API Reference

BlueprintEngine

The main orchestrator class.

Initialization

from flyto_blueprint import BlueprintEngine, MemoryBackend, get_engine

# Direct initialization
engine = BlueprintEngine(storage=MemoryBackend())

# Singleton pattern
engine = get_engine(storage=MemoryBackend())

Methods

list_blueprints()

List all non-retired blueprints, sorted by score.

blueprints = engine.list_blueprints()
# [{'id': 'browser_scrape', 'name': 'Scrape Page Content', 'score': 85, ...}, ...]

search(query)

Search blueprints by name, description, and tags.

results = engine.search("scrape")
# Filtered + ranked by relevance and quality

expand(blueprint_id, args)

Expand a blueprint into an executable workflow.

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

Returns warnings if optional arguments are missing (steps using skip_if_missing are omitted).

learn_from_workflow(workflow, name=None)

Abstract a concrete workflow into a reusable blueprint.

result = engine.learn_from_workflow(workflow, name="My Pattern")
# {'ok': True, 'data': {'id': 'my_pattern', 'score': 50, ...}}
# Or if duplicate: {'ok': True, 'action': 'boosted_existing', ...}

learn_from_execution(workflow, name=None)

Like learn_from_workflow but marks the blueprint as verified (initial score 70 instead of 50).

report_outcome(blueprint_id, success)

Report execution success or failure. Adjusts the blueprint's quality score.

result = engine.report_outcome("my_pattern", success=True)
# {'ok': True, 'score': 55}

result = engine.report_outcome("bad_pattern", success=False)
# {'ok': True, 'score': 40, 'retired': False}

MCP Tools

4 tools are exposed for LLM/agent integration:

Tool	Description
list_blueprints	List and search available blueprints
use_blueprint	Expand a blueprint with arguments into a workflow
save_as_blueprint	Learn a new blueprint from a workflow
report_blueprint_outcome	Report success/failure for scoring

Blueprint YAML Format

id: browser_scrape
name: "Scrape Page Content"
description: "Open a website and extract content"
tags: [browser, scrape, extract]
compose: [browser_init]          # Include reusable blocks
args:
  url:
    type: string
    required: true
  extract_selector:
    type: string
    required: true
  extract_type:
    type: string
    required: false
steps:
  - id: extract_data
    module: browser.extract
    params:
      selector: "{{extract_selector}}"
      extract_type: "{{extract_type}}"
      skip_if_missing: [extract_type]