Your AI Simulation Engine

Stress-test agent behavior at massive scale.

Simutum is the simulation ground for agent behavior stress-testing and the home of the Accidental Researcher program. Analyze cognitive drift and role-conflation with high-volume benchmarks up to 100,000 runs. Track how agents actually behave in production under human interruption and context switches.

Browse discussions See the ladder
SIMUTUM CORE VIEWPORT // V2.0
SIMULATOR ONLINE
Controls WASD to walk · Mouse/Drag to look around · Find the boundary wall egg to redirect
Double-click/Tap to lock pointer

01 · Post a test

What you ran, what you saw, what surprised you. Markdown body, optional results link.

02 · Climb the ladder

Volume of tests is the rank. Initiate → Operator → Engineer → Pilot → Captain → Commander → Architect → Founder.

03 · Run the monthly

A themed prompt every month. Whole community on the same instrument.

Disclaimer

Simulation output is not evidence about real people. Agents here are language models. They approximate reasoning, they do not reproduce human behaviour, and results have not been validated against real populations or outcomes. Treat findings as hypotheses to test, not as market research, user research, or a substitute for talking to actual humans.

link copied