Image 1 of 1
How do "simple statistical models" lead to AI agents that can write code, hold conversation and transform society, and how can you build one yourself?
• 22. Oct. - 10. Dec., 2026, evenings 18-21h
• One evening per week, 8-weeks, 3 hours/week
• In-Person at School of Machines in Berlin, Mitte
• Small class of participants
• Certificate of Completion
• No prior experience necessary
To book a call for more information and to have your questions answered, click here.
course
description
We are living through a strange and terrifying moment. Tools that were research curiosities five years ago are now reshaping labour, art, politics, and how we think. The chatbots in your pocket are engineered by a handful of corporations whose incentives do not align with their customers' or the planet's. But the power to use, misuse, adapt and adopt these technologies remains with us. In that spirit, you will pursue your own vision of an AI agent that runs on hardware you control, configured to your needs, with memory, access and tools you choose.
This eight-week after-hours program is a hands-on introduction to modern AI and autonomous agents. We begin by demystifying large language models (LLMs) with a recap of the history: how neural networks became LLMs and how LLMs became agents. We will ask what they actually do under the hood, and who built them and why. From there we move to practice, exploring the emerging art and science of prompt design across different models. We will test the limits of what we can ask from these systems and ask how they might fit into creative processes.
Our foundation will be Hermes Agent: an open-source AI agent framework you will install on your own private server. Unlike a corporate chatbot, Hermes is something you configure, not something you subscribe to. It has persistent memory — it remembers who you are and what you've discussed across days and weeks. It can execute code, browse the web, scrape and structure data, and connect to external services via Model Context Protocols (MCPs). It can be taught reusable skills that extend its capabilities. It can run on a schedule — autonomously generating briefings, monitoring sources, or performing tasks while you sleep. It connects to the messaging platforms you already use — Telegram, Signal, Discord — so your agent is reachable wherever you are.
You will choose how it is built, what services it interacts with, and what trade-offs you are prepared to make over cost, privacy, security, ethics, and utility. We discuss open-weight alternatives (Ollama, local inference) for those who want them but the class runs on models that are known to work well with Hermes.
By the end, you'll have a working AI agent running on your own infrastructure, configured with memory and skills, and a creative project — a web app, a bot, a generative artwork, a research tool, or whatever you choose — built in collaboration with the agent itself.
No prior technical experience is required. You will be guided through the command line, server provisioning, and agent configuration step by step. Participants with technical backgrounds will be encouraged to deepen their agent's capabilities — self-hosted models, custom tools, advanced memory systems.
In this program you’ll learn
How modern AI systems work — from neural networks to large language models
How to design effective prompts across different models
How to use the command line for creative and practical work
How to provision, secure, and maintain your own private server
How to install and configure a self-hosted AI agent (Hermes Agent)
How to connect your agent to messaging platforms (Telegram, Signal, Discord)
How agents build and use memory, skills and tools
How to use Syncthing for peer-to-peer file sync across your devices
How to build custom skills and connect agents to external tools via MCPs
How to use AI agents as collaborators for creative, critical, and technical work
How to think critically about who builds AI, who controls it, and what alternatives exist
course details
Weekly Outline
Week 0: Pre-start welcome (optional)
We’d love to see the humans taking this program connect with each other! We are offering a weekend opportunity to connect before the course begins starting on the Friday (6-9pm) before course begins and running through Saturday (11-4pm).
Week 1: Introductions and Artist Talk
We get to know each other, exchange backgrounds, interests, hopes and fears, and what brought everyone here. Meredith presents his own work and artistic practice.
This week you will:
Meet your cohort and share what drew you to the course
Get a grounding in Meredith's artistic practice and its relationship to AI
Map out what you hope to build over the coming weeks
Week 2: How We Got Here — AI from Perceptrons to Play
Lets pause, take a breath and look at how we got to where we are today. We trace the lineage from early neural networks through the transformer revolution to today's large language models. What is a token? What is attention? Why did this work now and not in 1995? We map the economic and political landscape: who built them, who controls them, and what that means for artists, activists, and citizens.
Then we get our hands dirty. We set up OpenRouter accounts and begin working directly with language models — building intuition through play across different models, exploring system prompts and the emerging craft of prompt design. Meredith shares insights from building interactive chatbot experiences for theatre and public installations, where an LLM is not a tool but a performer and prompt design becomes dramaturgy.
This week you will:
Understand how neural networks became LLMs — and what that actually means
Set up an OpenRouter account and run prompts across multiple models
Build a chatbot persona with a consistent voice and constraints — then break it
Week 3: Your Own Server — Provisioning and Security
This week we build the body. Each participant provisions their own virtual private server on Hetzner and connects it securely via Tailscale. This is your machine — no one else's code runs on it, no one else reads your research. We set up SSH keys, a basic firewall, and get comfortable navigating a remote Linux system.
We also introduce a foundational tool for the weeks ahead. The command line: the most direct interface to the machine, where we'll provision servers and manage our agents.
Before computing was a consumer product, it was a collaborative practice. The UNIX philosophy — small, composable tools, open source, text as universal interface — emerged from researchers sharing code and ideas. But running your own server also means you are responsible for its security. We discuss AI safety not as abstract existential risk, but as concrete choices: what can your agent do? What can it access? What happens if someone compromises it?
This week you will:
Get comfortable in the terminal: navigating, creating, editing, running commands
Provision your own Hetzner VPS and connect it via Tailscale
Set up SSH keys and a basic firewall
Get comfortable navigating a remote Linux system
Understand the security trade-offs of self-hosting
Week 4: Giving It a Mind — Installing Hermes Agent
You have a body. Now give it a mind. We install Hermes Agent on your VPS, configure it with an OpenRouter API key, and set up its initial memory and skill systems. Your first prompt exchange with an agent running on hardware you control is a threshold moment — it responds, it remembers, it can execute code. The prosthetic relationship begins.
An AI agent is shaped by every choice you make: which model, what it remembers, what it can do, how it speaks. This is fundamentally different from using a corporate chatbot — you are not a user, you are a configurator. We spend time actually using the agent: exploring what it can do, pushing at its limits, discovering what kind of collaborator it wants to be. What kind of relationship do you want with this entity? What should it know about you? What should it never do?
This week you will:
Install Hermes Agent and configure it with OpenRouter
Send your first prompt to an agent running on hardware you control
Explore your agent's capabilities — code, conversation, basic web use
Week 5: Connecting Your Agent — Messaging, Gateway, and Syncthing
Your agent is alive on your server. Now connect it to the world — and to your other devices. We bridge Hermes to messaging platforms — Telegram, Signal, Discord — so your agent is reachable wherever you are. We set up the Hermes gateway, enabling the agent to operate as a persistent systemd service that stays running even when you're not logged in. And we install Syncthing for peer-to-peer file synchronisation, so your files follow you from server to laptop to phone without touching a corporate cloud.
This week you will:
Connect your agent to Telegram, Signal, or Discord
Set up the gateway as a persistent systemd service
Install Syncthing for peer-to-peer file sync across your devices
Week 6: Memory and Persistence — Teaching Your Agent to Remember
Agents become useful when they remember. This week is dedicated to memory systems: how your agent retains context across conversations, how it builds knowledge over time, and how it can ingest other types of data. We explore Hermes' persistent memory and the OpenViking knowledge base for long-term indexed recall — so your agent's knowledge grows with use.
We also put the agent's web capabilities to practical use with a web scraping exercise: teaching your agent to fetch, extract, and structure data from the open web, building a research corpus it can reason about and draw from in future conversations.
Corporate AI platforms remember everything you write and use it to train their next model. But a self-hosted agent gives back some control over what persists, what gets indexed, what gets deleted. We treat memory not just as a technical feature but as a relationship: what should your agent remember? What should it forget? How does persistent memory change how you interact with it over weeks and months?
This week you will:
Configure Hermes persistent memory and understand how it works
Set up OpenViking for long-term indexed knowledge storage
Teach your agent to scrape and structure data from the web
Build a research corpus your agent can reason about and recall
Week 7: Agents in Depth — MCPs, Skills, and Autonomy
With memory in place, we go further. This week covers Model Context Protocols (MCPs) — the emerging standard for connecting agents to external tools and services — custom skills (reusable capabilities you can author for your agent), and autonomous scheduled tasks. Your agent can now do things without you: run daily briefings, monitor data sources, generate content on a schedule.
We step back and ask: what happens when agents become persistent, autonomous, and interconnected? We explore speculative futures — agents that manage your digital life, agents that negotiate with other agents, agents that outlive their creators. What does it mean to build a relationship with something that remembers you indefinitely? What are the politics of a world where some people have sophisticated AI prosthetics and others don't? What would an agent built for collective rather than individual use look like?
This week you will:
Build a custom skill that extends your agent's capabilities
Connect your agent to external services via MCPs
Set up a scheduled (cron) task so your agent acts autonomously
Week 8: Project Work
The final session is dedicated to working on your project with support from Meredith and your agent. This is open studio time — develop your prototype, debug with your agent, refine what you've built, and prepare for the final presentation.
This week you will:
Develop your project prototype with agent-assisted debugging and iteration
Prepare for the post-course showcase (date TBC)
Reflect on the agent you built and where you'll take it next
who is this
class for?
This program is for artists, designers, writers, researchers, activists, and curious citizens who want to understand modern AI and build their own tools rather than only using corporate platforms.
It is especially for people who:
Want practical, hands-on experience with AI beyond chatbots
Are interested in the politics and economics of who controls AI
Want to run their own infrastructure rather than depending on cloud services
Are drawn to building things — bots, apps, tools, artworks — with AI as collaborator
Want to learn the command line and server administration in a supportive environment
No prior technical experience is required. Participants with technical backgrounds will be encouraged to deepen their agent's capabilities — self-hosted models, custom tools, advanced workflows.
about this IRL program
Classes are 'in-person' meaning that they will be taking place at School of Machines in Berlin, Mitte.
If you would like to book a short 15 minute interview to have some questions answered before signing up, email us your availability here.
about fees
We strive to make all of our offerings accessible to as many people as possible and to counterbalance economic injustice, we have several payment tiers. While we feel this is important and necessary, it may lead to confusion, so hopefully this helps:
If you are making art full-time or are a full-time student then sign-up as an artist/student!
If you are paying VAT, for your own insurance, and working to find gigs to survive, then sign-up as a freelancer.
If you have a full-time job, have your insurance paid for by an employer, or have no financial struggle to make ends meet, then sign-up as professional.
We also offer a limited number of need-based scholarships and work study opportunities for this course. You can read more here.
For specific questions, please email us and we'll get back to you as soon as we can.
fees
Fees per Individual Course
Full-Time Artist / Student
€690
Freelancer
€740
Professional
€820
Generous Supporter
€870
Optional Add-ons
Server hosting and OpenRouter
€100
Use of space outside course hours
€125
Payments accepted via credit card, Paypal, and direct bank transfer. For alternative payment inquiries, get in touch.
Installment payments are available for the course fee (but not the optional add-ons due to Squarespace limitations).
About additional add-ons
We realize we are living in precarious times. We feel it too. Therefore, we are offering two optional purchases with your fee rather than baking it into the full cost of the program.
About server hosting and OpenRouter
This fee includes model credits (administered directly through the course, weekly budgets set per student from week 3)
About use of space outside course hours
We are offering a two-month membership for use our work spaces and tools for the duration of the full eight week program (Monday-Friday, 10am-6pm)
See membership page for more info.
meet the profs
Meredith Thomas
Creative Technologist
Meredith is an artist and creative technologist based in Berlin. He studied biomedical engineering and science communication at Imperial College London. After moving to Berlin he became interested in creative uses of technology. He has worked as a programmer and an artist to create transmedia experiences in virtual reality, for multimedia installations and for the stage. His work focuses in particular on novel uses of machine learning in creative domains and critiquing the broader technological and cultural manifestations of artificial intelligence.
Rachel Uwa
Artist, Educator
Rachel Uwa is an artist, educator, and organiser, who curates the programs of the School of Machines, Making & Make-Believe. Her work explores the intersection of art, technology, political inquiry and human connection, creating spaces where creativity and critical engagement can shape more inclusive technological futures.