Articles, how-to guides, blog posts, product launches, model trials, event notes, and weekend experiments from the Yolo-Auto team.
DeepSeek V4 Flash · 2× RTX PRO 6000 Blackwell · TP2 · 192 GB total VRAM
The custom TP2 stack worked and write-heavy throughput looked genuinely promising. Then one 180K request hit the memory ceiling on roughly $25,000 of hardware.
Goals, goal settings, loop police, and auditor checks in Yolo-Auto Desktop
Yolo-Auto Desktop took one /goals-set prompt and got Qwen3.6-35B to build a playable 3D blob game instead of spiraling on the same file.
Markdown memory, Obsidian, mem0, local vector search, docs, and MCP memory tools
A practical guide to shared memory for action agents that need to remember settled decisions, scratch work, durable preferences, and operational context without turning memory into a science project.
July 3–5, 2026 · 4th of July weekend · 72-hour build window · Slop AI game jam · 500 free requests/day for free-tier participants
A 3-day game dev contest to vibe code the most absurd, broken, hilarious AI-generated game you can make. Bad UI is allowed. Broken mechanics are allowed. Maximum slop wins.
Main-model rule: contestants must use yolo-auto3.6-35b-3a (Yolo-Auto Qwen3.6-35B-A3B; API route qwen3.6-35b-a3b) as their primary AI model.
Open-source desktop client · Windows and Mac downloads · Works with your Yolo-Auto API key
Last week we shipped a lightweight agentic chat interface for Yolo-Auto that works out of the box. No provider config files, no SDK setup, just paste your key and start using the API from a clean desktop app.