DeepWisdom (FoundationAgents)

What it is. DeepWisdom is the company behind MetaGPT, the open-source multi-agent framework, and MGX (MetaGPT X), a natural-language-programming product; the MetaGPT repository now lives under the FoundationAgents GitHub organization. Why it matters. It is the source of the “software company as multi-agent team” idea — assigning PM, architect and engineer roles to LLMs under the “Code = SOP(Team)” philosophy — which has shaped how the broader agent field thinks about role decomposition. Key characteristics. A pip-installable, MIT-licensed Python framework; a research lineage producing the SPO, AOT and AFlow papers; and the MGX product at mgx.dev. What a professional should know. The framework is model-agnostic (users supply their own model keys), targets Python 3.9–3.11, and the company’s headquarters and funding are not publicly documented.

DeepWisdom develops MetaGPT, an open-source multi-agent framework that “assigns different roles to GPTs to form a collaborative entity for complex tasks,” simulating a software company with product managers, architects, project managers and engineers. The repository recently moved from geekan/MetaGPT to FoundationAgents/MetaGPT. The team also ships MGX (MetaGPT X), launched February 2025 and described as “the world’s first AI agent development team,” a natural-language-programming product that productizes the same multi-agent idea behind a consumer-facing interface.

Why it matters

China AI Hub analysis indicates DeepWisdom matters as the idea-source of the multi-agent paradigm rather than a deployment target. Its “software company” SOP model and its papers have influenced how the field decomposes work across specialized agent roles, even as its direct production use remains a developer’s assembly job.

China AI Hub analysis indicates DeepWisdom’s structural role is framework-level idea-generation: where Dify and FastGPT are tools an operator runs, MetaGPT is a library a developer imports, so its influence shows up less in deployments than in the design patterns — role-based SOPs and self-optimizing workflows — that the rest of the agent field has adopted. That is why its commercial energy has shifted to MGX, which productizes the same multi-agent idea behind a natural-language interface.

China AI Hub analysis indicates the company runs the open-core pattern in a framework shape rather than a platform shape: an MIT core that is free and self-hosted, plus a commercial layer (MGX) that wraps it. The contrast with Dify and FastGPT is material — their open cores are systems you run, while DeepWisdom’s is a library you import — which explains why the repository’s move to FoundationAgents matters less than the split between the free framework and the paid product it feeds.

How it differs from Dify, FastGPT and the bound platforms

DeepWisdom sits in a distinct cell of the agent ecosystem from both the open platforms and the vendor-bound builders.

• Dify (LangGenius) and FastGPT (Labring) are platforms — self-hostable application builders (workflow/RAG/observability, or knowledge-base/RAG) that an operator runs.
• Coze and Yuanqi are vendor-bound managed builders — defaulting to Doubao and Hunyuan respectively — with no open weights and no self-hosted exit.
• MetaGPT (DeepWisdom) is a role-based multi-agent library — a pip-installable SDK whose core claim is not “run this to build apps” but “encode your team’s SOPs as agents and compose them in Python.”

China AI Hub analysis: the cleanest contrast is framework-versus-platform and framework-versus-vendor. A platform gives you a running system and a UI; a framework gives you building blocks and a loop. DeepWisdom is the clearest example in this database of the latter — a developer-facing library whose most famous artifact is an idea (the software-company SOP) rather than a hosted surface, which is why it pairs naturally with a productized layer (MGX) rather than replacing one.

Practical implications and limitations

For researchers and framework builders. MetaGPT is the reference to study role-based SOP orchestration — the SPO/AOT/AFlow lineage is a research asset in its own right, and the MIT license makes the code freely inspectable.

For developers wanting multi-agent codegen. The CLI and library paths are documented, but the developer must supply and pay for their own LLM keys and stay on Python 3.9–3.11 — a research-grade setup, not a turnkey product. Python 3.12 is not yet supported per the README.

For teams wanting a product. MGX is the productized path; the framework itself is the SDK underneath. Anyone evaluating DeepWisdom for production should be clear which layer they are actually adopting — the free MIT framework or the paid MGX surface — because the two carry different support and stability expectations.

Entity hub

Products

• MetaGPT

Agents

• MetaGPT

Research / Technology

• AI Agents
• Tool Calling
• Agent ecosystem structure and gaps

What is uncertain

• Headquarters and funding are not publicly documented.
• The repository’s move from geekan/MetaGPT to FoundationAgents/MetaGPT is documented in the redirect but not explained on the company site.
• Python 3.12 support timing and the formal relationship between the open framework and MGX are not documented.

Sources

evidence_id	source_name	source_url	source_type	published	verified	confidence	conflict
src-companies-deepwisdom-1	MetaGPT GitHub repository (FoundationAgents)	https://github.com/FoundationAgents/MetaGPT	Official documentation	—	2026-09-29	high	—
src-companies-deepwisdom-2	MetaGPT documentation (DeepWisdom)	https://docs.deepwisdom.ai/	Official documentation	—	2026-09-29	high	—
src-companies-deepwisdom-3	MGX (MetaGPT X) product site	https://mgx.dev/	Official	—	2026-09-29	high	—

Labels used above: Official fact (from the MetaGPT GitHub repository and DeepWisdom documentation), Vendor-reported claim (product statements by DeepWisdom), and China AI Hub analysis (our synthesis, always introduced as such). No third-party evaluation evidence is currently recorded for the MetaGPT framework.