FastGPT

Short answer. FastGPT is an open-source, knowledge-base-first platform by Labring (环界云计算) for building LLM applications on top of RAG retrieval and visual workflow orchestration — China’s most prominent self-hosted answer to enterprise knowledge Q&A, with a commercial edition and a managed cloud.

Key facts.

• Knowledge-base-first: data processing, RAG retrieval, chunk editing, hybrid retrieval with rerank, and TXT/MD/HTML/PDF/Docx ingestion.
• Visual workflow orchestration with Agent Skill editing, dialog and plugin workflows (including RPA nodes), and bidirectional MCP.
• Model-agnostic through Labring’s AI Proxy — a model aggregation and load-balancing service.
• Three deployment modes: Docker Compose self-hosting, the managed cloud (fastgpt.io), or one-click deployment on Sealos Cloud.
• Open source under the “FastGPT Open Source License” — commercial use as a backend service is permitted, but a SaaS service or commercial redistribution requires authorization.
• A commercial edition adds deeper support; its pricing is documented separately from the repo.

What this means. FastGPT is the self-hosted knowledge-infrastructure play: it competes with the closed platforms’ knowledge features (Baidu AppBuilder, Coze) on a bring-your-own-model, self-hosted basis — the segment of Chinese enterprises that cannot send internal documents to a managed cloud.

What is uncertain. The commercial edition’s feature set and pricing (documented separately), the exact SaaS-restriction boundary, Labring’s headquarters and funding, and independent retrieval-quality evaluation are not recorded in this database.

Sources.

evidence_id	source_name	source_url	source_type	published	verified	confidence	conflict
src-agents-fastgpt-1	FastGPT GitHub repository	https://github.com/labring/FastGPT	Official documentation	—	2026-09-30	high	—
src-agents-fastgpt-2	FastGPT documentation	https://doc.fastgpt.io/	Official documentation	—	2026-09-30	high	—
src-agents-fastgpt-3	FastGPT cloud service	https://fastgpt.io/	Official	—	2026-09-30	high	—
src-agents-fastgpt-4	FastGPT Open Source License	https://github.com/labring/FastGPT/blob/main/LICENSE	Official documentation	—	2026-09-30	high	—
src-agents-fastgpt-5	FastGPT home (labring.github.io)	https://labring.github.io/fastgpt-home/	Official	—	2026-09-30	high	—

Why it matters

FastGPT matters as the knowledge-base counterpoint to the workflow-first open platforms. Where Dify leads with general application workflows, FastGPT leads with document ingestion, chunk management, hybrid retrieval and rerank — the RAG plumbing that enterprise knowledge assistants are built on.

China AI Hub analysis indicates FastGPT’s structural role is self-hosted knowledge infrastructure: it occupies the exact segment of Chinese enterprises that cannot send internal documents to a managed cloud, and it serves them on a bring-your-own-model basis. That is why its retrieval depth is the moat — chunk management, hybrid retrieval and rerank are treated as the primary surface, not a secondary feature, which is what clusters it with enterprise document-Q&A use cases rather than general agent-building.

China AI Hub analysis: FastGPT’s non-OSI license locates it between two poles — more open than a closed platform (you can self-host the core) but less open than a pure-MIT framework (you cannot resell it as SaaS). That boundary, like Dify’s, protects a commercial business model; the difference is that FastGPT draws the line at “backend service use” rather than “multi-tenant SaaS redistribution,” which is a stricter commercial gate for anyone wanting to wrap it into a product.

How it differs from Dify and the closed platforms

FastGPT’s differentiation is its depth in one area against breadth elsewhere.

• Dify is the general application platform — workflow, RAG, agents, observability — for prototype-to-production shipping across many model providers.
• Baidu AppBuilder and Coze are closed, single-vendor managed platforms with knowledge features, but cloud-only and bound to ERNIE and Doubao respectively.
• FastGPT is the knowledge-base specialist: open-source, self-hostable, model-agnostic via AI Proxy, with hybrid retrieval and rerank as first-class — and a bidirectional MCP that the closed platforms’ internal MCP does not mirror.

China AI Hub analysis: for an enterprise whose primary need is answering questions over an internal document corpus, FastGPT’s self-hosted retrieval stack is the closest direct match among the platforms tracked here — the reverse of Dify’s breadth-first bet. The two are complementary: FastGPT for retrieval-heavy knowledge work, Dify for general application scope.

Practical implications

For enterprises with sensitive documents. FastGPT is the documented self-hosted path to keep a document corpus inside the firewall while still getting RAG, rerank and workflow orchestration. The cost is operational: Docker Compose self-hosting or the paid cloud, plus supplying and paying for your own model via AI Proxy.

For knowledge-base teams. The retrieval depth (chunk editing, hybrid retrieval, rerank, multi-format ingestion) is the reason to choose FastGPT over a general platform — it is the primary surface, not an add-on.

For those who might resell. The license permits commercial use as a backend service but requires authorization for a SaaS offering or commercial redistribution — a stricter gate than MIT, and a material consideration for anyone building a product on top of it.

What the evidence shows

The evidence is strong on FastGPT’s retrieval depth because the capability list is explicit in primary material: application orchestration (Agent Skill editing, dialog and plugin workflows with RPA nodes, bidirectional MCP), application debugging (knowledge-base search testing, full call-chain logs, evaluation), knowledge-base management (multi-library reuse, chunk editing, hybrid retrieval and rerank, multi-format ingestion), plugins, and operations. The AI Proxy model-aggregation layer is documented as the mechanism of its model neutrality.

The gaps are commercial and evaluative. The commercial edition’s features and pricing are documented separately from the open-source repo; the SaaS-restriction boundary is drawn by a non-OSI license; Labring’s corporate details and funding are not on official channels; and there is no independent evaluation of retrieval quality across the supported formats and rerank models. China AI Hub analysis indicates the retrieval claims are well-documented as capabilities but unverified as measured results — the same vendor-reported-versus-independent gap that runs across the open agent platforms.

Where this fits

Workload	Relevance
Knowledge-base Q&A over an internal document corpus	High
Enterprise knowledge management / AI customer service	High
Hybrid retrieval + rerank pipelines	High
Visual agent and plugin workflow orchestration	High
Self-hosted deployment (Docker Compose / Sealos)	High
General prototype-to-production application building	Moderate (Dify is broader)
Reselling as a SaaS	Restricted (authorization required)

Relevance judgments are China AI Hub analysis based on documented capabilities, not vendor claims.

Field reference

Field	Value	Evidence type
Underlying model	Model-agnostic (via AI Proxy aggregation)	Official
Target users	Enterprises, knowledge-base builders, self-hosters	Official
Platform	Docker self-hosted; cloud (fastgpt.io); Sealos Cloud	Official
OS	Cross-platform (Docker); Linux server	Official
Browser / computer use	Not publicly documented	Not publicly documented
Coding	Not a primary focus	Official
Autonomous task execution	Yes (agent-loop, workflow orchestration)	Vendor-reported
MCP	Yes (bidirectional MCP)	Official
Tool calling	Yes (plugin workflow, RPA nodes)	Official
Memory	Yes (knowledge base, hybrid retrieval)	Official
Workflow	Visual workflow + Agent Skill + plugin workflow	Official
API	Yes (OpenAPI)	Official
Pricing	Open source (free); paid cloud and commercial edition	Vendor-reported
Region	China (with global self-hosting)	Official
Open-source	Yes — FastGPT Open Source License (non-OSI)	Official
Deployment	Both (self-hosted + cloud)	Official
Limitations	Non-OSI license; self-hosting or paid cloud for production	Official
Source	GitHub · docs · fastgpt.io	Official
Last verified	2026-09-30	Official

See the Labring company profile, the self-hosting guide, the choosing-an-agent guide, and the site’s AI agents, RAG and MCP technology pages. For the structural reading of FastGPT as the self-hosted counterpoint to Dify and the closed platforms, see the research on the agent ecosystem structure.

Labels used above: Official fact (from the FastGPT GitHub repository, LICENSE, documentation and cloud site), Vendor-reported claim (feature and deployment statements by Labring), and China AI Hub analysis (our synthesis, always introduced as such). No third-party evaluation evidence is currently recorded for FastGPT-built applications.