Video: "Hermes Agent 2.0 is INSANE!" by Julian Goldie on YouTube.
What Hermes 2.0 is and where it sits in the release history
Hermes Agent has been releasing steadily through 2026. Version 0.19 — Quicksilver — cut cold-start times from 4.3 seconds to 0.9 seconds and added a persistent job database so tasks finish even if the agent app closes. That was covered in our Quicksilver summary.
Version 0.20 is the release Julian is covering in this video, and the version number maps to the "2.0" label Nous Research is using. The core addition is the Model Portal — a layer that sits above the individual model calls and decides which of the connected models should handle each task, rather than routing everything through a single configured model.
The Model Portal: what it does
The Model Portal connects Hermes to over 200 AI models through a single interface. Rather than choosing one model for all tasks, you configure which type of work goes to which model. SEO and content generation might go to Qwen 3.8 Max. Coding tasks to Sonnet 5. Outreach and email drafting to GPT. The Portal handles the routing once it is configured.
This matters because different models have genuine strengths in different areas. Using a single model for all agent tasks is a compromise that costs either quality or money — the best writing model is not always the cheapest coding model, and the cheapest model is rarely the best at both. The Portal is Hermes's answer to that trade-off: one agent stack, multiple models, each doing the task it is best suited for.
Julian demonstrated this routing live in the video. The Hermes dashboard shows which model is active for each running task. The configuration sits in the Hermes settings rather than requiring separate tool setups for each model.
Multi-agent routing from one gateway
The second significant change in 2.0 is multi-agent routing. A single Hermes gateway can now route work to separate specialised agent instances simultaneously — an SEO agent, a writing agent, a coding agent, an operations agent — rather than one instance handling everything in sequence.
Julian's demo showed a 100-agent SEO team running from one Hermes installation: each agent takes a keyword, runs the full pipeline (research, draft, optimise, publish), and hands off. The gateway coordinates the queue. This is not a use case for a solo operator, but it demonstrates what the architecture is capable of at scale for agencies or businesses with high content volume.
For a smaller team, the more immediate application is simpler: two or three specialised agents running in parallel, each focused on one category of task, rather than a general-purpose agent switching between them and loading different contexts each time.
How this fits alongside Hermes Cloud
The 2.0 release arrives alongside Hermes Cloud, covered in our Hermes Cloud summary. Cloud handles the always-on deployment problem — keeping agents running when your laptop is closed. Version 2.0 handles the model-selection problem — making sure each task goes to the right model once those agents are running.
Used together: you deploy a Hermes 2.0 instance to Hermes Cloud, configure the Model Portal to route tasks across your chosen models, and the whole stack runs 24/7 without local hardware. The cloud handles uptime; the Portal handles which model to use.
What is confirmed and what still needs checking
Julian's walkthrough is a product demo from someone closely involved with Hermes promotion, so the usual caveat applies. The Model Portal and multi-agent routing are shown live in the video, which puts them in the confirmed category. The 200+ model count is the figure Julian cites; the exact list of connected models was not fully enumerated in the video.
The 100-agent team demo is a stress test rather than a typical use case. Julian ran it to show the architecture holds, not to suggest every business needs 100 agents. The practical ceiling for a small team is probably two to five specialised agents, each with a defined role and model assignment.
API costs are the main unknown. Running multiple models simultaneously across multiple agents will generate more API spend than a single-model setup. The Model Portal makes it easier to use the right model, but it does not make the calls cheaper. Anyone deploying 2.0 at scale should model the cost before committing.
Where this connects to NordSys
The agents we configure for clients can already be matched to the best-available model for each task. Hermes 2.0's Model Portal formalises that approach at the framework level — which is a direction we welcome. If you want an AI agent stack configured for your business, with the right models for the right jobs, without tracking every Hermes release yourself, our AI Agents service handles that. From £6 a day.
See our AI Agents →