OpenCrab

OpenCrab is an ontology-centered Graph RAG platform that transforms messy documents into reusable AI knowledge packs. It ingests PDFs, HWP files, GitHub repositories, and expert notes, then uses OCR, CLIP, and parsing to structure content into ontology packs with nodes, edges, and evidence. These packs can be queried via Graph RAG, served through MCP (Model Context Protocol) to tools like Claude Desktop and GPT, or packaged as skills for agents. Target users include researchers building private evidence graphs, domain experts commercializing expertise, and AI builders grounding agents with auditable knowledge. The platform follows a six-stage pipeline: ingest data, parse into chunks, structure into ontology packs, host in a workspace, package as skills, and deploy to agents. A nine-space MetaOntology guides extraction across Subject, Evidence, Claim, Policy, and Lever. Users can query via OpenCrab's web app, desktop client, MCP clients, or custom agents. Key differentiators include ontology pack reuse/monetization through a marketplace, MCP-native integration, and evidence-grounded answers with source auditing. Specific features include reverse ingest from connected MCP tools (Expert tier), a CrabSkill workflow with validator and playbook, and support for Korean and English. Operated by AgentKorea, OpenCrab emphasizes ontology modeling over simple vector search, making it ideal for knowledge-heavy teams. Compared to tools like RAGFlow or Danswer, OpenCrab's ontology-first approach and pack marketplace offer a unique ecosystem for structured knowledge products. However, it lacks a public API and may be complex for users seeking turnkey Q&A solutions.