Hermes agent doesn't need a frontier model to run your computer well. Kimi K2.6 or Qwen 3.6 can do the same job at a fraction of the cost, and both are available fully private through the Venice API. 🧵 Here's the full Hermes "computer_use" setup:

Step 1: Install Hermes agent. Copy and paste this curl command into a terminal: curl -fsSL | bash Installs all dependencies, then auto-runs the Hermes setup. Quick-setup mode is fine for a first pass.

Step 2: Add Venice as a custom endpoint provider. Endpoint: Generate an API key in Venice: Paste it into the terminal setup. Venice has zero data retention: when you use private models, your prompts and screenshots stay yours.

Step 3: Find a private model. Kimi K2.6 and Qwen 3.6 27B do function calling + reasoning + vision + code at a fraction of GPT 5.5's cost.

Step 4: Skip the messaging platform. Telegram, Discord, WhatsApp: they store messages somewhere you don't control. That's a middleman. Run Hermes straight on your machine. Agent ↔ Private Inference ↔ You

Step 5: Let Hermes clean up its own config. By default Hermes includes a LOT of tools. You probably won't need most of them. "I want to use you for _____. Help me clean up the tools and skills we won't use." Hermes will edit itself, making its initial context load 50%+ smaller.

Step 6: Activate computer use. Cua driver + macOS Screen Recording + Accessibility permissions. Practice on simple tasks, like: • Using the Reminders app • Organizing your Desktop/Downloads folder Close other windows while it learns (focus theft is real).

Step 7 (bonus): Vibe-code native widgets into your desktop via Xcode. Widgets can't accept typed input, but you can build a dashboard with sessions + scheduled jobs + a button to open Hermes. Fully native. Fully private. At this point, the only limit is your imagination.

Result: a fully private AI managing your Mac, costing a fraction of the frontier alternatives. Don't sleep on open-source models on Venice. Generate your Venice API key and plug it into your Hermes Agent today:

Watch the full walkthrough from @jordanurbs here:

this is the way

Had this exact problem. Mnemosyne is a memory plugin for Hermes Agent that persists across sessions - hybrid search, entity extraction, auto-consolidates old memories. Self-hosted, your data stays local. Fixed this for me completely.

👀✨️

Cost often hinges more on action design than model size. Cheaper models work well when tasks have constrained tools, clear UI state, and a retry loop with screenshots after each step.

@NousResearch @AskVenice add native integration like ollama, xai and other platforms…

@jordanurbs 🔥🔥🔥

haven't tried kimi or qwen but every time I use grok or deepseek I end up going back to gpt-5.5 to fix what they broke

this is the direction ai agents need to go. no reason to send every screenshot and prompt through a cloud that stores it all. kimi and qwen running private with function calling is legit, been testing similar setups and the cost savings are real

Computer use

qwen 3.6 quietly eating hermes lunch

I showed the AI the video; it analyzed the process, provided me with guidance and support, and wrote the following prompt as part of the optimization process:

K2.6 running my desktop? challenge accepted

One of my favorite tokens. Proud bull of $VVV. 💯

Venice Ai supremacy

@willyogo It’ll only cost you an arm

This is starting to feel like

private inference plus Hermes is a setup more people should be running. built the marketing execution layer on top of this stack, Athena handles the pipeline work:

now i’m wondering how many people are still burning cash on frontier models for this exact job

Fully private computer use at a fraction of the cost. Venice just made a very compelling case

Efficiency always wins in the long run powerful tools at lower cost are changing the game fast, and smart traders are watching it all unfold on #MGBX

frontier flex feels pointless once you see thiw stack run circles at a fraction

so open source models were overkill all along, nice