Hermes AI: Automate & Build ANYTHING!
The video demonstrates several tools for managing Hermes AI agents, including Hermes Web UI, Hermes Desktop App, and Browser Use Box, comparing them to alternatives like Claude Code and OpenClaw. The host also covers the newly released Xiaomi Mimo V2.5, an open-source local model designed for agentic tasks. Throughout, the presenter emphasizes ease of use, customization, and free/open-source accessibility as key advantages of the Hermes ecosystem.
Summary
The video opens with a brief look at Browser Use Box, described as a 24/7 remote setup for running AI agents using Claude Opus, before pivoting to the main content: a hands-on walkthrough of Hermes Web UI. The presenter contrasts the messy terminal interface used to interact with Hermes against the cleaner, more organized web-based UI, noting that the latter is mobile-responsive and easier to navigate. Setup is shown to be extremely quick — essentially a single copy-paste command — and the UI supports multiple model providers via Open Router, Ollama, DeepSeek, and others.
The presenter walks through the various management features of Hermes Web UI, including scheduled tasks, skill management, memory, session history, and multiple agent profiles. He notes that he personally runs several distinct Hermes profiles with varying numbers of skills (e.g., 88 vs. 155), and that managing these through the terminal is significantly messier. He compares Hermes Web UI favorably to Open Web UI, describing it as more responsive and easier to configure, while clarifying that the two are entirely separate open-source projects with different purposes.
The video then transitions to testing the Hermes Desktop App, a newer free tool that offers a graphical interface for managing Hermes agents. Key features highlighted include: connecting to messaging platforms like Telegram, Discord, and email; switching between LLM providers and local/remote modes; exporting and importing configurations to avoid breakage during updates; a virtual 3D office view for agents ('Claude 3D'); and a unified session view showing conversation history across all channels (CLI, Web UI, Telegram, etc.). The presenter positions this desktop app as the most user-friendly option for non-technical users.
Throughout the video, the host repeatedly compares Hermes to Claude Code and OpenClaw. He argues Claude Code is easier and more reliable but less customizable and locked into Anthropic's ecosystem. OpenClaw, by contrast, is described as frequently buggy and unreliable, with usage data from Open Router shown to support the claim that Hermes is gaining popularity while OpenClaw's usage fluctuates downward after each new release.
The final segment covers the newly announced Xiaomi Mimo V2.5, a mixture-of-experts open-source model available on Hugging Face. The presenter explains its architecture — 310 billion total parameters but only 15 billion activated at runtime — making it relatively lightweight for local deployment. The Pro version has a 1 million token context window and 1 trillion total parameters. The model is shown performing well on agentic benchmarks, reportedly outperforming DeepSeek V4 Pro and Communicator 2.6. The presenter demonstrates it via Xiaomi's web chat interface and shows coding examples. He advises viewers to access it via Hugging Face or LM Studio for local use, noting it may not yet be indexed in LM Studio given how recently it was released.
Key Insights
- The presenter argues that Hermes is growing in Open Router usage while OpenClaw's usage drops after every new release due to bugs, showing two diverging trend charts as evidence that reliability is driving users toward Hermes.
- The presenter compares Claude Code to 'a car that's ready to drive straight away' versus Hermes, which he likens to building your own car — less convenient out of the box but far more customizable, including the ability to switch between any LLM provider.
- The presenter claims that Xiaomi Mimo V2.5 Pro outperforms DeepSeek V4 Pro and Communicator 2.6 on agentic benchmarks, and emphasizes that it was specifically built for agentic tasks like those run by Hermes and OpenClaw.
- The presenter notes that Mimo V2.5 uses a mixture-of-experts architecture where only 15 billion of 310 billion total parameters are activated at runtime, making it more feasible to run locally despite its large overall size.
- The presenter highlights that the Hermes Desktop App includes a session history view that consolidates conversations across all channels — CLI, Web UI, and Telegram — in one place, addressing a frequently requested feature from users who want unified message management.
Topics
Full transcript available for MurmurCast members
Sign Up to Access