agent-factory: Generate Executable AI Agent Workflows

Overview

Agent Factory is a command-line tool that generates Python code for AI agent workflows from a description of the task. It addresses the setup work involved in assembling an agent, its tools, and its operational files, with generated workflows built on the any-agent library.

It is intended for developers and teams exploring agentic workflows who want to start from a natural-language goal rather than write the entire workflow by hand. Its support for MCP servers and multiple agent frameworks makes it useful when tool integration or framework choice matters.

Key Features

• Generates executable Python workflows from natural-language prompts.
• Supports one-shot tasks and multi-turn conversations.
• Uses any-agent to work across supported agent frameworks and model providers.
• Identifies relevant tools and MCP servers for a described task.
• Produces runnable agents with operational instructions and dependency files.
• Supports evaluating generated agents against automatic or manual criteria.

Use Cases

• Developers can prototype an agent for a narrowly defined task before investing in a hand-built implementation.
• Teams can create agents that use MCP tools and servers to carry out work requiring external capabilities.
• Agent developers can compare workflow approaches across supported frameworks or model providers.
• Teams can generate a workflow and assess it against explicit criteria before adopting it for a task.

Getting Started

The README demonstrates invoking the CLI with a task description:

uv run agent-factory "Create an AI agent that searches the web for the latest news on open-source AI and creates a newsletter article summarizing the findings."

See the README and documentation for installation and configuration details.

Considerations

The README specifies Python 3.13 or later. Generated workflows depend on the capabilities of the selected models, frameworks, and tools, so review and evaluate generated code before relying on it. The repository lists 33 open issues, and the supplied project data does not establish production readiness or long-term support.