FastGPT
Type: low-code · Vendor: Sealos / Labring · Language: TypeScript (visual) · License: FastGPT Open Source License (custom Apache-2.0 — multi-tenant SaaS resale forbidden, LOGO removal forbidden) · Status: active · Status in practice: mature
Provide an open-source enterprise AI productivity engine that assembles knowledge-grounded agents from a visual workflow canvas — knowledge base (RAG) + hybrid retrieval + plugin/agent nodes — with a bidirectional MCP surface and self-hostable Docker deployment.
Description. FastGPT is an open-source AI agent platform built by Labring/Sealos. The homepage calls it an 'Enterprise AI Productivity Engine' that lets you 'Build secure, controllable enterprise-grade AI Agents' and 'Build AI agents for your business like stacking blocks.' Its README states 'FastGPT is a knowledge-based platform built on the LLMs, offers a comprehensive suite of out-of-the-box capabilities such as data processing, RAG retrieval, and visual AI workflow orchestration.' RAG is the centre of gravity: documented features include '混合检索 & 重排' (hybrid retrieval and rerank), multi-knowledge-base reuse and mixing, and broad file-format ingestion (txt, md, html, pdf, docx, pptx, csv, xlsx). The platform exposes a '规划 Agent 模式' (Planning Agent mode), workflow and plugin workflows with basic RPA nodes, and '双向 MCP' (bidirectional MCP).
Agent loop shape. A FastGPT app is a visual workflow on a Flow canvas. Conversation workflows and plugin workflows are first-class app types, both composed of typed nodes (LLM, knowledge-base search, tool/plugin, conditional/code, RPA). Knowledge bases are uploaded, automatically structured and queried via hybrid retrieval with reranking; multiple bases can be reused and mixed inside one workflow. A '规划 Agent 模式' (Planning Agent mode) places an LLM-driven planner over tools and knowledge nodes. Tools and external integrations are pluggable via bidirectional MCP — FastGPT can call external MCP servers and publish workflows as MCP endpoints.
Primary use cases
- enterprise knowledge-grounded chat (RAG over uploaded documents)
- low-code workflow orchestration with plugin and RPA nodes
- agent-style planning over knowledge bases and tools
- bidirectional MCP — consume external MCP tools and expose FastGPT workflows as MCP
- self-hosted on-prem deployment (FastGPT Open Source License)
Key concepts
- App (workflow) → visual-workflow-graph (docs) — Visual workflow canvas — conversation workflow or plugin workflow — composing LLM, knowledge, tool/plugin and RPA nodes.
- Knowledge Base (知识库) → agentic-rag (docs) — Uploaded documents auto-structured for retrieval; supports multi-base reuse, mixing, and broad file formats.
- Hybrid retrieval + rerank (混合检索 & 重排) → hybrid-search (docs) — Combined full-text + semantic search followed by a reranking model over candidate chunks.
- Planning Agent mode (规划 Agent 模式) (docs) — Agent-style loop where an LLM plans a sequence of calls over tools and knowledge nodes.
- Plugin / RPA workflow → tool-use (docs) — Composable plugin workflows containing basic RPA nodes; published as reusable plugins for other workflows.
- Bidirectional MCP (双向 MCP) → mcp (docs) — FastGPT consumes external MCP servers as tools and publishes its own workflows as MCP endpoints.
Patterns this low-code implements —
- ★★Agentic RAG
Knowledge Base is the headline product surface; the homepage frames it as 'reject hallucinations with an accurate enterprise knowledge base' and the README documents broad file-format ingestion plus…
- ★★Hybrid Search
FastGPT README and homepage both name hybrid retrieval + rerank as a primary capability of the knowledge layer.
- ★★Tool Use
Plugin workflows are first-class app type; FastGPT supports bidirectional MCP and Planning Agent mode that calls tools. README explicitly lists '对话工作流、插件工作流' alongside RPA nodes.
- ★★Model Context Protocol
FastGPT v4.9.6+ ships two complementary MCP surfaces: MCP Server exposes multiple FastGPT applications via MCP for external consumption (SSE transport); MCP Tools batch-creates tools by parsing an ex…
- ★★Visual Workflow Graph
Homepage and README both describe a visual workflow canvas with rich node types; the workflow IS the agent in FastGPT.
- ★★Fallback Chain
Original stub claimed fallback-chain as first-class. Re-verified 2026-05-24: neither the FastGPT README nor doc.fastgpt.io documents a named LLM fallback-chain, alternative-model routing, or automati…
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.
Alternatives & relatives
- low-code · visual builder
Dify
LangGenius's open-source low-code agent platform — wider surface (agents, workflows, RAG, MCP) than FastGPT's RAG-first emphasis.
- low-code · visual builder
Coze
ByteDance hosted no-code peer; comparable Chinese-market positioning.
- low-code · visual builder
Bisheng
Open-source Chinese visual LLM app builder; enterprise document focus.
Listed as alternative by (4)
References
labring/FastGPT README
“FastGPT is a knowledge-based platform built on the LLMs, offers a comprehensive suite of out-of-the-box capabilities such as data processing, RAG retrieval, and visual AI workflow orchestration”
“Production-grade AI Agents - One platform for enterprise AI delivery”
Provenance
- Last analyzed:
- Last updated:
- Verification status: needs-verification