Agenti e multi-agente

RAGFlow

infiniflow/ragflow

RAGFlow is an open-source Retrieval-Augmented Generation engine that combines document ingestion, retrieval, grounded citations, configurable language and embedding models, and agent capabilities to provide context for LLM applications.

Apri il repository ufficiale ↗ Pagina iniziale del progetto ↗
★ 86,7KStelle
⑂ 10,2KFork
2007Problemi aperti
GoLingua
Apache-2.0Licenza
Q@project.QualityScorePunteggio editoriale

Panoramica

The project is designed to turn structured and unstructured data into searchable knowledge for production-oriented RAG systems. It supports template-based chunking, document parsing, multiple retrieval methods with fused reranking, traceable citations, APIs, agent workflows, pre-built agent templates, memory, MCP, and optional code execution. Go is listed as the primary repository language, while source development also requires Python and a JavaScript frontend toolchain.

Funzionalità principali

  • Deep document understanding for complex, unstructured formats
  • Template-based, visualized, and human-reviewable chunking
  • Grounded responses with references and traceable citations
  • Support for Word, Slides, Excel, TXT, images, scanned documents, structured data, and web pages
  • Configurable LLM and embedding-model services
  • Multiple recall methods with fused reranking
  • Agentic workflows, pre-built agent templates, memory, and MCP support
  • Python and JavaScript code-executor component
  • APIs for application and business-system integration
  • Elasticsearch storage by default, with Infinity as an alternative document engine
  • Multilingual README files

Requisiti, installazione e avvio rapido

Recommended self-hosting path: clone https://github.com/infiniflow/ragflow, enter the docker directory, check out a release matching the selected image, and start the service with Docker Compose. The excerpt uses v0.26.4 and the command docker compose -f docker-compose.yml up -d. GPU acceleration can be selected for DeepDoc tasks through the documented DEVICE setting. Confirm initialization through the container logs before opening the server address in a browser. Configure the desired LLM provider and corresponding API key in service_conf.yaml.template. Full instructions are available at https://ragflow.io/docs/dev/.

Utilizzo

After startup, open http://IP_OF_YOUR_MACHINE when using the default HTTP port 80, log in, configure an LLM provider and API key, ingest data, choose parsing and chunking behavior, and create a retrieval or agent workflow. The hosted service is available at https://cloud.ragflow.io. API-specific examples are not stated in the supplied repository metadata.

Compatibilità del modello e casi d'uso

The system supports configurable LLMs and embedding models. The supplied update notes explicitly mention DeepSeek v4, Gemini 3 Pro, OpenAI GPT-5 series models, and multimodal models for interpreting images in PDF or DOCX files. A complete provider and model compatibility matrix is not stated in the repository metadata.

Note su licenza e rischi

The repository metadata identifies the license as Apache-2.0. The repository license file is available at https://github.com/infiniflow/ragflow/blob/main/LICENSE. No additional license exceptions or dependency-license analysis are stated in the repository metadata.

Rilascio e manutenzione

The excerpt identifies Docker image v0.26.4 in its startup example. Listed updates include agentic workflow and MCP support, a Python/JavaScript executor, agent memory, multimodal PDF and DOCX image understanding, synchronized sources, orchestrable ingestion, additional model support, an OpenClaw skill, and multiple chat channels. Formal release notes are available at https://ragflow.io/docs/dev/release_notes and releases at https://github.com/infiniflow/ragflow/releases/latest.

Firecrawl

firecrawl/firecrawl

★ 161,1KTypeScript

LangChain

langchain-ai/langchain

★ 143,6KPython