Capabilities

Darkstar Causal AI

Build transparent causal graphs, test evidence, interventions, counterfactuals, sensitivity, and Studio influence-diagram decisions, learn candidate structures from data, share review links, and move validated models into licensed SDK runtimes.

Start SparkView plans

Workbench

Self-serve Causal AI workbench

Start in Spark without a payment method, then upgrade when models need cloud save, sharing, Model Discovery, Learn from Data, support, or larger limits.

  • No-card Spark entry
  • Cloud save and autosave on paid plans
  • Studio influence diagrams and public read-only sharing
  • 13-language SDK portfolio

Causal graphs

Causal graph modeling

Build directed Bayesian-network models with visible variables, edges, node states, probability tables, graph layout, evidence, and scenario outputs.

  • Directed acyclic graph authoring
  • CPT editing and validation
  • Automatic graph layout
  • Cloud documents and review links

Counterfactuals

Counterfactual analysis

Separate observed facts, intervened values, and counterfactual targets so scenario comparisons stay legible and reviewable.

  • Factual and counterfactual states
  • Intervention-aware reasoning
  • Posterior comparison workflow
  • Read-only stakeholder review

Bayesian networks

Bayesian network software

Use the browser workspace for explainable probability models, directed graph structure, uncertainty reasoning, sampling, and portable model export.

  • Bayesian belief network workspace
  • Inference and sampling
  • Portable file export
  • Licensed SDK runtime path

Sensitivity

Sensitivity analysis

Rank CPT parameters by how much an up or down perturbation moves a selected target probability under the current evidence or intervention state.

  • Target-node and target-state selection
  • Proportional CPT row redistribution
  • Baseline and perturbed posterior comparison
  • Ranked maximum absolute probability delta

Studio

Influence diagrams and VOI

Studio workspaces add decision nodes, utility nodes, expected utility, policy solving, and value-of-information ranking on top of Bayesian-network uncertainty.

  • Studio-gated decision and utility nodes
  • Expected utility and policy solving
  • Exact value-of-perfect-information ranking
  • SDK examples for influence-diagram policy solving

Causal discovery

Causal discovery

Generate candidate structures from reviewed data and Model Discovery prompts, then validate the graph and assumptions before using outputs as decision models.

  • Model Discovery prompt lifecycle
  • Continuous and discrete data
  • Structure and parameter learning
  • Constraint-based, search-and-scoring, optimization
  • 28 registered algorithms: 13 discrete and 15 continuous

Deployment

SDK deployment path

When a validated model belongs inside a product or internal service, commercial SDKs provide licensed runtime delivery outside the hosted Online plan.

  • Separate SDK licenses
  • Language, platform, and rights matrix
  • Protected artifact delivery
  • Runtime use outside the browser