Packages & Components
Runtime packages organized by capability — Reliability, Observability, Governance, Orchestration, Evaluation, and Model Runtime — plus lower-level building blocks.
Flagship Packages
Observe
5 componentsObservability layer for LLM, tool, retrieval, policy, and agent workloads
ObservabilityService
Central service orchestrating tracers, redactors, and span lifecycle across workloads.
OpenTelemetry Tracer
Tracer adapter bridging to OpenTelemetry for production-grade distributed tracing.
InMemory Tracer
Zero-dependency tracer for test assertions. Captures spans in-process.
Default Redactor
Attribute redaction for PII, credentials, and sensitive data in spans.
Observe Decorators
Function-level @observe and @observe_tool decorators for automatic span instrumentation.
MCP
3 componentsTyped MCP client/server toolkit for interoperable tool and context integration
MCP Service
Core MCP service with transport abstraction, tool registration, and capability discovery.
Tool Registry
Typed tool registration with schema validation and execution dispatch.
Auth Hooks
Authentication and authorization hooks for MCP transport connections.
Router
3 componentsRouting engine for model/provider/workload selection with fallback and policy-aware decisioning
WorkloadRouter
Route workloads across models and providers based on capability, budget, and latency constraints.
Fallback Strategy
Configurable fallback chains with priority ordering and condition-based failover.
Capability Matcher
Match workload requirements against provider capabilities for optimal routing.
Evals
4 componentsEvaluation framework for offline datasets, retrieval scoring, tool-call evals, and regression checks
EvalService
Run evaluation suites against datasets with pluggable scorers and CI-friendly reporting.
Retrieval Scorers
Precision, recall, and relevance scoring for retrieval pipeline evaluation.
Baseline Comparisons
Compare current results against saved baselines for regression detection.
Dataset Loader
Load evaluation datasets from JSONL and structured fixtures.
Policy
4 componentsRuntime policy engine for action gating, approval requirements, evidence checks, and redaction
PolicyEngine
Core policy evaluation engine with typed decisions, violations, and audit trails.
Approval Workflows
Request and resolve approvals with evidence requirements and TTL enforcement.
Policy Packs
Versioned, composable rule collections for reusable policy configurations.
Policy Repository
In-memory and pluggable storage for policy rules and resource type filtering.
Skills
3 componentsVersioned packaging system for reusable AI skills, instruction assets, and execution metadata
SkillService
Load, validate, resolve, and register portable skill packages with typed manifests.
Skill Resolver
Resolve skill instructions and templates against execution context.
Skill Validator
Validate skill manifests and directory structures with diagnostic reporting.
Realtime
3 componentsRealtime session and event orchestration for streaming, interruption, and session lifecycles
RealtimeService
Session management and event orchestration for streaming agent interactions.
Event Pipeline
Typed event pipeline with cancellation, interruption, and lifecycle hooks.
Session State
Structured session state management with domain events and error handling.
Lower-Level Components
Core
5 componentsFoundational building blocks and utilities
Config Loader
Type-safe configuration loading with env, file, and override support.
Environment Manager
Structured environment variable resolution with validation.
Type Utilities
Runtime type guards, coercions, and assertion helpers.
Error Handling
Structured error classes, result types, and error boundary helpers.
Base Client
HTTP client foundation with retries, headers, and auth hooks.
Concurrency & Resilience
5 componentsResilience primitives for production workloads
Circuit Breaker
Trip-and-recover circuit breaker for external service calls.
Async Task Runner
Bounded concurrency task executor with backpressure and cancellation.
Rate Limiter
Token-bucket and sliding-window rate limiters for API calls.
Retry Utilities
Configurable retry logic with jitter, backoff, and predicate filtering.
Timeout Manager
Deadline propagation and timeout enforcement across async boundaries.
AI / LLM Infrastructure
7 componentsProvider-agnostic LLM infrastructure
LLM Gateway
Unified interface for LLM providers with policy hooks and telemetry.
Provider Adapters
Adapters for OpenAI, Anthropic, Ollama, and generic HTTP-JSON endpoints.
Structured Output Engine
Force structured JSON output from any LLM with schema validation.
Fallback Chain
Automatic provider fallback with configurable priority and conditions.
Batch Complete
Efficient batch completion with concurrency control and result aggregation.
LLM Caching
Semantic and exact-match caching for LLM responses.
Replay Tape
Record and replay LLM interactions for deterministic testing.
Safety / Policy / Validation
5 componentsPolicy seams, guardrails, and validation utilities
Policy Gateway
Pre/post request policy hooks with block, warn, and transform actions.
Sensitive Data Scanner
Detect and redact PII, credentials, and sensitive patterns in prompts.
JSON Repair
Heuristic JSON repair for malformed LLM output.
Prompt Fingerprint
Hash and track prompt templates for lineage and deduplication.
Input Validator
Schema-based input validation with structured error reporting.
RAG / Evaluation
4 componentsRetrieval-augmented generation evaluation tooling
RAG Evaluation Runner
Run evaluation suites against RAG pipelines with configurable scorers.
Eval Assertions
Assertion helpers for LLM output quality checks.
Retrieval Scorer
Precision/recall scoring for retrieval relevance.
Answer Quality Evaluator
LLM-as-judge evaluation for answer correctness and coherence.
Telemetry / Observability
4 componentsObservability and cost tracking for AI workloads
AI Telemetry
Structured telemetry events for LLM requests, latency, and outcomes.
Cost Ledger
Token-level cost tracking across providers and models.
Request Logger
Structured request/response logging with redaction and sampling.
Trace Exporter
Export traces to OpenTelemetry-compatible backends.
CLI Tooling
4 componentsCommand-line developer experience
Project Scaffolder
Generate new ElectriPy project skeletons with config and structure.
Config Inspector
Inspect and validate project configuration from the terminal.
Dev Commands
Local dev helpers: run, watch, reset, and fixture loading.