The archive · Developer & Business Tools · Strategic decision · 2023–2026
InfiniFlow bets RAG outlives the hype; RAGFlow tops GitHub 2025, Pre-A funded
InfiniFlow pivoted from its AI database to RAGFlow, an open-source RAG engine that made GitHub's 2025 fastest-growing list and drew two Pre-A rounds.
InfiniFlow (英飞流)
What the business is
RAGFlow is an open-source Retrieval-Augmented Generation engine with agent capabilities: deep document understanding, template-based chunking, grounded citations and orchestrated workflows, self-hostable or via cloud; InfiniFlow also builds Infinity, an AI-native index database that powers it.
Starting capital：Angel round led by Chenhui Ventures with Xiyu Capital (2023); two Pre-A rounds from Sinovation Ventures and Northern Light VC, disclosed together as Pre-A and totaling several million USD.
How it started
Zhang Yingfeng, a search and recommendation infra veteran, met Jin Hai, the early vector-database builder who led Milvus 1.0 at Zilliz, while both worked at Matrix Origin on an HTAP database. In 2023 they founded InfiniFlow in Shanghai (incorporated August 2023) under an 'AI-native database' thesis, raising an angel round from Chenhui Ventures and Xiyu Capital. Their first product was Infinity, an index database; RAGFlow was built from day one as its demo use case.
What happened
RAGFlow was open-sourced on 2024-04-01 and its traffic quickly dwarfed Infinity's. The team pivoted RAGFlow from demo to flagship product and repositioned Infinity as a component. The project became one of the few Chinese-team entries on GitHub's 2025 annual top-10 list, with 2,596% YoY contributor growth per Octoverse 2025. Two Pre-A rounds followed — Sinovation Ventures, then Northern Light VC, which invested at ~30-40k stars — totaling several million USD; the company restructured red-chip with a Hong Kong holding entity to prepare for overseas commercialization, and planned RAGFlow 1.0 in mid-2026 with online retrieval rewritten from Python to Go.
How it ended up
Still scaling: as of the May 2026 interview the company had closed both Pre-A rounds, completed restructuring, and was preparing its first paid SaaS and overseas GTM; the GitHub repository remained actively developed through June 2026.
Background
RAGFlow is an open-source Retrieval-Augmented Generation engine built by InfiniFlow (英飞流), a Shanghai company founded in August 2023 by Zhang Yingfeng, a search and recommendation infrastructure veteran, and Jin Hai, who led Milvus 1.0 at Zilliz. Their original thesis was an AI-native database called Infinity; RAGFlow began as its proof-of-concept, a demo of the database in the hottest use case of 2023.
When RAGFlow was open-sourced on 2024-04-01, its traffic exploded — tens of times higher than Infinity's. The co-founders made the strategic call to reverse the hierarchy: RAGFlow became the company's flagship product and the database its component. Their reasoning: a database sits too far from business problems, while RAGFlow itself is infrastructure that users can touch, see and run.
The bet that retrieval would outlast the 'RAG is dead' hype proved out in adoption. GitHub's Octoverse 2025 named RAGFlow one of the fastest-growing open-source projects by contributors, with 2,596% year-over-year contributor-engagement growth, and the 2025 GitHub annual top-10 list included RAGFlow as one of the few projects from a Chinese team. Northern Light VC invested when the repo had roughly 30-40k stars, citing its global open-source position, traffic and technical moat.
Two Pre-A rounds from Sinovation Ventures and Northern Light VC, totaling several million USD, followed, alongside a red-chip restructuring with a Hong Kong holding entity for overseas commercialization. As of mid-2026 the company planned RAGFlow 1.0, a Python-to-Go rewrite of online retrieval, and its first paid SaaS. The project remains actively developed as of September 2026.
What has to be true
- The proof-of-concept got tens of times more traction than the flagship database — market pull is hard evidence, and the team followed it.
- A database was a kilometer from business problems; RAGFlow closed most of that gap, making it the visible, self-hosting product enterprises could evaluate.
- Betting retrieval is permanent, not a passing feature, let the team stay anchored in its expertise while the ecosystem shifted toward agents and context engines.
- Open-source position and community velocity can substitute for revenue in early funding — Northern Light invested at 30-40k stars on ecosystem, traffic and moat.
What can be applied
A demo can reveal the real product: when the proof-of-concept outgrows the original, pivot around market pull — funders can value an open-source ecosystem position over current revenue.
Aftermath
As of September 2026, RAGFlow is still scaling: the GitHub repository (infiniflow/ragflow) shows active releases through 2026-06-15, and per the May 2026 interview the company had closed its two Pre-A rounds, completed a red-chip restructuring under a Hong Kong entity, and was preparing RAGFlow 1.0 plus its first paid SaaS and overseas go-to-market. The company reports GitHub Octoverse 2025 recognition and 2,596% YoY contributor growth; funding figures are per the interview and disclosed together as Pre-A.
Sources
- RAG 永不死：张颖峰和他藏在 Flow 里的执念 (RAG Never Dies: Zhang Yingfeng and the Obsession Hidden in Flow)
- RAGFlow Named Among GitHub's Fastest-Growing Open Source Projects
- infiniflow/ragflow — Open-source RAG engine
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card