is chatgpt a multi-agent system?
no. chatgpt is a single conversational agent with tool access and internal routing between models. some features, like deep research and agent mode, run sub-tasks that look agentic, but you cannot define separate agents with their own roles, memory, and permissions, which is what makes a system multi-agent.
what chatgpt actually is
One conversation, one agent, several models behind a router. When you send a message it may be dispatched to a faster or a stronger model, and the product may call tools: web search, code execution, file reading, image generation. All of that happens inside a single agent loop with a single context.
That is a sophisticated agent. It is not several agents, because there is only one entity holding goals and deciding actions.
the features that look multi-agent and are not
Deep research decomposes a question, runs many retrieval steps, and assembles a report. It genuinely spawns parallel sub-tasks. Those sub-tasks have no independent goals, no separate memory, and no ability to disagree with the parent. That is a search plan, not a team.
Agent mode operates a browser and a virtual machine over long horizons. Still one agent, with a bigger tool belt.
Custom GPTs let you configure different assistants. They cannot talk to each other or hand work between themselves, so you have several separately invoked agents, not a system.
the four-part test
A multi-agent system needs autonomy, local views, a shared environment, and decentralised control. Score chatgpt:
| property | chatgpt |
|---|---|
| several autonomous agents | no, one agent |
| separate local context per agent | no, one context |
| shared environment they both act on | no |
| decentralised control | no, one loop |
Zero of four. The reason this question keeps getting asked is that the internal routing between models is sometimes described as agents in coverage, which is a different use of the word.
where openai does do multi-agent
The Agents SDK is a developer library for exactly this: defining several agents with their own instructions and tools, and handing off between them. That is multi-agent, and it is a completely different product from chatgpt. If a comparison sounds like openai supports multi-agent, it is usually about the sdk, and it means you write the orchestration yourself.
how this works in aldena
Aldena is the thing chatgpt is not. Several agents, each with its own instructions, its own model, its own memory, and its own tool permissions, sharing the same work inside one isolated project.
They share the project and the work rather than a single context: 100 shared memory entries every agent reads, plus 50 private per agent. An org chart sets who reports to whom, so the intake agent delegates to build agents and the code review agent checks the result before it reaches you. You arrange it in the interface rather than writing orchestration code, and you watch it happen.
For the definition itself, see what is a multi-agent system. For the same question about microsoft, see is copilot a multi-agent system.
related questions
what is a multi-agent system?
several autonomous agents sharing an environment and coordinating toward a goal. the term is 40 years old and predates language models by decades.
is copilot a multi-agent system?
copilot itself is one agent. the multi-agent parts of microsoft's stack are copilot studio and autogen, which are separate products you configure yourself.
is chatgpt an agentic ai?
partly. plain chat is not agentic. agent mode and deep research are, because they plan, act, and revise without a prompt per step. the same product does both.
is chatgpt an agent or llm?
both, at different layers. gpt is the model. chatgpt is the product wrapped around it, and in agent mode that product behaves as an agent.
spin up your first project.
one project per client, product, or piece of work, staffed with intake, requirements, build and code review agents.