InfiniFlow · 2026-04-21 · major
RAGFlow v0.25.0 — Seven Pipeline Templates, Agent Publishing, User Memory, Mobile-Ready
Ships seven built-in ingestion pipeline templates aligned to document parsers, agent publishing with sandboxed code execution, user-level memory storage, RTL language support, and mobile-compatible embedded chat pages.
RAGFlow v0.25 ships production-ready agent publishing, persistent user memory, and ingestion pipelines for seven document types.
Key specs
| GitHub stars | 79,824 |
|---|
What is it?
RAGFlow is an open-source Retrieval-Augmented Generation engine with 79k GitHub stars that combines document parsing, agentic workflows, and model-agnostic retrieval. Version 0.25.0 landed April 21, 2026 with significant additions to both the pipeline and agent capabilities.
How does it work?
Seven built-in ingestion pipeline templates map to RAGFlow's native document parsers (PDF, Word, Excel, Markdown, image, email, audio), letting operators configure ingest without custom code. Agent publishing packages an agent with sandboxed code execution so it can be deployed as a standalone service. User-level memory stores and retrieves per-user context across sessions using RAGFlow's existing vector store. The v0.25.1 patch (April 30) added lazy chunked parsing for PDFs over 50 pages, cutting memory usage on large document ingestion.
Why does it matter?
The combination of pipeline templates and agent publishing turns RAGFlow from an R&D tool into a deployable production platform. User memory brings personalization to RAG pipelines without external memory infrastructure.
Who is it for?
Teams building enterprise document QA, knowledge bases, or customer support agents on top of an open-source RAG stack.
Try it
docker pull infiniflow/ragflow:v0.25.1 && docker compose up -d