Running a local AI agent on Windows with LM Studio, no Docker or admin rights

Description and use

Portable Hermes Agent for Windows that runs locally with LM Studio without Docker or admin rights, featuring a GUI, permission controls, and optional extensions.

Problem

Need an AI agent on Windows that can use tools and talk to a model on the local GPU without requiring Docker, system Python, or admin rights.

Outcome

A fully portable AI agent on Windows that can run locally on an NVIDIA GPU, use a GUI, manage permissions, and extend with TTS, music, and image generation servers, all without Docker or admin rights.

Workflow

• Download portable-hermes-agent-v*.zip from releases page
• Extract to a normal user folder (e.g., C:\Users\YourName\Portable-Hermes-Agent)
• Double‑click START.bat to run setup (downloads embedded Python, dependencies, LM Studio SDK, Node.js tools)
• Launch Hermes via START.bat (GUI) or hermes.bat (CLI)
• Set up LM Studio: install LM Studio, download a model (e.g., Qwen 2.5 7B), start server on port 1234
• In Hermes, open LM Studio panel, connect to http://localhost:1234, load model, set GPU and context length, click Use for Chat
• Open Tools > Permissions and set file, network, command execution levels as desired
• (Optional) Install extensions via Tools > Extensions (TTS, Music, ComfyUI)
• Use Workflow engine or Tool maker to create pipelines or new tools
• Update portable distribution with UPDATE.bat or hermes.bat update --backup --yes; update core via agent command

Evidence

I wanted an AI agent on Windows that could use tools (files, a shell, web search, a browser) and talk to a model running on my own GPU.

Source

So I built Portable Hermes Agent. It's a portable Windows build of NousResearch/hermes-agent (MIT) with a desktop GUI, an LM Studio panel, a permissions panel, and some extra tools added on top.

Source

On first launch, START.bat runs the portable setup. It downloads embedded Python, the dependencies, the LM Studio SDK, and Node.js tools into that folder only.

Source

The manual describes the panel like this: an endpoint field (default http://localhost:1234) with a Connect button, a status indicator, a model browser, a GPU selector, and a context length setting.

Source