Show HN: Datapizza AI – Lightweight open source framework for GenAI apps (github.com)

Datapizza AI is a new lightweight, open-source Python framework for building production-ready generative AI apps—agents, RAG systems, and multi-agent workflows—focused on speed, predictability, and observability. It provides a minimal, engineer-oriented API (pip-installable modules), pluggable provider clients (OpenAI, Google Gemini, Anthropic, Mistral, Azure), and a tool/agent model that helps you compose specialized agents (e.g., weather, search, planner) that can call each other. The project emphasizes shipping reliable systems quickly while keeping debugging fast and runtime behavior transparent. Technically, Datapizza bundles an end-to-end RAG and agent stack: document parsers (Docling), splitters, chunk embedders (OpenAI text-embedding-3-small), Qdrant vectorstore integration, and a DagPipeline for chaining rewrite→embed→retrieve→generate stages. Agents use a @tool decorator and a simple Agent.run API; observability is built-in via OpenTelemetry-style ContextTracing showing spans, token counts, and latencies. Extras include DuckDuckGo search and Redis caching, configurable embedders/re-rankers, and examples using models like gpt-4o-mini. Licensed MIT and community-friendly, Datapizza targets teams that want modular, auditable GenAI systems without heavy orchestration layers.