NodeCraft 0.2.0

A modular framework for building composable AI workflows

NodeCraft

A modular framework for building composable AI workflows with a "building blocks" design philosophy.

Overview

NodeCraft provides a pluggable Node and Scenario system that allows you to discover, create, and compose reusable AI workflow components. Think of it as building blocks for AI-powered code analysis and automation.

Key Features

• Node-based architecture with automatic discovery
• YAML-based scenario definition
• Template-based Node and Scenario creation
• Configuration file system for team collaboration
• Dynamic CLI generation from scenarios
• Built-in scenarios for common tasks

Installation

# Install from PyPI
pip install NodeCraft

# Set up API key
export ANTHROPIC_API_KEY="your-api-key-here"

After installation, the nodecraft command is available globally.

Quick Start

Discover Available Nodes and Scenarios

# List all available nodes
nodecraft nodes list

# Show node details
nodecraft nodes show @common/get_files

# List all scenarios
nodecraft scenarios list

# Show scenario details
nodecraft scenarios show simple_rag

Create New Nodes and Scenarios

# Create a function-based node
nodecraft nodes create --name my_analyzer

# Create a class-based node with retry logic
nodecraft nodes create --name api_caller --type class

# Create a scenario from template
nodecraft scenarios create --name my_workflow --template rag-query

# View available templates
nodecraft scenarios create --name test --template custom --dry-run

Register Custom Nodes and Scenarios

# Register a single node
nodecraft nodes register ./custom_nodes/my_node.py

# Register all nodes in a directory
nodecraft nodes register ./custom_nodes/ --recursive

# Register a scenario
nodecraft scenarios register ./my_scenario.yaml

Configuration

Create a .nodecraft.yaml file in your project root:

nodes:
  search_paths:
    - ".nodecraft/nodes"
    - "./custom_nodes"

scenarios:
  search_paths:
    - ".nodecraft/scenarios"
    - "./workflows"

llm:
  default_model: "claude-3-haiku-20240307"
  api_timeout: 60

Configuration priority: CLI arguments > Project config > Global config > Defaults

Global config location: ~/.nodecraft/config.yaml

Built-in Scenarios

OutcomeForge includes several built-in scenarios for common tasks:

• snapshot: Create snapshots of your codebase with AI analysis
• adapt: Analyze open-source repositories and generate adaptation plans
• regression: AI-powered quality gate based on test metrics
• arch-drift: Detect architecture violations and structural drift
• rag: Lightweight RAG for codebase Q&A
• code-review: Security, quality, and performance code review
• wiki: Generate structured wiki documentation from codebases

Example usage:

# Create a snapshot
nodecraft snapshot --patterns "**/*.py"

# Ask questions about your codebase
nodecraft rag --patterns "**/*.py" --query "How does this work?"

# Review code changes
nodecraft code-review --git-diff

# Generate wiki documentation
nodecraft wiki --local-dir ./my-project

CLI Usage

Node Management

# List nodes
nodecraft nodes list [--namespace <namespace>]

# Show node details
nodecraft nodes show <node_id>

# Create new node from template
nodecraft nodes create --name <name> [--type function|class]

# Register custom node
nodecraft nodes register <path> [--recursive]

Scenario Management

# List scenarios
nodecraft scenarios list

# Show scenario details
nodecraft scenarios show <scenario_id>

# Create new scenario from template
nodecraft scenarios create --name <name> --template <template>

# Register custom scenario
nodecraft scenarios register <path> [--recursive]

# Run a scenario
nodecraft <scenario_id> [OPTIONS]

Available scenario templates:

• rag-query: RAG-based codebase Q&A
• file-process: File collection and processing
• analyze-report: Code analysis with report generation
• gate-check: Quality gate enforcement
• snapshot-restore: Version control and rollback
• custom: Blank template for custom workflows

Tutorial System

# Get started tutorial
nodecraft tutorial

# Node creation tutorial
nodecraft tutorial nodes

# Scenario tutorial
nodecraft tutorial scenarios

Architecture

Node System

Nodes are the smallest units of work. Each node has three phases:

1. prep: Validate inputs and prepare parameters
2. exec: Execute the main operation
3. post: Process results and update context

Nodes are discovered automatically from:

• Framework built-in nodes (nodes/common/)
• Global user nodes (~/.nodecraft/nodes/)
• Project nodes (.nodecraft/nodes/)
• Configured search paths

Scenario System

Scenarios are workflows composed of nodes. They can be defined in:

1. Python (programmatic):

from engine import flow
from nodes.common import get_files_node, call_llm_node

def my_scenario(config):
    f = flow()
    f.add(get_files_node(), name="get_files")
    f.add(call_llm_node(), name="analyze")
    return f

2. YAML (declarative):

scenario:
  id: my_scenario
  name: My Scenario
  description: Custom workflow

parameters:
  patterns:
    type: list
    default: ["**/*.py"]

steps:
  - node: "@common/get_files"
    name: get_files
    params:
      patterns: "{{params.patterns}}"

  - node: "@common/call_llm"
    name: analyze

Context Flow

Data flows through the scenario via a shared context dictionary. Each node:

• Reads data from context using input_keys
• Writes results to context using output_keys
• Can access parameters via {{params.name}} syntax in YAML

Development

Project Structure

nodecraft/
├── core/                    # Core framework
│   ├── registry.py         # Node discovery and registration
│   ├── scenario_registry.py # Scenario management
│   ├── template_generator.py # Template generation
│   └── config.py           # Configuration management
├── nodes/
│   └── common/             # Built-in nodes
├── scenarios/              # Built-in scenarios
├── templates/              # Node and scenario templates
│   ├── nodes/
│   └── scenarios/
└── cli.py                  # CLI entry point

Running Tests

# Run all tests
pytest tests/

# Run specific test file
pytest tests/test_template_creation.py -v

Requirements

• Python 3.7+
• anthropic (Claude API)
• openai (OpenAI API)
• click (CLI framework)
• pyyaml (YAML parsing)
• gitpython (Git operations)
• jinja2 (Template rendering)

License

MIT