AI-NATIVE AGENT-BASED MODELING
Run agents.Reveal emergence.
AutoABM turns a research question into runnable agent worlds, automated experiments, and verifiable mechanism explanations. Start with micro rules. Discover how macro worlds emerge.
LIVE RESEARCH SYSTEM1,000 agents · tick 38 · deterministic seed 42
A NEW WAY TO DO ABM
Not faster coding. A faster path to discovery.
Traditional ABM work is split across model implementation, experiment scripts, visualization, and documentation. AutoABM reunites them in a Simulation-first research system — driven by AI and bounded by evidence.
PRODUCT WALKTHROUGH · 02:33
From question to evidence, one connected study.
A short product overview first explains the research loop and its core objects, then shows importable cases before entering the workbench for simulation design, live runs, deep experiments, mechanism graphs, ODD, and reproducible outputs.
00:00Product introduction00:05Research loop00:11Case library00:19Workbench walkthrough01:04Evidence-based explanation02:27Research handoff
THE SIMULATION IS THE PRODUCT
Turn complex systems into research objects.
This is not a code editor with a chat panel. Every research object has identity, version, run history, and an evidence trail.
Simulation
A versioned world
Model structure, agents, environment, parameters, and ODD evolve together as one Simulation.Run + Trace
Every run leaves evidence
State, metrics, events, and mechanism firings at every tick enter the Trace — the data root of explanation.Experiment
Automated research design
Run sweeps, interventions, and multi-seed repetitions in the same surface where results are compared and exported.Mechanism
From correlation to mechanism
Connect changes in macro curves back to micro agents, events, and causal mechanism paths.REAL-TIME SIMULATION CANVAS
Watch 1,000 agents change a system together.
Spatial, network, and continuous environments evolve in real time. Tune, run, pause, replay, and inspect metrics and agent states in the same time window.
LIVE · 1,000 AGENTS · 2 FPS
RUN CANVASSEED 42
DEEP EXPERIMENT
Turn a follow-up into an experiment.
Ask for an intervention or comparison in natural language. AutoABM builds the experiment view while keeping baselines, seeds, and metrics inspectable.
BASELINE ↔ INTERVENTION
EXPERIMENT UITRACE-LINKED
ONE QUESTION → A RESEARCH PROGRAM
Compress the research loop into one workbench.
ASK
Describe a phenomenon
Begin with an open question, an existing model, or research data.DESIGN
Compare mechanism worlds
AI proposes diverse drafts. Compare assumptions and boundaries before adoption.RUN
Make the world run
The local Mesa kernel executes deterministic Runs and continuously writes Trace.EXPERIMENT
Automate interventions
Sweep parameters, compare policies, repeat seeds, and aggregate real metrics.EXPLAIN
Trace the emergence
Every claim can return to a tick, an agent, an event, and a mechanism.REPRODUCE
Ship the full study
Export the Model, ODD, experiments, seeds, results, and Trace.More is different.
Emergence is not a metaphor. It is the object ABM must observe, experiment on, and explain. AutoABM builds an executable, replayable, verifiable path between micro rules and macro phenomena.
EVIDENCE, NOT ELOQUENCE
AI asks why. Trace shows the work.
Explanation begins with a selected Trace window, connecting agent states, mechanism firings, events, and metric changes. Facts and inferences stay visibly separate; the researcher keeps the final judgment.
- ✓Only real Runs produce metrics
- ✓Claims cite ticks and mechanisms
- ✓Unsupported content is marked inference
- ✓Every ODD aligns to a model version
MECHANISM GRAPHFIRING HEAT / TRACE
CHECKED EXPLANATIONSUPPORTED / UNSUPPORTED
AN EVIDENCE-BOUND RESEARCH SYSTEM
From hypothesis to methods, one evidence trail.
The versioned Simulation sits at the center. Modeling, execution, evidence, experiments, and documentation share one model identity; the immutable Trace defines the boundary of explanation.
AUTOABM / SYSTEM METHODVERSIONED · DETERMINISTIC · TRACE-BOUND
Research mode
Human in the loop
Researchers approve structural changes, batch experiments, and exports.Dialogue mode
Read-only explanation
Question the model, results, and ODD without changing the study.Autonomous mode
Automated research
Generate, run, compare, and report from a question — with every number tied to a real Run.AUTOABM 1.0
Build a world. Then ask it why.
The AI-native ABM research workbench for social science and complex systems. Available now for Windows.