Published: 2026-08-05

Qwen 3.8 Max as the Model Behind a Hermes Agentic OS

Chapters / key moments (click to jump — plays here on the page)

A screen tour of using Qwen 3.8 Max as the model behind an agentic operating system, with Hermes (or Claude Code) as the harness that does the work and a workflow UI on top — plus memory auto-saved to an Obsidian vault, a Hunter-API lead-outreach workflow, and a scheduled news-digest task. Goldie claims the model is a 2.4-trillion-parameter open-source release with a million-token context window that outperforms Fable 5 on several benchmarks; those figures are his, not independently verified here, and the video shows no install steps or commands — he estimates three to four hours a day to build the setup and points viewers to his paid community for a copy of it.

Watch on YouTube → · ← Back to News

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Reproducible steps

Think of the stack in three layers
The speaker splits the setup into the brain (Qwen 3.8 Max, which he says is open source with 2.4 trillion parameters and a million-token context window), the hands (an agent harness such as Hermes Agent or Claude Code), and the work surfaces (a UI with chat, voice, a kanban board and custom workflows).
Swap the brain, keep the harness
Hermes lets you switch the model for different jobs, and the same Qwen model can be used as the brain inside Claude Code as the harness. A Qwen coding plan can be plugged into Hermes; he says you can't use the Claude CLI with this system.
Run long jobs through a goal system
Plug the model into a goal loop so it builds for hours in the background, and come back to a history of the goals. He says Qwen is designed to code autonomously for hours.
Save every exchange to an Obsidian vault
A memory loop writes every exchange to an Obsidian vault so the model reads your business context instead of starting from zero.
Lead outreach workflow
In his outreach section he plugged in a Hunter API key and used Qwen as the brain: describe who you want (his example: SEO agencies interested in white-label backlinks), it finds and enriches leads, drafts an email campaign, schedules it and manages the inbox.
Use a frontier model as orchestrator over cheap workers
He runs Paperclip with Hermes and Qwen 3.8 as the CEO agent, and free or local models as sub-agents doing the underlying work.

Gotchas

The benchmark claims (Qwen 3.8 Max matching or beating Fable 5 and GPT 5.6 Sol, with side-by-side comparisons on a leaderboard site) are the speaker's, not verified here.
He says building a setup like this takes three to four hours a day, and the finished setup is sold through his paid community.
No install commands are shown; the video is a tour of a custom dashboard.