Let's kick things off with Nous Research, because there is a *lot* happening with Hermes Agent right now — and it's moving fast.
First, the big platform news. Nous Research launched hosted Hermes Agent on the Nous Portal. The pitch is simple: pick your model, pick your server size, and you're live in sixty seconds. No VPS. No terminal. No Docker. Just — running. Pricing tiers go from free pay-as-you-go all the way up to two hundred dollars a month. And what you get across those tiers is genuinely wild — over three hundred models, a Tool Gateway that covers web search, image generation, text-to-speech, and browser automation, plus native integrations with Telegram, Discord, Slack, WhatsApp, and Signal. On top of that: profiles, cron jobs, a kanban board, skills, memory, sub-agents, mixture-of-agents, goal tracking, a learn mode, journey visualization — the list keeps going. The MIT self-hosted version is still there, unchanged, for anyone who wants to run it themselves.
Then Teknium dropped Hermes Agent v0.18, which they're calling the "Judgement Release." The headline? They cleared roughly two thousand open issues and pull requests — and knocked out every single P0 and P1 bug. That's a serious cleanup. But the new features are just as interesting: a first-class Mixture of Agents virtual model, a slash-learn command that pulls from open sources, Journey visualization so you can actually see your agent's progress, Gemini Vertex as a new provider, support for Fable 5, Sonnet 5, and Fugu models, desktop Projects, self-verifying work, editable memory, multi-model consulting, and background sub-agents running while you do other things.
And the model integrations keep stacking up. Teknium added Grok 4.5 to Hermes Agent — routed through the Nous Portal, the Grok slash X API, or OpenRouter. Then, day-zero, they also dropped support for GPT-5.6 Sol. Same routing options: Nous Portal, your OpenAI subscription, or OpenRouter. So if a new frontier model drops, it sounds like Hermes Agent is gunning to have it available immediately.
Now here's a story that really illustrates what people are actually *doing* with this thing. Samuel Cardillo shared feedback from a neighbor — an SVP at a fintech company — who ran Qwen 3.6 35B A3B on an NVIDIA DGX Spark through Hermes Agent's desktop client. The model plus the Hermes harness apparently delivered what they called mission-success results. That's a pretty strong signal that local enterprise hardware plus Hermes Agent is a combo people are taking seriously.
And on the community side — someone built an unofficial open-source browser extension that puts Hermes Agent in a side panel on any browser tab. We're talking model swapping, vision and screenshot support, session picking, and the ability to connect to either a local or remote gateway. It's not official, but the demo video made the rounds and it looks genuinely polished.
That's your AI digest for 10 Jul 2026.