Introduction

Anvil is a JIT (Just-In-Time) Infrastructure SDK for AI agents. Instead of hardcoding tool implementations that break when APIs change, you define intents and Anvil generates the code on the fly.

The Problem: Tool Rot

Every AI agent needs tools - functions to search the web, call APIs, process data. Traditional approaches have a critical flaw:

# Traditional approach - brittle, breaks when API changes
def search_notion(query: str) -> dict:
    response = requests.post(
        "https://api.notion.com/v1/search",
        headers={"Authorization": f"Bearer {NOTION_KEY}"},
        json={"query": query}
    )
    return response.json()

This works until:

The API endpoint changes
Authentication requirements update
Response format evolves
Rate limiting is introduced

Then your agent breaks, and you’re manually fixing code.

The Solution: Intent-Based Tools

With Anvil, you describe what you want, not how to do it:

from anvil import Anvil

anvil = Anvil()

search_tool = anvil.use_tool(
    name="search_notion",
    intent="Search a Notion workspace for pages matching a query",
    docs_url="https://developers.notion.com/reference/post-search"
)

Anvil:

Fetches the latest API documentation
Generates working Python code using an LLM
Caches the generated code for future use
Automatically regenerates if the tool fails

Key Features

JIT Code Generation

Tools are generated at runtime using LLMs. Provide the intent and optionally a documentation URL, and Anvil produces working code.

Self-Healing

When a tool fails, Anvil captures the error context and regenerates the code automatically. No manual intervention needed.

Multi-Provider Support

Use Claude (Anthropic), GPT-4 (OpenAI), or Grok (xAI). Bring your own API keys.

Framework Adapters

One tool works everywhere:

# 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()

Glass-Box Transparency

All generated code is saved to disk, visible, and editable. You can:

Inspect what Anvil generated
Manually edit and “eject” tools
Version control everything

When to Use Anvil

Anvil is ideal when:

You’re building AI agents that need dynamic tool capabilities
You want tools that adapt to API changes automatically
You’re working with multiple AI frameworks and want portable tools
You need transparency into what your tools are doing

Next Steps

Ready to get started?