01What is Gemma 4 and how it works

Gemma 4 is Google's new open-source AI model that you can run locally for free. It works as the "brain" inside Hermes Agent, an AI agent that acts on your machine. This combination gives you a free, local AI agent that can automate tasks without cloud costs.

Gemma 4 comes in sizes like 12B and 26B. The 12B model is new and designed for agentic reasoning, meaning it can plan and execute tasks. Hermes Agent is the "body" that runs 24/7, while Gemma 4 provides the intelligence.

Together, they can handle everything from daily briefs to building apps.

Because it runs locally, your data stays private, and you can use it offline. This is a huge difference from cloud models like Claude, which need an internet connection.

02Key features of Gemma 4 and Hermes Agent

Gemma 4 offers several features that make it powerful yet affordable:

  • Free and local: No token costs or subscription fees.
  • Offline capability: Works without Wi-Fi, perfect for travel.
  • Laptop-ready: Runs on just 16GB of VRAM.
  • Open-source: You can customize and use it freely.

Hermes Agent adds the ability to act: it can manage files, run background tasks, use browser automation, and schedule autonomous work. It also integrates with other tools like OpenCore, Codex, Cloud Code, and OpenCode.

With the new web UI, you can manage everything in one dashboard, including models, tasks, logs, and sessions.

03How to set up Gemma 4 with Hermes Agent

Getting started with Gemma 4 and Hermes Agent is simple. Follow these steps:

  1. Install Ollama on your machine. Go to the Ollama website and download the latest version.
  2. Run the command to launch Hermes with Gemma 4: ollama launch Hermes model Gemma 4. This will set up the model.
  3. Verify the model is ready by checking the terminal message.
  4. Use the web UI (optional): Go to the models section, change the main model, and select Ollama with Gemma 4.
  5. Configure as sub-agent (optional): In the web UI, set your main model to something like GPT-3.7 Flash, and then set Gemma 4 as the sub-agent for auxiliary tasks.
  6. Or use a free API: If you don't have local hardware, choose Gemma 4 26B or 31B from OpenRouter in the model list.

04How to use Gemma 4 with Hermes Agent

Once set up, you can use Gemma 4 with Hermes Agent for various tasks. The episode shows several examples:

  • Morning brief: Schedule a 7 a.m. brief that summarizes overnight events and daily priorities.
  • Email triage: Have Hermes review and organize your inboxes.
  • Weekly review: Connect Hermes to your Obsidian vault for memory and schedule a weekly review of your notes.
  • Research and content ideas: Point it at a topic and let it research or generate ideas.
  • Build mini apps: It can create games, animations, timers, and color palettes.

You can also schedule tasks in the dashboard under "Schedule tasks" and manage everything in one place. For direct chats, use the web UI or terminal.

05Results, tests, and benchmarks with Gemma 4

In the episode, Julian ran tests with Gemma 4 for general day-to-day tasks. He says it's "not bad," but honest that it's "not amazing at writing." However, it excelled in creating functional tools like a snake game and animations.

Benchmarks show Gemma 4 uses advanced reasoning and is laptop-ready with 16GB of VRAM. Compared to larger models like the 27B, it performs well for its size, offering near-quality results of models twice its size.

One key comparison: Claude Opus is far more powerful, but Gemma 4 is free, local, and offline-capable. For simple or auxiliary tasks, the free model saves tokens.

06Tips, limits, and mistakes to avoid with Gemma 4

When using Gemma 4, keep these tips and warnings in mind:

  • Use as a sub-agent: Since Gemma 4 is fast but not a super reasoner, use it for smaller tasks to save tokens.
  • Check hardware: Ensure your machine has at least 16GB of VRAM for smooth local running.
  • Be patient with writing: It's not the best at creative writing; use it for functional tasks.
  • Use offline: One big advantage is offline use, so plan for flights or no-Wi-Fi times.
  • Avoid using it for complex reasoning: Stick to simpler agentic tasks.

Mistakes to avoid include not configuring the main model correctly in the web UI, and forgetting to schedule tasks if you want automation.

07Who Gemma 4 is for and why it matters

Gemma 4 is perfect for developers, hobbyists, and small businesses who want free AI automation without cloud reliance. It's also great for privacy-conscious users who don't want their data sent to data centers.

This matters because it democratizes AI: you don't need a huge budget to run a capable agent. With Hermes Agent, you can automate daily work like inbox management, research, and content creation for free.

It also opens doors for offline use, making AI available on planes, in remote areas, or during outages. The community aspect, like the AI Profit Boarding, offers training and support for beginners.