Packages

Command center core library for AI agent orchestration

Vector embeddings service for the NSAI ecosystem

Dataset management library for ML experiments with support for GSM8K, HumanEval, MMLU loaders and evaluation metrics (BLEU, ROUGE, F1).

Phoenix LiveView interface for sample generation and human labeling workflows. Thin wrapper around Forge (sample factory) and Anvil (labeling queue) with real-time progress updates, keyboard shortcuts, and inter-rater reliability dashboards.

Metrics aggregation service for experiment results and system health monitoring.

Synapse integration for altar_ai - SDK-backed LLM providers for multi-agent workflows. Provides unified adapter layer for Gemini, Claude, and Codex with automatic fallback, workflow actions, signal handlers, and telemetry bridging.

FlowStone integration for altar_ai - AI-powered data pipeline assets. Provides FlowStone.AI.Resource for unified AI access and FlowStone.AI.Assets DSL helpers (classify_each, enrich_each, embed_each) with telemetry bridging.

North Shore AI Registry: Distributed model registry with PubSub coordination, health monitoring, and multi-backend storage for ML reliability research.

Unified API Gateway for North Shore AI services - handles routing, authentication, rate limiting, and circuit breaking

Code Intelligence Platform built on the Portfolio RAG ecosystem. Repository analysis, semantic code search, dependency graphs, and AI-powered code understanding with multi-language support.

Interactive CLI/REPL for the NSAI ecosystem: unified interface for Crucible experiments, CNS dialectical synthesis, and ML reliability research.

Elixir SDK for Linear with committed upstream schema artifacts and generated GraphQL API reference docs.

Shared GraphQL runtime for Prismatic-based Elixir SDKs.

Lineage IR for cross-system traces, spans, artifacts, and provenance edges. Provides a shared event envelope and sink interface for consolidation across runtimes.

Service-runtime kernel for self-hosted inference backends, owning readiness, health, lease reuse, and endpoint publication above the transport seam.

TelemetryReporter is a transport-agnostic telemetry batching library for Elixir/BEAM apps. It uses Pachka for efficient size/time batch flushing, drops on overload to protect producers, and isolates encoding failures so a single bad event never poisons a batch.

NsaiWork - Unified job scheduler for the North-Shore-AI platform. Protocol-first, multi-tenant job scheduling with priority queues, resource-aware scheduling, and pluggable backend execution.

Low-level BEAM↔Python concurrency substrate. ETS-backed shared state with Python views, batched fan-out with real backpressure, and stage composition over BEAM + Python steps. Built on SnakeBridge/Snakepit.

First concrete self-hosted inference backend for llama.cpp, owning llama-server boot specs, readiness probes, stop semantics, and endpoint publication through self_hosted_inference_core.