⭐ If Verantyx inspires you, give us a Star on GitHub — it keeps this research alive.

⚡ Verantyx

24.0%
240 / 1,000 ARC-AGI-2 tasks solved
The hardest general intelligence benchmark — solved with zero neural networks, on a single MacBook.
✓ Zero LLM calls ✓ Zero GPU ✓ 100% Symbolic ✓ Every answer verifiable
⭐ Star on GitHub 🔍 View Source Code

Your Star Keeps This Research Going

Verantyx is built by one student on a personal MacBook — no lab, no funding, no GPU cluster. Every GitHub Star directly helps this project survive and grow.

📣

Visibility = Survival

Stars push the repo into GitHub Trending and search results. More eyes → more contributors → faster progress toward AGI.

🎓

Grant Applications

Star count is one of the first things grant committees and sponsors check. Your star is a vote of confidence.

🤝

Community Signal

Every star tells other researchers: "This matters." It costs you nothing but means everything to independent research.

Free & Takes 2 Seconds

Click the ⭐ button on GitHub. That's it. You just supported cutting-edge AGI research from Kyoto, Japan.

⭐ Star Verantyx on GitHub

Why This Matters

Frontier LLMs like o3 and Grok 4 attack ARC-AGI-2 with billions of parameters, thousands of GPU-hours, and inference costs of $10–100+ per task.

Verantyx solves 240 out of 1,000 tasks at ~$0.00 per task. No API calls. No cloud. No neural networks. Every solution is a human-readable symbolic program that can be formally verified.

240
Tasks Solved
0.5s
Avg per Task
$0
Inference Cost
1
MacBook
Approach Hardware Cost / Task Interpretable?
LLM-based (o3, Grok 4) GPU cluster $10–100+ No
Verantyx MacBook (M-series) $0.00 Yes — every answer is a program

🚀 Verantyx-CLI: Personal AI with Evolving Memory

Beyond ARC-AGI research, Verantyx introduces a groundbreaking CLI tool that transforms how you interact with AI. Verantyx-CLI wraps Claude Code with a unique 6-axis Cross Structure memory system, creating an AI that evolves with every conversation.

6-Axis
Cross Structure
.jcross
Unified Language
100%
Local Storage
Learning Capacity

💡 What Makes Verantyx-CLI Different?

Traditional AI assistants forget your conversations. Verantyx-CLI remembers everything using a revolutionary Cross Structure that organizes knowledge across six dimensions:

Every interaction is stored in .jcross files — a unified language where code, data, and logic coexist. No separate databases. No JSON bloat. Just pure, human-readable Cross Structure.

Feature Traditional AI Chat Verantyx-CLI
Memory Persistence Session-only or cloud-synced 6-axis local Cross Structure
Data Format JSON, SQLite, proprietary .jcross (code + data unified)
Learning Static model updates Evolves per conversation
Privacy Cloud-dependent 100% local storage
📦 Get Verantyx-CLI on GitHub 📚 Learn .jcross Language

🇯🇵 Built by a Student in Kyoto

Verantyx is built by kofdai, a student in Kyoto, Japan, who doesn't have access to massive GPU clusters or enterprise API budgets. Instead, he uses Claude as a development partner — designing the architecture himself, then collaborating with AI to implement, test, and iterate at unprecedented speed.

One laptop. One human vision. One AI pair-programmer. Competing at the frontier of AGI research.

The engine itself is 100% symbolic — no LLM is used in the solving pipeline. AI assists only in the development process. This is what human-AI collaboration looks like when the human brings the architectural insight and the AI brings implementation velocity.

⭐ Star = Support

No donation required. No sign-up. Just click the Star button on GitHub.
It takes 2 seconds and tells the world this research matters.

⭐ Star on GitHub 💜 Sponsor Development 📂 View Code

Building AGI research costs real money — Claude API credits for development add up fast.
Even $5/month sponsorship keeps the development velocity high. Stars help with everything else.