Probabilistic and causal modeling

Causal AI for changing outcomes.

Explore what could happen, which actions could change it, and what past decisions can teach you—in one visual workspace with explicit models, assumptions, and results.

Start modeling freeExplore the platformEnterprise integration

Free Spark plan. No credit card required.

Darkstar visual workspace with model parameters on the left, the model canvas in the center, and analysis and results on the right
A real local Darkstar workspace using the maintained synthetic Simpson Recovery example. Inspect the model and assumptions

Three questions, one inspectable model

Move from uncertainty to action—and learn from what happened.

What is likely to happen?

Update predictions as evidence changes and keep the uncertainty visible.

What could change the outcome?

Compare supported interventions under the model’s declared causal assumptions.

What could a different decision have changed?

Examine supported counterfactual alternatives to a factual outcome without rewriting it.

A connected modeling platform

Build the model. Inspect the reasoning. Put it to work.

Start from expertise or supported data workflows, ask model-appropriate questions, and carry supported results into software. Not every model supports every operation.

Build

Build and learn models

Turn domain knowledge and supported learning workflows into explicit model structure and parameters.

Explore model families

Infer

Understand uncertainty

Use evidence updates, posterior distributions, prediction, and family-appropriate queries.

See probabilistic inference

Compare

Reason about causes and alternatives

Apply supported interventions, effect methods, diagnostics, and counterfactual analysis.

See causal analysis

Integrate

Bring reasoning into applications and agents

Embed supported capabilities with SDKs or let authorized agents request explicit model operations through MCP.

Explore Agentic Causal AI

Worked example

See why a prediction is not an action.

Simpson Recovery is a synthetic Bayesian-network example. Observe Treatment = treated, then set the same value as an intervention. The model stays fixed; the answer changes because Severity affects both treatment selection and recovery. This example does not demonstrate learning or a complete agent workflow.

See the worked exampleOpen Spark and load Simpson Recovery

Choose your next step

Explore a question—or make the reasoning reusable.

For individuals and practitioners

Explore a question. Test a change.

Open a model, edit assumptions, compare scenarios, and inspect results before deploying anything.

Start modeling free

For organizations

Put causal reasoning to work across your organization.

Connect recurring decisions to supported models, analyses, applications, services, and agent workflows.

Explore enterprise integration