The most complete causal AI library for TypeScript — 6 modular packages for anomaly detection, causal discovery, root cause analysis, effect estimation, counterfactual reasoning, and visualization. Enterprise-grade security, CI-verified quality, browser + Node.js + PostgreSQL + Neo4j storage.
| Feature | DoWhy (Python) | causal-js (TS) | Causality Analyzer |
|---|---|---|---|
| Causal Discovery | ✗ | 10 algos | 10 algos (PC/FCI/GES/LiNGAM/NOTEARS/Grow-Shrink/KCI/RCD/CD-NOD/MVPC/GRaSP/TS-ICD) |
| Causal Inference | ✓ | ✗ | Backdoor (5 variants)/IV/PS/DR/Frontdoor |
| Root Cause Analysis | ✗ | ✗ | 4 RCA + CIRCA pipeline + OnlinePC streaming |
| do-Calculus | ✓ | ✗ | Full recursive ID algorithm + hedge criterion |
| SCM + Counterfactuals | ✓ | ✗ | ANM/PN + auto-assign + abduction framework |
| Bayesian Networks | ✗ | ✗ | 5 inference engines + Dirichlet learning |
| Sensitivity + Refutation | ✓ | ✗ | E-value + partial R² + 5 refuters (Bootstrap/Placebo/DataSubset/RandomCommonCause/DummyOutcome) |
| Enterprise Security | ✗ | ✗ | mTLS + Bearer Token + AES-256-GCM + audit trail |
| Web Components | ✗ | ✗ | Lit 3 + Canvas2D + ARIA |
| HTTP API + Docker | ✗ | ✗ | 11 endpoints (OpenAPI 3.1) + compose (PostgreSQL + Neo4j) |
| Browser Storage | ✗ | ✗ | WASM SQLite + OPFS (offline persistence) |
| Streaming | ✗ | ✗ | Sliding window OnlinePC + drift detection |
| Model Serialization | ✗ | ✗ | CausalGraph + SCM + FFNMechanism JSON |
| Graph Similarity | ✗ | ✗ | Causal structural fingerprint + Cosine |
| CI Quality | Basic | None | Lint(6 pkgs) + Build(6) + Test(1746) + Playwright E2E + Neo4j |
npm install @agentix-e/causality-analyzer-core @agentix-e/causality-analyzer-pipelineimport { CausalGraph, HeuristicPathRCA } from '@agentix-e/causality-analyzer-pipeline';
import { Matrix } from 'ml-matrix';
const graph = new CausalGraph(['Memory', 'CPU', 'Latency']);
graph.addEdge('Memory', 'CPU');
graph.addEdge('CPU', 'Latency');
const data = new Matrix(100, 3); // your metrics
const rca = new HeuristicPathRCA();
rca.train(graph, new Set(['CPU', 'Latency']), data);
const result = rca.findRootCauses(['CPU', 'Latency']);
console.log(result.rootCauses[0].name); // "Memory"import { pcAlgorithm, gesAlgorithm, notearsAlgorithm, directLiNGAM, fciAlgorithm, kciTest } from '@agentix-e/causality-analyzer-pipeline';
import { Matrix } from 'ml-matrix';
// PC algorithm (constraint-based)
const { graph } = pcAlgorithm(data, ['X', 'Y', 'Z']);
// NOTEARS (neural/continuous optimization)
const { graph: dag } = notearsAlgorithm(rawData, ['X', 'Y', 'Z']);
// GES (score-based)
const dag2 = gesAlgorithm(data, ['X', 'Y', 'Z']);
// LiNGAM (non-Gaussian ICA)
const { graph: dag3 } = directLiNGAM(data, ['X', 'Y', 'Z']);import { adjustBackdoor, findBackdoorSet } from '@agentix-e/causality-analyzer-pipeline';
const adj = findBackdoorSet(graph, 'Treatment', 'Outcome'); // {Confounder}
const { ate, se } = adjustBackdoor(graph, 'Treatment', 'Outcome', data, nodeIndex);
console.log(`ATE = ${ate.toFixed(3)} ± ${(se * 1.96).toFixed(3)}`);import { computeEValue, computePartialR2 } from '@agentix-e/causality-analyzer-pipeline';
const eValue = computeEValue(ate, se); // "How strong must an unmeasured confounder be?"
const r2 = computePartialR2(ate, se, n); // "How much variance would it explain?"npx causal-analyzer serve --port 3000
# GET /health /ready /live /metrics
# POST /discover /analyze /estimatedocker compose up -d # pipeline + PostgreSQL + Neo4jCausal Discovery: PC (stable), FCI (R1-R10), BOSS (permutation+GST, NeurIPS 2023), GFCI (hybrid PAG), RFCI (fast PAG), GES (BIC), LiNGAM (non-Gaussian), NOTEARS (neural+L-BFGS), Grow-Shrink, KCI (kernel), RCD (hybrid BIC), CD-NOD (domain shifts), MVPC (missing values), GRaSP (L1-regularized), TS-ICD (time-series), targeted discovery, OnlinePC (streaming)
Root Cause Analysis: HeuristicPathRCA, RandomWalkRCA, HTRCA, FPGrowthRCA, CIRCA pipeline, Shapley attribution, FusionAnalyzer (weighted/nested/voting)
Causal Inference: Backdoor adjustment (5 variants: minimal/maximal/efficient/exhaustive/mincost), Frontdoor, IV/2SLS, Propensity Score (IRLS), PS Matching, Doubly Robust (O(n) performance), CATE, IPW, Mediation (Baron-Kenny)
Sensitivity + Refutation: E-value (Cohen's d conversion), Partial R² (Cinelli & Hazlett 2020), Robustness value, Bootstrap, Placebo treatment, Data subset, Random Common Cause, Dummy Outcome refutation
do-Calculus: Full recursive ID Algorithm (Shpitser & Pearl 2006), c-component decomposition, hedge criterion, Pearl's 3 rules
SCM + GCM: Additive noise, PostNonlinear, auto-assign mechanisms, counterfactuals, Shapley RCA, mechanism change detection, graph falsification
Bayesian Networks: Variable Elimination, Junction Tree, Loopy BP, Likelihood Weighting, Gibbs Sampling, online Dirichlet learning, brute-force oracle
Uplift Modeling: Qini curve, AUUC (normalized), uplift@k, model comparison
Stability + Robustness: Stability Selection (bootstrap+edge threshold), StARS (auto regularization selection)
Infrastructure: Audit trail (SHA-256), AES-256-GCM encryption, Prometheus metrics, Rate limiter, mTLS, L-BFGS/Adam optimizers, structured error hierarchy
CI Tests: Fisher Z, Chi-Square, G-Square + Kernel CI (KCI)
Streaming + Drift: OnlinePC sliding-window discovery, Welford incremental covariance, stability scoring, graph drift detection (SHD-based)
Visualization: Canvas2D causal DAG (directed edges + arrowheads), uPlot time series, Lit 3 Web Components, screen-reader ARIA support, keyboard navigation
| Package | npm | Description |
|---|---|---|
core |
 |
Types, interfaces, math, ColumnarTable, graph-similarity, OTel |
pipeline |
 |
32 algorithms, RCA, inference, GCM, HTTP API (mTLS+OpenAPI) |
storage-embed |
 |
Node.js: node:sqlite + OverGraph |
storage-browser |
 |
Browser: WASM SQLite + OPFS |
storage-remote |
 |
PostgreSQL + Neo4j (mTLS) |
visual |
 |
Lit 3 Web Components + Canvas2D |
pnpm install
pnpm run --filter @agentix-e/causality-analyzer-core build
pnpm -r test # 1746 tests (292 core + 1233 pipeline + 33 embed + 20 browser + 52 remote + 112 visual)
pnpm -r lint # ESLint, 6 packages, 0 errors| Algorithm | Paper |
|---|---|
| PC | Spirtes, Glymour & Scheines (2000). Causation, Prediction, and Search |
| FCI | Zhang (2008). On the completeness of orientation rules for causal discovery |
| NOTEARS | Zheng et al. (NeurIPS 2018). DAGs with NOTEARS |
| ID Algorithm | Shpitser & Pearl (2006). Identification of Joint Interventional Distributions |
| CIRCA | Li et al. (KDD 2022). Causal Inference-Based Root Cause Analysis |
| SPOT | Siffer et al. (KDD 2017). Anomaly Detection in Streams with Extreme Value Theory |
| GES | Chickering (2002). Optimal Structure Identification With Greedy Search |
| LiNGAM | Shimizu et al. (JMLR 2006). A Linear Non-Gaussian Acyclic Model |
| GRaSP | Lam et al. (UAI 2022). Greedy Relaxations of the Sparsity Penalty |
| TS-ICD | Rohekar et al. (ICML 2023). Time-Series Iterative Causal Discovery |
| MVPC | Tu et al. (2019). Causal Discovery in the Presence of Missing Data |
| RCD | Ma et al. (2022). Reinforced Causal Discovery |
| CD-NOD | Zhang et al. (2017). Causal Discovery from Nonstationary Data |
| E-value | VanderWeele & Ding (2017). Sensitivity Analysis in Observational Research |
| DoWhy | py-why/dowhy |
| causal-js | Kanaries/causal-js |
| Intel Causal Lab | IntelLabs/causality-lab |
MIT