An ongoing research project
A terminal.
A world to explore.
An ordinary terminal opens onto an LLM-emulated world. A scenario sets the direction; DREAMLAND builds and maintains the detail as you explore.
Find your way inFOLLOW YOUR CURIOSITY
Where will you start?
Step into a world
Follow a story through familiar commands. Every setting includes a beginner-friendly quickstart.
Choose a setting CREATEBuild and guide worlds
Use Spiritguide to shape a scenario, save a profile, or guide a running session.
Meet Spiritguide UNDERSTANDLook inside the engine
Follow the command loop, state model, enrichment agents, and the decisions behind them.
Explore the engineering EVALUATEExamine the evidence
Browse graphs, observed failures, and methods from world-building and live simulation studies.
Read the research AUTOMATETest through API & MCP
Drive sessions, compare configurations, and connect model behavior with recorded state and traces.
See the testing workflow WATCH & READFind the presentation
A place for the slides and recording, with links into the deeper technical notes.
Presentation resourcesFIVE STARTING POINTS
Choose your world.
ABOUT THE RESEARCHER
Andrew Stein
I'm exploring how language models can sustain believable computing worlds. DREAMLAND is where I share the engineering decisions, experiments, and questions that emerge.
With thanks to Langfuse for session tracing and model-call observability. How it supports the research ↗