🚀 Projects

Have an AI chat that works without an internet connection, running on your computer.

It’s the simplest and most direct proof that local works—and the foundation for all the other projects.

In the end, the model responds with Wi-Fi turned off.

Download (or confirm that it already exists) the course’s fast model with ollama pull qwen3:30b-a3b-q4_K_M.

Without the model on disk, there’s nothing to run offline.

ollama list shows the model.

Start the conversation from the terminal (ollama run ...) or through the Ollama app.

Both paths lead to the same model; choose whichever feels more comfortable.

The model answers an initial question.

Disconnect from the internet and ask another question to prove that nothing depends on the cloud.

And it’s the test that turns “I believe it” into “I saw it work.”

The response comes through even without a network connection.

Create a “branch” (branch/fork) of a chat to explore two responses from the same point.

Lets you compare options without losing the original conversation.

There are two conversation threads branching from the same message.

A completely yours AI chat: private, free, and works anywhere.

It’s the course’s first “this is mine”—the foundation of the Project 2 agent.

Answers offline, at no cost, without sending data outside.

Run the Hermes Agent end to end using only the local model, in Vault mode (isolated from the internet).

It’s the course’s “holy grail”: an agent that’s capable, 100% private, and costs $0.

The agent completes a task with the internet disconnected.

Confirm that the 64k context model created in Module 2.4 is available in Ollama.

Hermes requires 64k of context; a smaller model won’t work for the agent.

ollama list shows qwen3-coder-64k.

In the Hermes selector, choose the Ollama model; it appears in the lower-right corner.

It’s the step that makes the agent “think” with your local model instead of the cloud.

The local model’s name appears selected in the interface.

Enable Vault mode, which isolates the agent from the network — like unplugging the internet cable.

It’s the technical guarantee that nothing leaks: the agent has no way out.

Hermes indicates that it’s in Vault / offline.

Give the agent a concrete task (summarize, organize, draft) and watch it use tools.

Proof that the agent doesn't just chat—it takes action, even offline.

The task finishes with a useful result.

Check that everything happened locally and no data went to the internet.

It’s the confirmation that makes the agent usable with sensitive data.

Everything worked with the network disconnected, in Vault.

Turn the generic agent into a system that feels like yours: what it remembers, how it acts, and what it can do.

It’s what separates a “chatbot” from a personal AI OS.

The agent remembers something about you and acts according to the defined persona.

Enable persistent memory so the agent can remember facts between conversations.

Without memory, you repeat everything each time; with it, the agent grows with you.

In a new session, it remembers something said earlier.

Set the agent's behavior and tone (formal, direct, teacherly...).

The persona makes the agent fit your way of working.

Responses change tone based on the selected persona.

Plug in a specific skill that the agent can then execute.

Skills expand what the agent can do without switching models.

The skill appears as available, and the agent uses it when asked.

Let the agent offer suggestions on its own, based on memory and skills.

It’s the “agent that helps you grow”—it proposes, not just responds.

The agent suggests something useful without you asking.

An agent that remembers, acts as your persona, and has its own skills—your AI OS.

It’s the foundation for the connection and automation projects that follow.

Memory + persona + skill working together.

Connect the agent to your information sources—code and documents—so it can work with them.

And that’s what takes the agent out of the void and puts it to work on YOUR material.

The agent responds based on something found only in your repo/doc.

Add the GitHub connection so the agent can read and understand a repository.

Code is the classic use case for a private local agent.

The agent lists/describes files in the connected repo.

Point the agent to your documents so it can consult their contents.

Lets you ask questions about PDFs, notes, and reports without sending anything to the cloud.

The agent cites something that appears only in the document.

Ask a real question (summarize a file, find a bug, explain a function).

And it’s where the connection turns into concrete value.

The response uses the source’s actual content.

Ensure sensitive code and documents are processed in Vault mode.

Proprietary data must not leak; Vault ensures that.

Work with the sources happens offline, without a network connection.

An agent that works with your code and documents while preserving your privacy.

It’s the agent moving from “general-purpose” to becoming a work tool.

Useful answers based on your sources, in Vault.

Have agents running tasks in the background, on a schedule, all the time.

It’s a unique advantage of running locally: in the cloud, this would be expensive.

There’s a scheduled task that runs on its own.

Use hermes cron for programming when the agent runs a task.

It’s the command that starts recurring automation.

The task appears as scheduled and runs on time.

Because the model runs locally, each execution has no per-token charge.

Changes the economics: you can leave it running without worrying about the bill.

No charges appear for using the agent.

Run scheduled agents in Vault mode, without sending data outside.

Continuous automation + total privacy is a rare combination.

Scheduled tasks run offline.

Monitor health with hermes status e hermes doctor.

Automation without monitoring is a trap; these commands alert you to problems.

The commands show that the agents are healthy.

Agents working 24/7, private and at zero cost, monitored by you.

And it’s the kind of leverage that only cheap local use makes possible.

The scheduled task produces results at no cost.

Choose which mode to use for each task based on data sensitivity and the quality required.

And that’s what gets you out of “all local” or “all cloud” and gives you the best of both.

You can tell which mode each task should use.

Send tasks with sensitive data to Vault mode, isolated from the network.

It’s the non-negotiable rule for customer, health, or IP data.

The sensitive task runs offline.

Use Connected mode when you need more power without going all the way to the cloud.

It’s the balance for mid-level tasks.

The task runs in Connected with good results.

Turn on the cloud for a difficult task that requires a frontier model—and has no sensitive data.

No ideology: the cloud delivers better quality when privacy isn’t the priority.

The difficult task gets better in Cloud.

Switch modes using natural language by telling the agent to send the task to the right mode.

It’s the smooth, dynamic workflow, without having to change settings at every step.

The agent switches modes when asked.

A workflow that routes each task to the ideal mode, with a clear decision rule.

It’s the practical summary of the “best tool for each task” philosophy.

You know which mode to use for each type of task.

Talk to your local agent from your phone, wherever you are, without losing privacy.

And that’s what turns the agent from a “desktop thing” into an always-available assistant.

You send a message from your phone, and the agent replies.

Use Telegram as the entry point to chat with the Hermes agent.

It’s the simplest way to talk to the agent without opening the desktop.

Telegram messages reach the agent.

Activate the agent running at home even when you’re out, traveling, or somewhere else.

The brain stays on your machine; you just send the request.

Works when you’re away from home.

Ensure processing stays on your machine; the phone is just the remote control.

Mobility doesn't have to cost you your privacy.

The model remains your home local model.

Points to watch so remote access stays secure (network, who can talk to the agent).

Access from anywhere requires care about who has access.

Only you can trigger the agent.

The agent in your pocket—and the course wrap-up, with a recap of the three tracks.

Brings the journey together: fundamentals, hands-on work, and projects, all connected.

You have a local, private AI OS that you can access even from your phone.