9:00 AM
to 10:00 AM
Doors and coffee
Badge pickup at the front. Claim a seat, claim an outlet, get your laptop on the wifi. The good outlets go first.
Two days, one room
The schedule
One room, two days, and every slot now has a name on it. A few talk write-ups are still landing as we lock specifics with each speaker, so check back. The loading bar finally hit 100.
August 15
Saturday
10:00 AM
to 10:20 AM
Welcome to the Underground
What this weekend is, how the room works, and what you should walk out with on Sunday. Short on purpose so we can get to the real thing.
Stephen DeLorme10:30 AM
to 11:15 AM
Running Local Models: From Desktop to Data Center
You can run your first local model in about ten seconds. That is the easy part. The harder questions start immediately. How do you keep it private? How do you make it secure? How do you power an agent? What changes when one agent becomes ten, then a swarm? Austin walks through the full path from a first model on your laptop to a working inference cluster on hardware you control. Along the way, he covers the foundations of inference engineering: model choice, serving, runtimes, hardware, and the systems that let many agents run together. Come as a beginner or an experienced builder. Leave with a clear map of what to build next.
Austin11:15 AM
to 12:00 PM
How models get made: training and fine-tuning, end to end
Ben has been going deep on how models actually get trained, and this talk walks the whole process end to end: what pretraining does, where fine-tuning picks up, how the data and the hardware shape the result, and how the stages fit together into a model you can actually run. Less mystique, more mechanics. If a model has ever felt like a black box that arrives fully formed from somewhere else, this is the talk that takes the lid off.
Ben Carman12:00 PM
to 1:00 PM
Freedom tech grants: how HRF backs the builders
Harrison leads freedom tech grants at the Human Rights Foundation, the program that funds open-source, privacy-preserving, censorship-resistant tools for human rights defenders around the world. He walks through how the program actually works: what HRF looks for, how projects get funded, and where the AI for Individual Rights program fits into supporting the kind of open model and open agent work this room is about. If you are building freedom tech and wondering how work like yours gets supported, this is the hour to find out. Come with questions about your own project.
Harrison Friedes1:00 PM
to 2:00 PM
Lunch
Lunch is on us. Half the value of this hour is the person next to you. Ask them what they are building.
2:00 PM
to 2:45 PM
Mimir: the agent that knows everything and spends nothing
Odin gave an eye for one drink from the well of wisdom. HODLBarbarian gave up cloud convenience instead, and got a personal agent that can see his entire financial life without any of it leaving the house. The problem is familiar: your money is scattered across a dozen institutions that each show you one slice, and every tool that offers to assemble the whole picture wants your credentials sitting on its servers. This talk is the architecture of the alternative. A local model, a sealed vault, a broker that hands out one credential at a time, and watch-only wallet descriptors, arranged so the agent can know everything and touch nothing. Mimir was a severed head: all counsel, no hands. That is a security model, not just a joke. The middle of the talk is the part nobody writes up, the approval economy. Every secret costs a tap, and a system that asks you fourteen times a night trains you to stop reading the prompts. He brings the failure log with him, including six denied nights in nine days where the guard was right every time, and a fix that turned out to be half a fix and broke again the same day. Then a live demo and the uncomfortable lessons, including an agent that recorded a security finding which never happened and how that fiction survived three review passes and became fact. You will leave knowing how to give an agent read access to things that matter without giving it the ability to move a single satoshi, and how to design approvals that people do not train themselves to ignore.
HODLBarbarian2:45 PM
to 3:30 PM
Agentic AI for developers
Adewale is an ambassador at the Agentic AI Foundation, where he helps agentic AI become understandable, credible, and usable for the developers who actually have to build with it. Expect a practical session on open agents and the developer experience around them. Exact talk details land once we lock them with the speaker.
Adewale Abati3:30 PM
to 4:30 PM
LLM compression and self hosted inference
Sherif is an applied AI researcher at Sybil Solutions working on LLM compression and self hosted inference: the craft of getting capable models to run on hardware you actually own. Exact talk details land once we lock them with the speaker.
Sherif Cherfa4:30 PM
to 6:30 PM
Happy hour
Day one winds down over drinks. Pair up, unstick each other, show what you built. The talks are the excuse, this is the conference.
August 16
Sunday
9:00 AM
to 10:00 AM
Doors and coffee
Day two. Same drill: coffee, outlets, wifi. Come early if yesterday left you with something half-built.
10:00 AM
to 10:45 AM
Fireside: open AI for activists and human rights defenders
Alex builds private AI at Finite Computer, where the whole point is that using a model should not mean handing your data to somebody else. Harrison runs freedom tech grants at the Human Rights Foundation, funding the open source, privacy-preserving, censorship-resistant tools that human rights defenders actually depend on. They sit down together to talk about what happens where those two things meet: open models in the hands of activists, organizers, and defenders working under governments that would rather they had no tools at all. Where open AI is already doing real work, where it is still too fragile to trust with somebody's safety, and what the builders in this room could go and fix. Come with questions.
Alex LewinHarrison Friedes11:00 AM
to 11:45 AM
Building at the intersection of AI and Bitcoin
Nitesh is the developer behind Hunk, an AI code review tool, and works on Noah, a Bitcoin Ark wallet. Expect a session from the intersection of AI and Bitcoin, with working code over roadmap theater. Exact talk details land once we lock them with the speaker.
Nitesh B12:00 PM
to 1:00 PM
Proven correct: business logic with Lean and AI
Lean is a programming language for writing mathematical proofs, so instead of testing your software and hoping, you prove it correct. Matthias has been pairing Lean with AI to try exactly that on real business problems: building proofs for accountants, construction projects, and other domains where the books have to balance and the schedule has to hold. Come see what a theorem prover actually does and how AI changes the work of writing proofs, and leave motivated to write robust software that is mathematically proven to be correct.
Matthias Debernardini1:00 PM
to 2:00 PM
Lunch
Last chance to corner a speaker over food. They came here to be cornered.
2:00 PM
to 2:45 PM
Confidential AI Beyond Inference
Running model inference inside a trusted execution environment can protect user data while it is being processed. But private inference is only one component of a confidential AI system. Anthony, co-founder and CTO of Maple, draws on several years of operating TEE-based AI infrastructure in production to explore what it takes to protect the larger system around inference, and where local execution complements confidential remote compute.
Anthony Ronning3:00 PM
to 3:45 PM
Designing payments for people and agents
UX and agent experience for agentic payments: making money move cleanly whether a person or an agent is holding the wallet, and what changes about design when your user might be a loop.
Stephen DeLorme3:45 PM
to 4:30 PM
Buzz for startups
James is an Entrepreneur in Residence at ATDC, Georgia Tech's startup incubator, where he coaches a portfolio of tech-enabled founders. This session makes the case for Buzz, the platform from Block, as a place to actually run a startup: how a small team coordinates, ships, and works alongside agents in the open.
James Harris4:30 PM
to 5:30 PM
Lightning demos
Open mic for the room. Five minutes, your screen, something that runs. Sign up at the door during the day. No slides about what you are going to build, only things that exist.
5:30 PM
to 6:00 PM
Closing
What happened this weekend, who to stay in touch with, and how to keep building after the room empties out.
One badge, both days
Every session above is included. Move early, pay less, bring your laptop.