Agentic simulations.

Run up to a million agents in a simulated environment, fork the run at any decision, and trace any outcome back to its cause.

Branching timeline 400 branches
t+0 FORK t+31d +75 BPS BASELINE −25 BPS

What it does.

Regulated customers ask for these most often. Use any combination of them.

Exact replay

Start from the same state and the run repeats identically on any supported hardware. Each run produces a signed record of its inputs, agent definitions, and interactions.

“What if” branching

Pause at any step, change one variable, and continue both versions. Compare them at any point to see what the change did.

Mix agent types

Combine large-language-model agents, rule-based bots, reinforcement-learning agents, and hand-written ones in the same simulation. Each agent can use a different model.

What customers simulate.

SectorScenarioPopulation
Finance How financial stress spreads between institutions under regulatory capital rules; test alternative policy responses 1.2 × 10⁴
Public health Outbreak modelling that accounts for human behaviour and hospital capacity limits 4.8 × 10⁵
Defense Supply-chain resilience in difficult operating conditions; test alternative routes and response plans 3.1 × 10³
Energy markets How prices form in an electricity market with diverse participants and intermittent renewables 2.0 × 10⁴

Writing a simulation.

# define agents and environment, then run
from bitqubic import Simulation, Agent, replay

sim = Simulation(name="macroprudential-v12", seed=0xB17Q)
sim.populate(Agent.from_manifest("cohorts/banks.yaml"), n=12_000)
sim.attach_environment("basel-iv-2026q1")

with sim.run(steps=90, parallel_branches=400) as run:
    run.fork_at(t=31, mutate={"rate_hike_bps": 75})

report = replay(run.manifest).attribute("insolvencies")
# -> signed and reproducible

Run your own scenario with us.