Framework Adapters

One of Anvil’s key features is framework portability. A single tool can be used with LangChain, CrewAI, AutoGen, OpenAI Agents SDK, or standalone.

Why Adapters?

Each AI framework has its own tool interface:

Framework	Tool Interface
LangChain	`BaseTool` with `_run()` method
CrewAI	`BaseTool` with Pydantic models
AutoGen	`FunctionTool` with type hints
OpenAI Agents	`FunctionTool` with JSON schema

Writing tools for each framework means duplicating code. Anvil solves this with automatic adapters.

Quick Example

from anvil import Anvil

anvil = Anvil()

# Create one tool
search = anvil.use_tool(
    name="web_search",
    intent="Search the web for information"
)

# Use it everywhere
lc_tool = search.to_langchain()      # LangChain
crew_tool = search.to_crewai()       # CrewAI
ag_tool = search.to_autogen()        # AutoGen
oai_tool = search.to_openai_agents() # OpenAI Agents SDK

Adapter Methods

Every Anvil Tool object has these methods:

Method	Returns	Framework
`to_langchain()`	`BaseTool`	LangChain
`to_crewai()`	`BaseTool`	CrewAI
`to_autogen()`	`FunctionTool`	AutoGen
`to_openai_agents()`	`FunctionTool`	OpenAI Agents SDK

What Gets Converted

Adapters automatically convert:

Name → Tool name/function name
Intent → Description for the LLM
Inputs → Typed parameters (Pydantic, JSON Schema, etc.)
Execution → Wrapped run() call

Installation

Install the adapter for your framework:

# LangChain
pip install "anvil-agent[langchain]"

# CrewAI
pip install "anvil-agent[crewai]"

# AutoGen
pip install "anvil-agent[autogen]"

# OpenAI Agents SDK
pip install "anvil-agent[openai-agents]"

# All frameworks
pip install "anvil-agent[all-frameworks]"

Framework Guides

Detailed integration guides for each framework:

Async Support

All adapters support async execution:

# Sync
result = search.run(query="test")

# Async
result = await search.run_async(query="test")

Framework adapters inherit this capability where supported.

Type Safety

Adapters generate proper type annotations from your InputParam definitions:

from anvil import InputParam

tool = anvil.use_tool(
    name="calculate",
    intent="Perform calculation",
    inputs=[
        InputParam(name="x", param_type="float", required=True),
        InputParam(name="y", param_type="float", required=True),
        InputParam(name="operation", param_type="str", default="add"),
    ]
)

# LangChain adapter creates a Pydantic model:
# class CalculateInput(BaseModel):
#     x: float
#     y: float
#     operation: str = "add"

Fallback Behavior

If a framework isn’t installed, calling its adapter raises a helpful error:

tool.to_langchain()
# ImportError: LangChain is not installed.
# Install it with: pip install "anvil-agent[langchain]"