manager-gpt

ManagerGPT

Feature: gpt CLI: autogpt man

ManagerGPT is the top-level orchestrator in the AutoGPT agent mesh. It receives a high-level project goal, decomposes it into discrete subtasks, assigns each subtask to the appropriate specialist agent, and consolidates their outputs into a coherent final result.

What ManagerGPT Solves

Complex software projects involve multiple domains, e.g. architecture, backend, frontend, design, that must be developed in a coordinated way. ManagerGPT eliminates the need to manually decompose tasks and coordinate specialists. You give it one goal; it handles the rest.

How It Works

User Goal
    │
    ▼
ManagerGPT (decomposes)
    │
    ├──▶ ArchitectGPT  →  Architecture diagram
    ├──▶ BackendGPT    →  Server-side code
    ├──▶ FrontendGPT   →  UI code
    └──▶ DesignerGPT   →  UI mockups (optional)
    │
    ▼
ManagerGPT (consolidates)
    │
    ▼
Final Output to User

ManagerGPT communicates its decomposed subtasks to each agent by constructing specialized Task descriptions enriched with the original goal context. Each agent receives not just its slice of the work but enough context to make coherent decisions.

CLI Usage

autogpt man

AutoGPT prompts for your project goal interactively. For example:

> Develop a full stack app that fetches today's weather in Python using FastAPI.

ManagerGPT dispatches to ArchitectGPT, DesignerGPT, BackendGPT, and FrontendGPT. Terminal output:

[*] "ManagerGPT": Executing task: "Develop a full stack app that fetches today's weather in python using FastAPI."
[*] "ArchitectGPT": Executing tasks: Task { description: "- Design the user interface for the weather app..." }
[*] "BackendGPT": Executing tasks: Task { description: "- Using FastAPI in Python, create a backend..." }
[*] "FrontendGPT": Executing tasks: Task { description: "- Using FastAPI in Python, create a user interface..." }
[*] "ManagerGPT": Completed Task: Task { ... }