Introduction
At the AI Resilience Hub Village, we explore the latest developments in artificial intelligence and its application in cybersecurity. Our goal is to build resilient AI models and systems that defend against modern-day threats.
After our debut in Miami, we're bringing the Village back home to Buenos Aires for the 22nd Ekoparty — three days of talks, live demos, hands-on workshops, and AI-powered chaos in the heart of the city where it all started.
3
Days
100+
Tools
22
Eko Edition
TokioAI
The philosophy behind everything.
TokioAI is the open-source AI engine that powers the AI Resilience Hub. It's not a product — it's a philosophy. A way of building AI systems that actually work in production, not just in demos.
Why TokioAI exists
The AI industry is drowning in frameworks, wrappers, and abstractions that add complexity without adding capability. Every week there's a new "AI framework" that promises to solve everything but requires you to learn its own language, its own patterns, its own ecosystem.
TokioAI was born from a simple observation: the model is the brain, not the code. The code is just the body — hands, eyes, legs. You don't need a framework to give an AI hands. You just need clean tool definitions and a loop that lets the AI think, act, observe, and repeat.
Radical minimalism. ~1,000 lines of core engine. Zero frameworks. Zero dependencies beyond the API clients. If you need a framework to use an AI model, you're doing it wrong.
Core principles
- The Model is the Brain: Don't fight the model. Don't micromanage it. Give it tools, context, and memory — then get out of its way. The intelligence is in the model, not in your orchestration code.
- Body, Not Mind: Your code is the body. Tools are hands. SSH is legs. File operations are eyes. The model decides what to do — your code just executes. Build the body, trust the brain.
- Tools, Not Frameworks: Each tool is a self-contained function. No inheritance, no base classes, no plugin systems. A tool is a function with a name, a description, and parameters. That's it. The model knows how to use it from the description alone.
- Native Tool Calling: Real structured function calls through the API — not regex parsing, not prompt hacking, not string matching. Claude, GPT, and Gemini all support native tool calling. Use it. Everything else is a hack.
- The Agent Loop: Think → Act → Observe → Learn. The AI receives a task, decides what tool to call, executes it, reads the result, and decides the next step. Up to 25 rounds of autonomous action. No human in the loop unless the AI asks for help.
- Multi-Provider, Zero Lock-in: Claude, GPT, Gemini, DeepSeek, Ollama. Same tools, same loop, same interface. Switch models with one flag. The engine doesn't care which brain is driving.
What it looks like in production
TokioAI runs in production right now with 100+ tools: shell execution, SSH into any server, file operations, network scans, Docker management, WAF control, SOAR integration, Jira, SpyCloud, Defender, robot commands, drone flight, vision analysis, Kafka consumers, database queries, and more.
One brain. Many hands. Zero frameworks.
~1K
Lines of core
100+
Tools
6
AI Providers
0
Frameworks
The evolution of AI capabilities
AI went from text generation to reasoning to tool calling to autonomous agents in three years. The models got smarter. The interfaces got dumber. Most "AI apps" are still glorified chatbots with a nice UI.
The real revolution isn't in the model — it's in what the model can do. When you give an AI the ability to execute shell commands, SSH into servers, read logs, patch code, query databases, and control hardware — it stops being a chatbot and becomes an operator.
TokioAI is built for this reality. Not for the AI of 2023. For the AI of right now — one that can think, decide, act, and learn from the result. Autonomously. In production. Under pressure.
Built from Patagonia. Open source. No VC funding. No corporate agenda. Just hackers building tools for hackers.
GITHUB → TOKIOAI
TOKIOIA.COM
Workshops & Talks
Our workshops cover a range of topics, from building robust AI systems to understanding how AI can be applied in penetration testing and defense. Speakers will share insights on securing machine learning models and detecting threats in real-time environments.
CFP is open! Submit your talk proposal for AI Resilience Hub at Ekoparty Buenos Aires 2026. Deadline: August 14, 2026. More info.
Schedule
// Schedule will be announced after CFP closes — August 2026
| Talk Title |
Speaker(s) |
Day, Time & Room |
| To be announced — Submit your talk at Call for Papers |
About AI Resilience Hub
The AI Resilience Hub Village is part of Ekoparty, bringing together experts and enthusiasts from around the world to discuss the intersection of artificial intelligence and security. Our mission is to advance the understanding and application of AI in creating more secure systems.
Born from the Latin American hacker community, the Village is a space where technical curiosity thrives — where real systems get attacked, defended, and rebuilt live on stage. No slides. No marketing. Just working AI systems facing real threats.
The Statement
This is not a demo. This is live.
The Statement is not a traditional CTF and not a sales pitch. It's a live experiment where real AI systems face real attacks, real defenses, and real consequences — all happening in front of you.
We brought TokioAI — a full autonomous security platform built from Patagonia — and we're putting it to the test on stage. Everything you see is running live: the monitoring, the attacks, the defense, the patching, the robots. Nothing is pre-recorded. Nothing is scripted.
What's happening on the floor:
- Live WAF under attack: A real Web Application Firewall with 26 signatures, behavioral rules, honeypot traps, and IP reputation scoring. We will attack it live — SQLi, XSS, path traversal, log4j, zero-day payloads — and watch TokioAI detect, classify, correlate, and block in real time. Every request. Every decision. On screen.
- Real-time monitoring dashboards: Live panels showing attack traffic, threat classification, multi-phase correlation (recon → probe → exploit → exfiltration), and autonomous blocking decisions. You see the AI think.
- Infrastructure patching in real time: When we find a vulnerability, TokioAI patches it live — SSHing into servers, editing configs, restarting services, verifying the fix. One prompt, multiple tool calls, zero human intervention.
- Robot fleet control: Robots on the floor controlled by TokioAI through natural language. Patrol commands, camera feeds, sensor telemetry, all orchestrated by the AI agent. Speak to it, and the robot moves.
- Bug hunting session: We pick a target (with authorization), point TokioAI at it, and let the AI scan, enumerate, probe, and report. Live vulnerability assessment driven by AI tool calling — not a wrapper around nmap, but an AI that decides what to scan, interprets results, and chains tools intelligently.
- WiFi defense: A live wireless defense system monitoring the airspace — detecting deauth attacks, evil twins, probe floods, and rogue access points. If someone tries something funny on the WiFi, the screen lights up.
- AI vs AI — The Trial: An attacking AI agent tries to compromise an IoT device. A defending AI agent protects it, reacting and adapting in real time. Humans can jump in with their own tools, scripts, SDRs, fuzzers, or LLMs. Whoever establishes Command & Control wins the round.
- Computer vision + face recognition: A Raspberry Pi with a Hailo AI accelerator running real-time object detection and face identification — the AI identifies who's standing at the booth and reacts accordingly.
- Drone control: DJI Tello flight control via natural language — takeoff, patrol patterns, camera feed, telemetry. Tell the AI to fly, and it flies.
The engine behind it all: TokioAI
Everything running on stage is powered by TokioAI — an open source AI engine with native tool calling. Not regex parsing. Not prompt hacking. Real structured function calls through Claude, GPT, and Gemini APIs. The AI decides what tool to use, executes it, reads the result, and decides the next step. Up to 25 rounds of autonomous action per task.
100+ tools in production: shell execution, SSH into any server, file operations, network scans, Docker management, WAF control, robot commands, drone flight, vision analysis, and more. One brain, many hands.
It's open source. You can install it right here.
Try it yourself:
- Install TokioAI at the booth: Bring your laptop, we'll help you set it up in 2 minutes. Works with any API key — Claude, GPT, Gemini, or run it free with Ollama locally.
- Attack the WAF: We'll give you a target and let you try to bypass the defenses. Can you evade the signatures and a behavioral AI? Bring your payloads.
- Control the robots: Walk up to the terminal, type a command in natural language, and watch the robot execute it. No programming required.
- Join The Trial: Just stop by the AI Resilience Hub booth and jump in. No registration needed. Bring your tools, your scripts, your creativity.
- Learn to build: We'll show you how TokioAI works under the hood — how to add tools, connect providers, and build your own AI-powered security platform.
Practical info:
- When & where: October 7–9 — Ekoparty Buenos Aires 2026 — AI Resilience Hub Village.
- Bring your laptop and your tools. SDRs, boards, frameworks, LLMs — whatever you work with.
- Rounds are timed and moderated. Safety rules enforced. Everything stays inside the Statement environment.
- No attacking external infrastructure. Play hard, play smart, play fair.
Why come?
Because you won't see this anywhere else. Not a slideshow. Not a keynote about what AI “could” do. This is an AI defending real infrastructure against real attacks, controlling real robots, patching real servers — live, in front of you. And you can be part of it.
Come break things. Come build things. Come see what happens when AI stops being a buzzword and starts being a weapon.
GET TOKIOAI → GITHUB
Call for Papers
We are accepting submissions for papers to be presented at AI Resilience Hub Village at Ekoparty Buenos Aires 2026. Whether your focus is on AI security, offensive and defensive applications, or securing machine learning models, we want to hear from you.
Topics:
- AI Applied to Cyber Defense and Attack (Purple Team AI): How AI is used both to attack and defend security infrastructures.
- Adversarial Attack Techniques: Exploiting AI Models: How adversarial attacks can alter machine learning models.
- Prompt Injection in Generative Models: Challenges and Solutions: Exploiting and defending against prompt injections.
- Attacks and Defense of AI-Powered Computer Vision Systems: Defending neural-powered computer vision systems.
- Risk Mitigation in Machine Learning: Building Robust AI Systems: Strategies for securing machine learning models.
- Ethical Use and Concerns Regarding AI in Cybersecurity: Ethical implications of AI in cybersecurity.
- Detection and Prevention of Attacks on AI Models through Self-Learning: How self-learning models can prevent future attacks.
- Prompt Injection: Crafting Challenges and Defending Against LLM Manipulation: Creating challenges and defending against prompt injection.
- Challenges of Using AI for Real-Time Vulnerability Detection: AI for continuous, real-time vulnerability detection.
- Deploying AI Systems in Offensive and Defensive Environments: Hands-on deployment of AI in real-world scenarios.
- Automating CTFs with Artificial Intelligence: How AI can automate challenges in CTFs.
- Strengthening AI Models Against Prompt Injection and Emerging Attacks: Techniques to secure AI models from new threats.
- AI for Detecting Advanced Threats in IoT Networks: Using AI to detect threats in IoT environments.
- AI vs. AI: Competitions of Intelligent Models in Simulated Attacks and Defenses: Competitive AI-driven offensive and defensive simulations.
- Security in the AI Data Supply Chain: Securing data inputs to prevent vulnerabilities in AI systems.
- Agentic AI Security: Securing autonomous AI agents, MCP tooling, and multi-agent systems.
Contact Us
Email: contact@airesiliencehub.space
