Video: "Hermes Agent OS + Kimi K3 is CRAZYY GOOD!" by Julian Goldie on YouTube.

What Hermes Agent OS actually is

Hermes Agent OS is not a single tool — it is a layer of connected workflows running on top of the Hermes agent. Think of it as a dashboard with named jobs: Oracle watches for trending topics and publishes SEO articles to WordPress; Astros monitors competitor keywords, titles and fresh content angles; Jarvis takes voice commands and turns them into running tasks; a shared Obsidian memory layer keeps context persistent across all of them.

On its own, this is useful. Each workflow runs from a goal prompt with no manual prompting at each step. The model brain underneath is the thing that determines how well each job gets done — and up to now, most people running Hermes were paying for a frontier API to get reliable results.

Why Kimi K3 changes the economics

Kimi K3 from Moonshot AI sits in an unusual position. It is free to use at kimi.com, open-source from July 2026, and performs at roughly the same level as Claude Fable 5 on Julian's own benchmarks — edging out GPT-5.6 and Opus 4.8 on long-context reasoning tasks. The 1-million-token context window is not just a headline number; it is practically relevant inside Hermes, where an agent reading a long content plan, a competitor site, or a 200-article knowledge base needs to hold all of it in one call.

That said, free-tier usage at kimi.com is rate-limited. A $20 deposit to the Moonshot API moves you to a higher tier — enough for daily Hermes sessions without the workflow stalling. Worth knowing before you assume it will run all day for nothing.

The marketing workflows in practice

Oracle is the most immediately useful part for most business owners. Give it a niche and a publishing target, and it finds trending topics, drafts a structured SEO article, and pushes it to WordPress — no human in the loop once you have set the goal. With a 1-million-token context window, Kimi K3 can hold the site's existing articles in context while writing, which reduces internal duplication and improves topical coherence across a cluster.

Astros runs in the background and monitors competitor sites for new content, keyword movements and fresh angles. It flags what is worth responding to and queues it for Oracle. In practice this means your agent is watching the market for you rather than waiting to be prompted.

Jarvis is less about automation and more about speed. Voice commands go in — tasks come out. Useful if you are the kind of person who finds typing a prompt slower than just saying what you want. Not a replacement for structured goals, but a reasonable interface for ad-hoc work.

The /learn command and Blender MCP

One addition Julian covers in detail is the /learn command. Point it at a URL — a guide, a how-to, a product page — and Hermes reads it, extracts the key instructions, and saves them as a local skill. Next time you run a related task, the agent already knows the workflow. This is how you build a custom knowledge base inside your agent without manually writing prompts for every use case.

The Blender MCP connection is more niche but worth noting if you sell physical products. Connecting Hermes to Blender through the Model Context Protocol lets Kimi K3 model a 3D product and generate a short promo video from a single prompt. Earlier models struggled with the multi-step coordination this requires. Kimi K3's stronger long-horizon reasoning handles it more reliably.

What is genuinely new here and what is not

The Hermes Agent OS architecture has been building for months. Oracle, Astros and Jarvis have all featured in earlier Julian Goldie videos. What is new is the combination: Kimi K3 as the reasoning layer, free (or very cheap), with a context window large enough for the whole stack to work properly without truncation errors on longer jobs.

To be fair: the quality ceiling depends on how well you configure the system. A poorly structured Oracle goal will still produce generic content; Astros is only as useful as the competitor list you give it. Kimi K3 improves what the stack can do, but it does not fix a setup that was not working in the first place. Start with the simplest workflow — Oracle on one topic — before wiring in the full OS.

Where this connects to NordSys

NordSys sets up and maintains Hermes Agent and its workflows for UK businesses. That includes choosing and configuring the model brain — switching to Kimi K3, or to whichever open-source release makes sense when the landscape moves again. If you want Oracle, Astros or Jarvis running for your site without managing the configuration yourself, that is what the AI Agents service covers.

See our AI Agents service →