Intent
LLM gives intent. The agent proposes through typed intents; proposals may be wrong — that is fine.
Deterministic systems
meet atmospheric poetry.
Building a World OS where AI agents are first-class citizens — not tools.
Developing Khan Engine 0.8.0 Zig 0.17.0
Verified 2026-11-05 — this page holds true
An AI-native World OS for simulated interactive worlds — an embodied AI runtime that looks like a game engine.
LLM gives intent. The agent proposes through typed intents; proposals may be wrong — that is fine.
Engine gives truth. Validation runs deterministic, contract-checked simulation. Proposals either survive or are rejected.
Editor gives human control. Branch and merge for multi-agent worlds. No session context → no mutation.
Evidence gives trust. Proven, reproducible, rollback-able evidence. Nothing is believed on faith.
Each project lives on its own page — status, stack, and a world of its own. This page is itself a project, world #0.
An AI-native World OS for simulated interactive worlds — an embodied AI runtime that looks like a game engine.
Zig 0.17.0 ECS MCP 02A rainy world under simulation — mysticism, weather, and care. Itrains stays poetry; Khan Engine stays engineering.
Simulation Poetry 03This page is the first citizen of its own philosophy — deterministic, atmospheric, only own endpoints at load.
Single-file Zero depsAgents live inside verifiable physics and scenarios — not in loose text. Each has identity, capabilities, a budget, and an audit trail.
The Semantic Mirror gives them sight; the Agent Session gives them citizenship: no session context, no mutation.
Itrains — a world under simulation; poetry, weather, and care.
The negative space is part of the design: what we refuse to add defines the system as much as what we ship.
This page is the first citizen of its own philosophy.