Argos: a stateful simulation engine turns video models into a renderer for a persistent world
mallow610 · x · 2026-09-05
Developer whyarethis demoed Argos, a stateful simulation engine rendered through the H3 video model via Fal Director (with fal credits).
- He texted a simulated character who had to physically go upstairs and find her phone before replying — the video model didn't decide the plot; a persistent world of characters, objects, locations, and systems did
- Each character has its own cognition, memories, and goals, choosing actions autonomously while the simulation determines outcomes
- World state is real: phones have locations, messages have delivery/read states, typing takes time — move the phone before she arrives and it's gone
- Argos converts changing world states into instructions for H3 to depict, and can expose the character's thought stream
The takeaway: video models stop inventing stories and become a rendering layer over a stateful world — a reference architecture for persistent AI-simulated environments.
More from coding & agent
- Two 45-minute loops with a critic: how GPT-6 Astra iterates on 3D reconstruction — bilawalsidhu · 2026-09-05
- GPT-6 Astra rebuilds a living room in Blender from a 3D scan, no downloaded assets — bilawalsidhu · 2026-09-05
- m3e-canvas hits 2.2k GitHub stars: sketch UI in browser, export prompts for AI coding tools — benaratame · 2026-09-05
- Anthropic's Model Hardware Standard Lets Claude Agents Autonomously Run Lab Experiments — CurieuxExplorer · 2026-09-05
- Qwen3.8-27B beats the Wikipedia game in 6 clicks using Playwright — swagonflyyyy · 2026-09-05
- Claude built 95% of a Three.js cathedral, Astra fixed the rest in 1 hour — Silver-Chipmunk7744 · 2026-09-05