ragbits and deep-thinking-rag
The mature, production-ready building blocks library (A) provides foundational components that could be extended or wrapped by specialized RAG pipeline implementations (B), making them complements rather than direct competitors, though B appears to be an early-stage research project rather than a stable ecosystem sibling.
About ragbits
deepsense-ai/ragbits
Building blocks for rapid development of GenAI applications
Provides modular Python packages for LLM integration (100+ models via LiteLLM), RAG pipelines with 20+ document formats, and multi-agent coordination using the A2A protocol and Model Context Protocol. Features type-safe prompt execution with Python generics, support for Qdrant/PgVector and other vector stores, Ray-based distributed document ingestion, and OpenTelemetry observability—installable as granular components (core, agents, document-search, evaluate, guardrails, chat, CLI) rather than monolithic framework.
About deep-thinking-rag
FareedKhan-dev/deep-thinking-rag
A Deep Thinking RAG Pipeline to Solve Complex Queries
Implements a multi-stage agentic RAG system that decomposes complex queries into structured research plans, then iteratively retrieves, reranks, and synthesizes evidence using supervisor agents, cross-encoders, and hybrid search strategies (vector/keyword/semantic). Built on LangChain with configurable LLM providers, it includes self-critique and policy-based control flow to decide when to refine the plan, continue research, or synthesize final answers—enabling multi-hop reasoning across both internal documents and web sources.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work