🔧 agent‑generator

From one plain‑English sentence to a production‑ready multi‑agent workflow.

---

✨ Why agent‑generator?

Designing multi‑agent systems is powerful yet verbose—boilerplate, framework quirks, deployment plumbing.
agent‑generator turns intent (“I need an AI team that …”) into runnable artefacts:

Output	Where it runs
Python scripts for CrewAI, CrewAI Flow, LangGraph, ReAct	Any Python 3.9+ environment
YAML skill for WatsonX Orchestrate	IBM WatsonX Orchestrate
FastAPI MCP server (wrapper)	MCP Gateway / Kubernetes
JSON spec + Mermaid/DOT	Docs, audits, architecture diagrams

---

🚀 Features

Category	Details
Multi‑framework	WatsonX Orchestrate · CrewAI · CrewAI Flow · LangGraph · ReAct
Provider‑agnostic	IBM WatsonX (default) · OpenAI (extra)
MCP integration	Python outputs include a FastAPI /invoke endpoint ready for MCP Gateway registration
Web UI (Flask)	Form‑based generator with live diagram preview & code download
Cost estimator	Token counts & USD approximation before you run the prompt
Pluggable	Add providers / frameworks via a single subclass & registry
Typed models	Pydantic schemas for Agents · Tasks · Workflow
CI / CD	Ruff · Mypy · Pytest in GitHub Actions + automatic PyPI publish
Docker‑ready	Alpine image with Gunicorn server

---

📦 Installation

# Core (WatsonX only)
pip install agent-generator

# + OpenAI provider
pip install "agent-generator[openai]"

# + Flask Web UI and dev tools
pip install "agent-generator[dev,web]"

Requires Python ≥ 3.9.

---

⚡ Quick Start

1 Export WatsonX credentials (or put them in .env)

export WATSONX_API_KEY=...
export WATSONX_PROJECT_ID=...
export WATSONX_URL=https://us-south.ml.cloud.ibm.com
export AGENTGEN_PROVIDER=watsonx                   

---

2 Generate a WatsonX Orchestrate YAML skill

agent-generator "I need a research assistant that summarises papers" --framework watsonx_orchestrate 

or

agent-generator \
  "I need a research assistant that summarises papers" \
  --framework watsonx_orchestrate \
  --output research_assistant.yaml

You will get

✓ Written to research_assistant.yaml

Import the resulting research_assistant.yaml via:

orchestrate agents import -f research_assistant.yaml

---

3 Generate a CrewAI Python skill with MCP wrapper

agent-generator \
  "I need a research assistant that summarises papers" \
  --framework crewai  

or

agent-generator \
  "I need a research assistant that summarises papers" \
  --framework crewai --mcp \
  --output research_team.py

✓ Written to research_team.py

4 Run the Python skill locally

python research_team.py          # one‑off execution
python research_team.py serve    # FastAPI /invoke on :8080

---

🖥 CLI Reference

agent-generator [OPTIONS] "requirement sentence"

Option / Flag	Default	Description
-f, --framework *	—	watsonx_orchestrate · crewai · crewai_flow · langgraph · react
-p, --provider	watsonx	watsonx (default) | openai
--model	provider default	Override LLM model
--mcp / --no-mcp	off	Append FastAPI MCP wrapper (Python only)
-o, --output PATH	stdout	Write artefact to file
--dry-run		Build spec & code skeleton, no LLM calls
--show-cost		Print token usage + USD estimate

See the full Usage guide for examples.

---

🌐 Web UI

FLASK_APP=agent_generator.web FLASK_ENV=development flask run

Navigate to http://localhost:5000 → fill the form → generate → download code/YAML.

Deployed via Docker:

docker build -t agent-genenerator .
docker run -e WATSONX_API_KEY=... -p 8000:8000 agent-genenerator

🛠 Extending

• New provider → subclass BaseProvider, register in providers/__init__.py.
• New framework → subclass BaseFrameworkGenerator, register in frameworks/__init__.py. The MCP wrapper is added automatically for any Python output.

---

🧑‍💻 Development

git clone https://github.com/ruslanmv/agent-generator.git
cd agent-generator
pip install -e ".[dev,web,openai]"
pre-commit install
make lint test
mkdocs serve        # live docs

Documentation

https://ruslanmv.com/agent-generator/

---

🗺️ Roadmap

• Azure OpenAI provider
• MCP Gateway dashboard helper
• VS Code “Generate agent” command
• Real‑time collaborative flow editor

---

📄 License

Apache License, Version 2.0 © 2025 Ruslan M. V. – contributions are welcome, PRs + issues encouraged!