Build and run open simulations anywhere

Every paradigm on one canvas

Mix queues, populations and stocks in one model.

One document, three views

Diagram, Stage and Code show the same OSDL document.

Diagram Stage Code

Concept render. The Stage view is in development.

Streamed results and interactive simulations

Charts, the diagram and the event stream update as the run plays. Pause, step, scrub or update a parameter at any point.

Collaborate with an AI Assistant

Ask the Assistant to edit a model and run an experiment, or connect Claude or ChatGPT over MCP.

The Assistant and MCP need an account. Invitation-only beta.

Assistant · Set Couriers capacity

Edit history

Set Couriers capacityAssistant Undo

InspectAssistant

Orders back up at lunch. Try one more courier and tell me what changes.

Selected: Couriers

Assistant

1. DoneRead the couriers parameterget_model
2. DoneAdd a third courieredit_model
3. DoneNo problemsvalidate_model
4. DoneOne day, completedrun_experiment

With three couriers the bakery delivers 30 orders instead of 28, and next-day orders fall from 7 to 5. Couriers are busy 51% of the time, down from 73%.

Selected: Couriers

Ask the AssistantSend

One document, many tools

{
  "id": "doctors",
  "type": "des.server",
  "label": "Doctors",
  "params": {
    "serviceTime": {
      "dist": "exponential",
      "rate": { "expr": "1 / treatmentTime" }
    },
    "capacity": { "expr": "effectiveDoctors.value" }
  }
},

{
  "id": "effectiveDoctors",
  "type": "sd.auxiliary",
  "label": "Effective doctors",
  "params": {
    "value": {
      "expr": "max(1, round(baseDoctors * (1 - 0.5 * fatigue.value)))"
    }
  }
}

from simcraft import Model, des

m = Model("mm1-queue")
arrivals = m.add(des.Source(id="arrivals", interarrival=des.Exponential(rate=0.9)))
waiting  = m.add(des.Queue(id="waiting", discipline="fifo"))
teller   = m.add(des.Server(id="teller", service_time=des.Exponential(rate=1.0)))
done     = m.add(des.Sink(id="done"))

arrivals >> waiting >> teller >> done

m.save("mm1-queue.osdl.json")
results = m.run(duration=10_000, replications=30)

{
  "model_id": "5248dc3e-01a9-4e28-bcaf-9e0d578a2575",
  "label": "Add a courier shelf",
  "ops": [
    {
      "op": "add_component",
      "component": { "id": "shelf", "type": "des.queue" },
      "near": "dispatch"
    },
    { "op": "connect", "from": "dispatch.out", "to": "shelf.in" }
  ]
}

$ simcraft validate clinic-hybrid.osdl.json
clinic-hybrid.osdl.json: valid (schema + diagnostics + buildable, 7 components)

$ simcraft run clinic-hybrid.osdl.json --seed 42 --events a.ndjson > a.json
$ simcraft run clinic-hybrid.osdl.json --seed 42 --events b.ndjson > b.json
$ shasum -a 256 a.ndjson b.ndjson
e8ad4d777536259434fb152974dbe5659db5a8cc59d311d7b11955d8501f189a  a.ndjson
e8ad4d777536259434fb152974dbe5659db5a8cc59d311d7b11955d8501f189a  b.ndjson

const sim = Simulation.fromEngine(new SimcraftEngine(), document);

sim.events.on("entity.moved", (event) => {
  console.log(event.time, event.from, event.to);
});

const results = await sim.run({ experiment: "one-week", seed: 42 });

{"seq":2000,"time":27.087630178280868,"type":"entity.moved","source":"triage","payload":{"entityId":"patients#62","from":"patients","to":"triage"}}
{"seq":2001,"time":27.087630178280868,"type":"state.changed","source":"patients","payload":{"path":"patients.held","value":0.0}}
{"seq":2002,"time":27.100000000000115,"type":"metric.updated","source":"fatigue","payload":{"name":"fatigue.value","value":0.24883315774918094}}
{"seq":2005,"time":27.154559380806752,"type":"state.changed","source":"doctors","payload":{"path":"doctors.busy","value":2.0}}

/-- Dispatch one calendar entry and its post-event checks: exactly one
calendar-dispatch semantic unit. -/
def stepOnce (state : MachineState) : StepDisposition :=
  match state.popNextCalendarEntry with
  | none => .completed .quiescent state
  | some (entry, popped) =>
    if state.pastEndTime entry.time then .completed .endTime state
    else
      let state := { popped with time := entry.time }
      let step := dispatchCalendarTarget entry.target state

Live in your browser

Both models run on the Simcraft engine, compiled to WebAssembly.

• 

  A weekday on the Tube

  Published Tube service and central London bus timetables with simulated passenger demand. Not a live transport feed.

• 

  The Kessler cascade

  How orbital collisions create debris and raise the risk of later collisions. An illustrative scenario, not an orbital forecast.

Powered by TfL Open Data. Contains OS data © Crown copyright and database rights 2016. Track alignment © OpenStreetMap contributors, ODbL. NASA Blue Marble and Black Marble imagery

• Repeatable traces

  Same document, experiment and seed give the same event stream.

• Live branching

  Change a parameter mid-run and the run forks a branch.

• Collaborative AI

  The Assistant in the IDE and connected tools over MCP edit your model with you. Invitation-only beta.

• Conformance-tested kernel

  Engine traces are compared with a Lean reference machine.

• Clean Git diffs

  Models are plain JSON, so changes review like code.

• Portable component libraries

  Package your own components as WebAssembly for the CLI and server. Not yet public.

• Open format

  OSDL is Apache-2.0, with published schemas and a conformance suite.

• Portable engine

  The same model runs in the browser, on the CLI or on a server. CLI and server not yet public.

• Python integrations

  Dagster, MLflow and ClearML through the Python SDK. Not yet published.