create-agent-app

Scaffold production-ready Agentic AI projects in Python with one command.

create-agent-app generates opinionated, extensible project templates built around LangGraph patterns, modern provider support, API-first structure, and practical defaults for real-world teams.

Quick Demo

Why create-agent-app

• Production-oriented templates (not toy examples)
• Consistent project structure across agent architectures
• Interactive setup flow for provider and feature selection
• Optional FastAPI backend, streaming, tests, Docker, and observability hooks
• Ready-to-customize prompts, graph logic, tools, and config files

Supported Templates

Template	Best For	Architecture
react_agent	Tool-using assistants with iterative reasoning	ReAct loop with LangGraph + tool execution
rag_agent	Grounded answers from your documents	Retrieval + grading + generation pipeline with optional guards/cache
multi_agent	Coordinated specialist workflows	Supervisor graph orchestrating worker agents
conversational	Alias for conversational assistants	Delegates to react_agent generation
hitl	Human-in-the-loop orchestration baseline	Delegates to multi_agent generation

Supported LLM Providers

• Groq
• Gemini
• Azure OpenAI
• Ollama

Provider credentials are configured via generated .env files.

Installation

pip install create-agent-app

For local development:

pip install -e .[dev]

Usage

create-agent-app my-agent-project

If your shell cannot find create-agent-app, use:

python -m create_agent_app my-agent-project

Optional output directory:

create-agent-app my-agent-project --output ./workspace

Alias command (equivalent):

create_agent_app my-agent-project

The CLI prompts you to configure:

• Template type
• LLM provider
• API backend and streaming support
• Pre-installed tools
• RAG options (semantic cache and security guards)
• Optional features (Docker, tests, observability, agent description)

What Gets Generated

Each project includes:

• A runnable main.py entrypoint
• Config-first setup (config.yaml, .env.example)
• Template-specific modules (agent/, rag/, agents/, tools/, etc.)
• Optional api/ routes and schemas
• Optional tests/
• Optional Docker artifacts

The generator also initializes git and creates a data/ directory scaffold.

Feature Flags (Template Rendering)

The generator conditionally includes files based on selected options:

• include_api: includes or skips api/
• include_tests: includes or skips tests/
• include_docker: includes or skips Docker artifacts
• include_guards (RAG): includes or skips security/
• include_semantic_cache (RAG): includes or skips semantic cache template

Example Developer Workflow

create-agent-app customer-support-agent
cd customer-support-agent
python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env
# Add provider credentials
python main.py

Project Quality Notes

• Templates are generated from Jinja files under create_agent_app/templates/
• Shared and template-specific rendering are centralized in create_agent_app/generator.py
• Generation summary is displayed in rich tables for clarity
• Templates are structured for straightforward extension, not lock-in

Development

pip install -e .[dev]
python -m build
python -m twine check dist/*

Troubleshooting (Windows)

If you see:

create-agent-app : The term 'create-agent-app' is not recognized ...

then the package is usually installed, but your Python Scripts directory is not on PATH.

Check install location:

py -m pip show create-agent-app
py -m site --user-base

Typical scripts path to add to PATH:

%APPDATA%\Python\Python3x\Scripts

After updating PATH, restart PowerShell and run:

create-agent-app --help

Fallback that always works when package is installed:

py -m create_agent_app --help

Publishing to PyPI

If a version already exists on PyPI, bump project.version in pyproject.toml before upload.

python -m build
twine upload dist/*

Optional:

twine upload --skip-existing dist/*

Contributing

1. Fork the repository.
2. Create a branch for your change.
3. Validate generated templates locally.
4. Submit a pull request with a clear summary and sample output.

License

MIT