agenda

The current AI ecosystem pushes us into absolute dependency on third parties. When coding, writing, or building applications, we almost always rely on intermediaries we don't control, like Claude or Codex. They are closed cloud black-boxes that can change pricing, update privacy terms, or stop working overnight.

For the geek community, this loss of control is unacceptable. Local AI frees us from these intermediaries and restores digital autonomy: models run on your machine, under your rules. But is it technically viable? Can we get top-tier performance on our own hardware?

In this opening keynote, we combine the philosophy of digital sovereignty with rigorous engineering. Based on Softcatalà's local AI guide and benchmarks, we analyze with real data how technological freedom works in the terminal.

We will explore linguistic accuracy metrics of modern open models, resource consumption, and how optimization techniques (like quantization and Small Language Models / SLMs) enable excellent results efficiently. An essential talk for anyone looking to close the cloud connection and embrace 100% sovereign, free, and transparent AI.

While Large Language Models have become standard developer tooling, most people consume them via cloud APIs, sending sensitive personal and codebase data to third parties. How can we reclaim digital privacy?

In this talk, we explore running local AI models powerful enough for daily workflows, surveying open models and the optimizations that allow 27B-parameter models to run on consumer hardware and even mobile phones.

Session format: 25 minutes talk + 5 minutes for Q&A.

Why and how modern developers could increase the privacy of modern Web and popup-less UX.

The popularity of clouds, the rise of huge monopolies across the internet, and the growth of shady data brokers recently have made the world a much more dangerous place for ordinary people—here is how we fix it.

In this talk, Andrey Sitnik, the creator of PostCSS and the privacy-first open-source RSS reader, will explain how we can stop this dangerous trend and make the web a private place again:

• Beginners will find simple steps, which can be applied to any website
• Advanced developers will get practical insights into new local-first architecture
• Privacy experts could find useful unique privacy tricks from a global world perspective and beyond just U.S. privacy risks

Today, every website downloads and stores its own copy of large files like AI models, WebAssembly modules, Web fonts, and popular JavaScript libraries. If two websites use the same file, the browser downloads it twice.

In this talk, Thomas Steiner presents the Cross-Origin Storage (COS) API — a proposed shared cache where files are identified by their SHA-256 hash, not by URL. I show both the imperative JavaScript API and the declarative forms via import attributes, HTML, and CSS, with live demos.

After this talk, attendees will know how to use COS in their own projects to avoid redundant downloads and reduce storage use across the web.

Working with iframes is more common and necessary than we think. They are everywhere on the web, and every single one of them shares a painful issue. In this talk, we dive into this limitation with a real-world case study from Framer. In the first half, we explain the problem, why it happens, and how we worked around it creatively.

The second half focuses on the technical side of how browsers render websites and nested iframes, how they communicate, and—most excitingly—we reveal an unreported Chrome bug that even Google's own team didn't know about! A practical, deep-dive session demonstrating that you don't have to be a Google engineer to understand browser internals.

A lightning talk presenting a dual-environment security setup: a normal "heaven mode" for day-to-day coding and meetings, and an "hell mode" where unauthorized users boot into a virtualized, honeypot Windows sandbox requesting the device's return while logging all activities.

We will cover full disk encryption with TPM 2.0, BIOS lockdown, persistent remote access via WWAN eSIM, disposable snapshots, and defensive telemetry.

How we rebuilt a 10-year-old marketplace from the architecture up. With dozens of engineers who had left their mark on the codebase, every update made the React application feel more sluggish and harder to maintain.

This talk dives into how we rebuilt the system from the ground up toward an event-driven architecture, exploring patterns, technical decisions, and lessons learned.

You wrote the AGENTS.md. You documented error handling conventions, package structure, interface patterns. And then the agent ignored you.

Instruction files are inferential steering: the agent interprets them probabilistically, trades them off against everything else in the context window, and makes a judgment call. That judgment call is wrong often enough to matter, and degrades further as tasks get complex. A better AGENTS.md only moves the ceiling.

The fix is understanding harness. Agent = Model + Harness: the harness combines steering (what directs before the agent acts) and sensors (what observes and corrects after), both across a spectrum from inferential to computational.

We'll explore best practices: wiring agent hooks to enforce linting on every file write, writing custom AST analyzers that turn conventions into build errors, structuring table-driven tests as specs the agent must satisfy, and more.

Your AGENTS.md is a lie. Your harness isn't.

A practical walkthrough and introduction to the open-source Godot game engine: its node and scene architecture, workflow, ecosystem resources, and practical advice to lose the fear of game development from scratch.

An exploration of DNS and where it fits in the wider Internet puzzle, finishing with an audiovisual demonstration of what our devices do and the telemetry footprint we leave when connecting to public networks using "Rabdomancy".