flyto-blueprint

Self-evolving workflow pattern engine. flyto-blueprint enables AI agents to select pre-built workflow blueprints and fill in arguments rather than building workflows from scratch.

How It Works

1. Agent receives task
2. Search blueprints by keywords/tags
3. Select matching blueprint
4. Fill in arguments (url, selector, etc.)
5. Expand into executable workflow
6. Execute workflow
7. Report outcome → score adjusts
8. Low-scoring blueprints auto-retire

Key Features

• Pre-built patterns — 11 builtin blueprints for common tasks
• Self-learning — Successful workflows become reusable blueprints
• Scoring system — Quality tracking with auto-retirement
• Deduplication — Structural fingerprinting prevents duplicates
• Compose blocks — Reusable multi-step patterns
• Type-aware templates — Preserves int/dict/bool in substitutions

Quick Start

Installation

pip install flyto-blueprint

Basic Usage

from flyto_blueprint import BlueprintEngine, MemoryBackend

engine = BlueprintEngine(storage=MemoryBackend())

# List available blueprints
blueprints = engine.list_blueprints()

# Search by keyword
results = engine.search("scrape")

# Expand a blueprint into a workflow
result = engine.expand("browser_scrape", {
    "url": "https://example.com",
    "extract_selector": "#content",
})
# Returns: {'ok': True, 'data': {'steps': [...], 'edges': [...], 'yaml': '...'}}