Packages

A modern Elixir client for Weaviate vector database with support for collections, objects, batch operations, GraphQL queries, and vector search. Includes health checks and friendly error messages for missing configuration.

An Elixir SDK for Claude Code - Build AI-powered CLI tools with Claude

AI Firewall and guardrails for LLM-based Elixir applications. Provides prompt injection detection, data leakage prevention, jailbreak detection, and comprehensive security guardrails.

The definitive CI/CD platform for AI Quality.

Intermediate Representation for the Crucible ML reliability ecosystem

Fairness and bias detection library for Elixir AI/ML systems. Provides comprehensive fairness metrics, bias detection algorithms, and mitigation techniques.

Asset-first data orchestration framework for Elixir/BEAM.

Comprehensive benchmarking framework for AI research. Measures latency, throughput, cost, and reliability with percentile analysis and Nx numerical computations.

Model evaluation harness for standardized benchmarking with semantic similarity, exact match, and custom metrics.

Elixir client for HuggingFace Hub—dataset/model metadata, file downloads, caching, and authentication

Production adapters and pipelines for PortfolioCore. Vector stores, graph stores, embedders, Broadway pipelines, and advanced RAG strategies.

Hexagonal architecture core for building flexible RAG systems in Elixir.

Explainable AI (XAI) tools for the Crucible framework. Includes LIME implementations, SHAP-like explanations, feature attribution, and model interpretability for local and global explanations.

Unified schema definition, validation, and JSON generation for Elixir. Sinter is a focused, high-performance schema validation library designed specifically for dynamic frameworks like DSPy. It distills complex validation APIs into a single, powerful, and unified engine.

Elixir browser automation with WSL-to-Windows support. Control visible Windows browsers from WSL.

Adversarial testing and robustness framework for AI models with 25 attacks (character/word/semantic perturbations, prompt injection, jailbreak, extraction, inversion), defenses (detection/filtering/sanitization), certified robustness metrics, and attack composition.

Experimental research framework for running AI benchmarks at scale. Provides orchestration, streaming processing with Flow/GenStage, and statistical analysis.

Multi-model ensemble prediction with voting strategies for AI reliability. Leverages BEAM parallelism for massively concurrent LLM queries.

Request hedging for tail latency reduction in distributed systems. Implements Google's 'Tail at Scale' with adaptive strategies. Reduces P99 latency by 75-96%.

The command and control layer for the AI-powered enterprise.