/7 on your own infrastructure while building long-term memory, reusable skills, and even a deeper understanding of its user itself.” 00:20
“it is a free open-source co-work platform that allows AI agents like Hermes to work alongside you directly on your computer, essentially creating a fully autonomous agentic AI operating system.” 01:53
“we now have an agentic OS that is running autonomously powered by Hermes agent that is able to work with us to do anything like managing apps, creating automation pipelines.” 09:54
Verified Claims
Hermes Agent is built by Nous Research under the MIT license. 00:16 — Sources: NousResearch/hermes-agent, MIT LICENSE file. Verdict: Confirmed (video says “News Research”; correct name is Nous Research).
Hermes runs continuously on your own infrastructure with persistent memory, self-improving skills, and a deepening user model. 00:18 — Sources: Hermes Agent README, hermes-agent.nousresearch.com. Verdict: Confirmed — README describes agent-curated memory, autonomous skill creation, user modeling, and deployment from “$5 VPS” up.
AionUi is a free, open-source cowork platform where agents read/manage files, write and execute code, browse the web, and automate workflows. 01:51 — Sources: iOfficeAI/AionUi, aionui.com. Verdict: Confirmed (Apache-2.0, Electron + React, cross-platform).
System requirements: macOS 10.15+, 4 GB RAM, 500 MB available storage. 03:01 — Source: AionUi GitHub. Verdict: Confirmed (repo lists macOS 10.15+, Windows 10+, Linux distros, 4GB+ RAM, 500MB+ storage).
AionUi auto-detects AI agents installed locally, including Hermes. 03:59 — Source: iOfficeAI/AionUi. Verdict: Confirmed — repo advertises auto-detection of 12–20+ CLI agents (Hermes Agent, Claude Code, Codex, Gemini CLI, etc.).
You can control AionUi from your phone via WhatsApp or Telegram. 05:54 — Sources: iOfficeAI/AionUi, AionUi wiki. Verdict: Disputed — AionUi documents Telegram, Lark, DingTalk, WeChat and a remote WebUI; WhatsApp is not listed (WhatsApp is a channel of Hermes itself, per the Hermes README).
AionUi supports adding skills and MCP servers. 05:16 — Source: iOfficeAI/AionUi. Verdict: Confirmed — three-tier skill system and “configure MCP tools once, automatically sync to all agents.”
Multiple autonomous agents can run simultaneously on different tasks. 07:11 — Source: iOfficeAI/AionUi. Verdict: Confirmed — repo describes multi-agent coordination / Team Mode with parallel execution.
Tools, Papers & Standards Mentioned
- Hermes Agent (Nous Research) — github.com/NousResearch/hermes-agent · official site
- AionUi (“IonUI” in the video) — github.com/iOfficeAI/AionUi · aionui.com · wiki
- MCP (Model Context Protocol) — referenced as AionUi’s extension mechanism; configured per-app, synced to agents (AionUi GitHub)
- Telegram / WhatsApp — messaging channels; Hermes supports both (Hermes README), AionUi supports Telegram (AionUi GitHub)
- Gemini, Codex — mentioned in passing as alternate agents selectable inside AionUi (AionUi GitHub)
Follow-up Questions
- What does the permission model actually enforce when AionUi runs Hermes in “Yolo” mode — is there any sandboxing of file, shell, and browser access, or full user-level privileges?
- How do Hermes’s long-term memory and self-created skills persist and compose when driven through AionUi versus its native gateway (Telegram/CLI) — one shared memory store or per-frontend state?
- What are realistic token/API costs and failure rates for the multi-agent parallel workloads shown (spreadsheet + white paper + file ops), and how does AionUi surface or cap runaway agent loops?