Compute Core
ECS world, deterministic simulation, physics, render, audio. The truth the engine commits to.
Project / Khan-Engine
An AI-native World OS for simulated interactive worlds — an operating system in which agents perceive the world through semantic interfaces, act through typed intents, verify results through deterministic simulation, and accumulate experience through observable evidence. An embodied AI runtime that looks like a game engine. Agents are first-class citizens, not tools.
💮 formula: LLM gives intent. Engine gives truth. Editor gives human control. Evidence gives trust.
01 / Positioning
Khan Engine is not codegen bolted onto a renderer. It is a domain World OS — the engine is the runtime, the verification, and the trust layer where agents live inside a verifiable world.
02 / Architecture
GPC gives the agent a body. BDI gives the agent a mind. GRACE contracts give the world its laws. Semantic Mirror gives sight. Agent Session gives citizenship. Physics gives the world its physics.
ECS world, deterministic simulation, physics, render, audio. The truth the engine commits to.
Editor, branch/merge, capability safety, pre-execution gate. The human control surface.
Semantic interfaces, typed intents, MCP gateway, agent citizenship. How agents live in the world.
03 / Core loop
Intent is generative and may be wrong. Validation is deterministic, contractual, checkable. Evidence is provable, reproducible, rollback-safe.
04 / Technology
05 / Log element
Latest verified entries from /log.json, filtered to world=khan-engine. The full log lives at /log.html?project=khan-engine.