Module catalogue¶

The Apogee ecosystem is 29 independently versioned Python packages plus one npm package. Generated projects depend only on the ones their use case needs — nothing here is mandatory except apogee-core.

Foundational layer¶

Package	What it is for
`apogee-core`	Base classes, contracts and HTTP abstractions for projects generated by Apogee
`apogee-ai`	Agent orchestration over LLM providers and RAG pipelines
`apogee-auth`	RBAC, MFA, audit log and session management for Apogee projects
`apogee-auth-runtime`	FastAPI/Starlette middleware that wires `apogee-auth` into a running app.
`apogee-frontend`	React SPA generator — shadcn/ui, MUI, Bootstrap, Chakra, PrimeReact
`apogee-ai-providers`	Unified LLM provider layer (OpenRouter, OpenAI, Anthropic, Bedrock, Gemini)
`apogee-ai-rag`	Clean Architecture RAG toolkit — 28 RAG variants, 60+ integrations
`apogee-ai-comunication`	Clean Architecture toolkit for AI-agent communication protocols (MCP, A2A, AG-UI, ACP, A2UI, MCP-UI)

AI-native layer¶

Package	What it is for
`apogee-ai-prompt`	Content-addressed registry for prompts with versioning, A/B comparison and golden-set replay.
`apogee-ai-eval`	Offline + online evaluation harness with G-Eval, faithfulness, bias and judge-LLM scorers.
`apogee-ai-workflow`	Long-running, durable agent workflows with HIL pause/resume.
`apogee-ai-serving`	Tier-aware router (CHEAP/BALANCED/EXPENSIVE), shadow traffic, canaries, batch inference.
`apogee-ai-guardrails`	Input/output filters: PII, prompt injection, jailbreak; commercial-grade adapters.
`apogee-ai-observability`	OpenTelemetry tracing tailored for LLM calls — token counts, model, latency, cost.
`apogee-ai-sandbox`	Run untrusted code (or agent code) in isolated sandboxes with policy gates.
`apogee-ai-memory`	Layered memory for agents: short-term, episodic, semantic and procedural.
`apogee-ai-voice`	STT, TTS, VAD, barge-in and hot-word detection for voice agents.
`apogee-ai-data`	DVC-style versioning + synthetic data + labelling + feature store, all in one package.
`apogee-ai-browser`	Agentic headless browser automation on top of Playwright.
`apogee-ai-tools`	Curated tool registry wired against `apogee-ai`'s tool-use loop.
`apogee-ai-ui`	React widgets for AI applications — also published as `@apogee/ai-ui-react` on npm.
`apogee-ai-ops`	Production runbooks for AI systems: rollouts, rollbacks, drift, retraining.
`apogee-ai-cost`	Cost-aware router that picks the cheapest model meeting a quality bar.
`apogee-ai-billing`	Multi-tenant usage metering, plan management and RLS-isolated invoicing.
`apogee-ai-connectors`	SaaS integrations for CRM and helpdesk on a shared `RestIntegrationBase`.
`apogee-ai-training`	Fine-tuning recipes (LoRA/QLoRA), distillation, RLHF/DPO entrypoints.
`apogee-ai-templates`	End-to-end project templates seeding `apogee new --template …`.
`apogee-ai-search`	Hybrid retrieval (BM25 + dense + reranker) over `apogee-core` index abstractions.

Installing¶

Each package installs on its own from PyPI:

Bash

pip install apogee-ai-prompt
pip install apogee-ai-eval
pip install apogee-ai-workflow

Heavy SDKs are always optional extras, so the base install stays small and the test suite runs offline:

Package	Extras
`apogee-core`	mongo, redis
`apogee-ai`	agno, autogen, claude-sdk, crewai, google-adk, langchain, langgraph, llamaindex, microsoft-af, openai-sdk, pydantic-ai, semantic-kernel, strands
`apogee-auth`	fastapi, jinja, runtime
`apogee-auth-runtime`	argon2, bcrypt, joserfc, pyjwt, redis
`apogee-ai-rag`	all, bge, chroma, cohere, colbert, deepeval, docling, dspy, elasticsearch, embedchain, faiss, giskard, graphrag, haystack, huggingface, instructor-embed, jina, lancedb, langchain, langfuse, langgraph, langsmith, llama-stack, llamaindex, llamaparse, marker, milvus, mlflow, mongo, neo4j, nomic, nougat, openllmetry, opensearch, otel, pdfplumber, pgvector, phoenix, pinecone, pymupdf, qdrant, ragas, redis, semantic-kernel, sentence-transformers, tika, trulens, txtai, unstructured, voyage, weave, weaviate
`apogee-ai-comunication`	a2a, acp, ag-ui, all, grpc, mcp, oauth2, otel, websocket
`apogee-ai-prompt`	all, git, mustache, sql, tiktoken
`apogee-ai-eval`	all, anthropic, deepeval, openai, ragas, trulens
`apogee-ai-workflow`	all, dbos, redis, restate, sql, temporal
`apogee-ai-serving`	all, embeddings, llamacpp, mlx, ollama, redis, vllm
`apogee-ai-guardrails`	detoxify, presidio
`apogee-ai-observability`	all, datadog, helicone, langfuse, langsmith, otel, phoenix
`apogee-ai-sandbox`	all, daytona, docker, e2b, modal
`apogee-ai-memory`	all, letta, mem0, pgvector, sqlite, zep
`apogee-ai-voice`	all, assemblyai, cartesia, deepgram, elevenlabs, livekit, openai, runway
`apogee-ai-data`	chroma, faiss, html, markdown, pdf, qdrant, sentence-transformers
`apogee-ai-browser`	html, httpx, playwright
`apogee-ai-tools`	jsonschema, pydantic
`apogee-ai-ui`	markdown, rich
`apogee-ai-ops`	docker, k8s, slack
`apogee-ai-cost`	tiktoken
`apogee-ai-billing`	stripe
`apogee-ai-connectors`	discord, httpx, slack
`apogee-ai-training`	huggingface, openai
`apogee-ai-templates`	jinja
`apogee-ai-search`	elasticsearch, opensearch

Shared conventions¶

Every package follows the same rules: Clean Architecture (domain, application, infrastructure, presentation), lazy imports for heavy SDKs, injectable transport for testability, an async-first API, and registration as an apogee CLI plugin through the apogee.plugins entry point.