From aa29d210fb9a9ea0436c0e631ae787958caa2725 Mon Sep 17 00:00:00 2001 From: chengke <404835780@qq.com> Date: Thu, 14 May 2026 23:27:58 +0800 Subject: [PATCH] docs: add What We Are heading, convert intro from blockquote to normal text --- README.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 709ffa770..57e54f7f8 100644 --- a/README.md +++ b/README.md @@ -29,9 +29,11 @@ 🖥️ Dashboard

-> **We're not building the next document parser — we're building document memory infrastructure that agents can actually consume.** +## What We Are ->Knowhere is the open-source infrastructure that turns unstructured documents into persistent, navigable memory for AI agents. It automates the full pipeline — parsing, hierarchy extraction, multi-modal structuring, and graph construction — so your agents get structured, high-quality context optimized for *Agentic RAG*, *traditional vector-based RAG*, and *any LLM-powered workflow*. +**We're not building the next document parser — we're building document memory infrastructure that agents can actually consume.** + +Knowhere is the open-source infrastructure that turns unstructured documents into persistent, navigable memory for AI agents. It automates the full pipeline — parsing, hierarchy extraction, multi-modal structuring, and graph construction — so your agents get structured, high-quality context optimized for *Agentic RAG*, *traditional vector-based RAG*, and *any LLM-powered workflow*. > [!TIP] > Knowhere stands on the shoulders of excellent open-source parsers (MinerU, Docling, Marker, and others) — and actively improves upon them. We contribute **memory-oriented parsing optimizations** that fix real-world deficiencies in existing tools, and we're pursuing **agentic parsing** to push accuracy even further. Beyond parsing, Knowhere focuses on what turns raw extraction into usable agent memory: **reconstructing hierarchical structure**, **extracting and normalizing multi-modal assets**, **building navigable cross-document graphs**, and **persisting it all as long-term, citable memory** that AI agents can search, navigate, and reason over.