Anvil SDK

JIT Infrastructure & Self-Healing SDK for AI Agents

Documentation | PyPI | GitHub

Anvil prevents "Tool Rot" in AI agents. Instead of hard-coding tool implementations that break when APIs change, define intents and Anvil generates the code on the fly.

Features

• JIT Code Generation - Generate tool code at runtime using LLMs
• Smart Parameterization - Automatically generates reusable, parameterized tools (no hardcoded values)
• Self-Healing - Automatic regeneration when tools fail
• Multi-Provider - Works with Claude, GPT, Grok (BYO API keys)
• Glass-Box - All generated code is visible and editable
• Framework Adapters - Works with LangChain, CrewAI, AutoGen, OpenAI Agents SDK
• Beautiful CLI - Professional terminal experience with Rich

Quick Start

1. Install

pip install anvil-agent

With Claude support (recommended):

pip install "anvil-agent[anthropic]"

2. Initialize Your Project

anvil init

This interactive wizard will:

• Create the anvil_tools/ directory for generated tools
• Set up .gitignore to protect your API keys
• Securely prompt for your API keys and save them to .env
• Create an example script to get you started

3. Start Building

from dotenv import load_dotenv
from anvil import Anvil

load_dotenv()

# Initialize Anvil
anvil = Anvil(tools_dir="./anvil_tools")

# Define what you want, not how
search_tool = anvil.use_tool(
    name="search_notion",
    intent="Search the user's Notion workspace using the official API",
    docs_url="https://developers.notion.com/reference/post-search"
)

# Execute
result = search_tool.run(query="Project Anvil")

# Code is saved to ./anvil_tools/search_notion.py
print(anvil.get_tool_code("search_notion"))

CLI Commands

anvil init      # Initialize a new project with interactive setup
anvil doctor    # Check system requirements and API keys
anvil list      # List all generated tools
anvil clean     # Clear tool cache to force regeneration
anvil verify    # Verify tool code in sandbox

Installation Options

# Basic installation
pip install anvil-agent

# With LLM provider
pip install "anvil-agent[anthropic]"  # Claude (recommended)
pip install "anvil-agent[openai]"     # GPT-4

# With framework adapter
pip install "anvil-agent[langchain]"
pip install "anvil-agent[crewai]"
pip install "anvil-agent[autogen]"
pip install "anvil-agent[openai-agents]"

# Everything
pip install "anvil-agent[all]"

Multi-Provider Support

# Use Claude (default)
anvil = Anvil(provider="anthropic")

# Use OpenAI GPT-4
anvil = Anvil(provider="openai", model="gpt-4o")

# Use Grok
anvil = Anvil(provider="grok", model="grok-2")

Framework Integration

# LangChain
lc_tool = search_tool.to_langchain()

# CrewAI
crew_tool = search_tool.to_crewai()

# AutoGen
autogen_tool = search_tool.to_autogen()

# OpenAI Agents SDK
oai_tool = search_tool.to_openai_agents()

Testing Without API Keys

anvil = Anvil(use_stub=True)  # Returns mock implementations

Anvil Cloud (Coming Soon)

For instant cached tools without LLM latency:

pip install anvil-cloud

anvil = Anvil(mode="cloud")  # Instant retrieval from global cache

Development

# Clone the repository
git clone https://github.com/Kart-ing/anvil-sdk.git
cd anvil-sdk

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest -v

# Run examples
python examples/basic_usage.py --stub

Documentation

Full documentation is available at anvil-sdk-docs.vercel.app

• Getting Started
• How It Works
• Self-Healing
• LangChain Integration
• CrewAI Integration
• AutoGen Integration
• OpenAI Agents SDK
• API Reference

Building & Publishing

# Build the package
python -m build

# Upload to TestPyPI first
twine upload --repository testpypi dist/*

# Upload to PyPI
twine upload dist/*

License

MIT License - See LICENSE for details.