Quick Start

AutoFlow runs as a Python CLI tool. It uses a mock LLM provider by default so you can explore the full pipeline without any API keys.

git clone https://github.com/your-org/autoflow.git
cd autoflow
pip install -e ".[dev]"

python -m autoflow "Summarize daily intelligence reports into a SITREP"

python -m autoflow "Classify incoming emails and route to the right team" --mode generate

# List all 20 starter templates
python -m autoflow --catalog

# Search for a workflow
python -m autoflow --search "classification"

# View a detailed workflow card
python -m autoflow --card classification

What Happens Under the Hood

When you run a pipeline command, AutoFlow:

1. Navigate — Determines if Agent Builder is the right tool
2. Normalize — Extracts structured task requirements from your input
3. Generate/Synthesize — Produces the workflow JSON or SITREP
4. Evaluate — Scores the output on correctness, completeness, format validity, efficiency, and reusability
5. Critique — If scores are below threshold, identifies issues and regenerates (up to 2 iterations)
6. Envelope — Wraps the final output in a Universal Output Envelope

Next Steps