FastGPT — Knowledge-Based AI Platform with Visual Workflow Builder

Introduction

FastGPT is a self-hosted platform for building knowledge-based AI applications. It combines RAG retrieval, visual workflow orchestration, and data processing into an integrated system where teams can create and deploy Q&A bots and AI assistants without writing code.

What FastGPT Does

• Processes documents into searchable knowledge bases with automatic chunking and embedding
• Provides a visual drag-and-drop workflow editor for building AI application logic
• Implements RAG retrieval with hybrid search combining vector and full-text methods
• Supports multi-model routing across OpenAI, Claude, and open-source models
• Offers API endpoints for integrating built applications into external systems

Architecture Overview

FastGPT is built with Next.js for the frontend and Node.js for the backend, backed by MongoDB for application data and PostgreSQL with pgvector for vector storage. The workflow engine executes directed acyclic graphs of processing nodes. A plugin system allows extending workflows with custom nodes for HTTP calls, code execution, and database queries.

Self-Hosting & Configuration

• Deploy with Docker Compose using the provided configuration files
• Configure LLM provider API keys and model endpoints in the environment file
• Set up MongoDB and PostgreSQL with pgvector for data storage
• Customize embedding models and chunk sizes for knowledge base ingestion
• Configure SSO and team permissions through the admin panel

Key Features

• Visual workflow builder with drag-and-drop node composition
• Hybrid RAG search combining vector similarity and keyword matching
• Multi-model support with automatic routing and fallback
• Plugin system for extending workflows with custom logic
• Team collaboration with role-based access control

Comparison with Similar Tools

• Dify — similar visual builder; FastGPT has deeper knowledge base processing features
• RAGFlow — focuses on document understanding; FastGPT adds a full workflow orchestration layer
• Flowise — LangChain-based flow builder; FastGPT integrates its own RAG pipeline natively
• Quivr — simpler RAG interface; FastGPT offers more advanced workflow composition
• AnythingLLM — all-in-one desktop RAG; FastGPT is a team-oriented web platform

FAQ

Q: What document formats does FastGPT support? A: PDF, Word, Excel, CSV, Markdown, plain text, and web page URLs.

Q: Can I use local models? A: Yes. Configure any OpenAI-compatible endpoint including Ollama, vLLM, or LocalAI.

Q: Is there a free hosted version? A: The team offers a cloud version, but the open-source edition is fully functional for self-hosting.

Q: How does the workflow editor work? A: It provides a canvas where you connect nodes for input, knowledge retrieval, LLM calls, conditional logic, and output.

Sources

• https://github.com/labring/FastGPT
• https://fastgpt.in