Running Local AI Agents with Intelligent Terminal on a Project Zenith Dev Box

This session puts a Project Zenith class device to work entirely offline, pairing Intelligent Terminal with a 122-billion-parameter Qwen 3.5 model running through Ollama. The hosts take a repetitive Windows image and Hyper-V VM task that normally requires many manual steps and hand it to the agent, then follow the result from prompt to a booted virtual machine.

■ Automated VM and Imaging Workflow
- The manual steps involved in spinning up a VM and applying a new image for smoke testing
- Asking the agent pane to turn a repeated task into a reusable skill
- Locating a Windows image in a folder and requesting a bootable VM from it
- Permission prompts, ISO mounting, and the VHD booting into the Windows out-of-box experience

■ Intelligent Terminal Setup
- Agent pane, agent status bar, and hotkey access within an open-source fork of Windows Terminal
- Connecting a local model with an endpoint URL, ID, and API key
- Model-agnostic and ACP-compatible use with GitHub Copilot, Claude, or Gemini
- Memory usage and responsiveness observed while the large local model runs
- Installation through the Microsoft Store, GitHub, or winget

If you automate shell tasks or evaluate on-device AI, this walkthrough shows how a local model can take over a repeated administrative routine without cloud dependencies, and what it looks like end to end on capable local hardware.