Pasha
ptPortuguêsenEnglish

Hermes

June 8, 2026

In this practical guide, we'll cover two essential procedures for keeping your local artificial intelligence environment optimized. First, we'll see how to update Hermes Agent running via Docker Compose without losing any memory or configuration. Then, we'll set up LM Studio Headless on Arch Linux to serve the innovative Qwen3.6-35b-a3b model locally.

Part 1: Updating Hermes Agent in Docker

If you followed the standard installation structured via Docker Compose, updating Hermes Agent requires just a few simple commands. Since agent memories and data are mapped to a persistent volume on your host machine, this process is entirely safe.

Navigate to the project folder in your Linux terminal:

bash
cd ~/hermes-agent

1. Pull the Latest Image

The first step is to pull the latest official build from the image repository:

bash
docker compose pull

2. Recreate the Updated Container

This command cleanly shuts down the old instance and initializes the container based on the image you just downloaded, preserving UID/GID environment variables:

bash
HERMES_UID=$(id -u) HERMES_GID=$(id -g) docker compose up -d

3. Disk Cleanup (Optional)

To prevent the accumulation of unused ("dangling") images that waste storage unnecessarily, run:

bash
docker image prune -f

4. Initialize Internal Agent Services

Remember to restart the gateway and APIs in the background inside the new updated container so it resumes integrated operation:

bash
docker exec -d hermes hermes gateway run

Part 2: LM Studio Headless + Qwen3.6-35b-a3b on Arch Linux

LM Studio includes an official CLI (lms) that allows running the local inference engine entirely in headless mode (via terminal, ideal for servers or pure Arch Linux machines without a heavy graphical interface). We'll use the Qwen3.6-35b-a3b model, a highly efficient MoE (Mixture of Experts) with 35 billion total parameters and only 3 billion active parameters per token, ideal for tool-calling and complex agentic tasks.

1. Installing the LM Studio CLI (lms)

On Arch Linux, the cleanest way to manage the LM Studio CLI without depending on the full graphical interface is using the official command-line installer or local package manager.

Open your Arch terminal and run the LM Studio CLI installation script:

bash
curl -fsSL https://lmstudio.ai | bash

(Note: Make sure to reload your terminal with source ~/.bashrc or source ~/.zshrc to export the lms binary to your PATH.)

2. Starting the LM Studio Service

With the binary active, start the LM Studio headless server in the background:

bash
lms server start

3. Downloading the Qwen3.6-35b-a3b Model

The lms get command lets you fetch and download models directly from the Hugging Face repository via the terminal. Download the GGUF-optimized version of Qwen3.6-35b-a3b:

bash
lms get qwen/qwen3.6-35b-a3b

(Tip: If you prefer a specific quantization to save VRAM on your GPU, you can list the available options or pass the exact tag of the desired GGUF file.)

4. Loading the Model into Memory

Once the download is complete, load the model into the local inference engine:

bash
lms load qwen/qwen3.6-35b-a3b

You can monitor the loading status and memory usage at any time by running lms ps.

5. Integrating Hermes Agent with Local LM Studio

By default, the LM Studio headless server exposes an API fully compatible with the OpenAI format at http://localhost:1234/v1.

To make the Hermes container (Docker) see the LM Studio running on your Arch host, configure the endpoint using the Docker bridge internal IP:

  1. Access the Hermes setup assistant:
    bash
    docker exec -it hermes hermes setup
  2. When prompted for the custom provider or endpoint, point to:
    text
    http://docker.internal
  3. Set the model ID exactly as qwen/qwen3.6-35b-a3b.

That's it! Your Hermes Agent is now running on the latest version and leveraging the raw power of locally managed Qwen3.6 via LM Studio Headless.