Open-source RAG frameworks, agents and workflows for 2024–2025

Machine-assisted translation of the archived article. Read the Chinese original.

These ten retrieval-augmented generation (RAG) frameworks were featured on GitHub in 2024. They combine information retrieval with text generation

to produce answers grounded in relevant context. Each entry below includes a short description and a link to its repository.

1. Haystack offers modular components for building question-answering and search systems.

deepset-ai/haystack

2. RAGFlow provides ready-made components and workflows for building retrieval applications.

infiniflow/ragflow

3. txtai provides semantic search, language-model workflows and document processing.

neuml/txtai

4. STORM is a Stanford research project that uses language models to research topics and write reports.

stanford-oval/storm

5. LLM App provides templates and tools for building RAG applications that work with changing data.

pathwaycom/llm-app

6. Cognita brings together tools for developing AI applications and operating them using MLOps practices.

truefoundry/cognita

7. R2R is a RAG framework focused on improving retrieval.

truefoundry/r2r

8. Neurite focuses on developer experience and rapid prototyping.

satellitecomponent/Neurite

9. FlashRAG is a lightweight RAG toolkit developed at Renmin University of China.

RUC-NLPIR/FlashRAG

10. Canopy is Pinecone’s framework for building retrieval applications that scale within its ecosystem.

pinecone-io/canopy

These frameworks give developers a choice of tools for building and testing AI applications.

This 2025 roundup features five open-source AI agent projects that had attracted large numbers of GitHub stars. They range from development platforms to multi-agent frameworks.

1. langgenius/dify combines visual AI workflows, RAG pipelines and agents in an application development platform for large language models.

langgenius/dify

2. geekan/MetaGPT is a multi-agent framework that assigns different parts of a software development task to collaborating agents.

geekan/MetaGPT

3. langflow-ai/langflow is a low-code tool for RAG and multi-agent applications. It can connect models, APIs and databases.

langflow-ai/langflow

4. All-Hands-AI/OpenHands provides tools for automating parts of the software development process.

All-Hands-AI/OpenHands

5. microsoft/autogen is Microsoft’s framework for developing agent applications. Its tools include a graphical interface for building workflows and facilities for evaluating agents.

microsoft/autogen

These projects offer different starting points for designing, building and evaluating agent applications.