OpenAPI-LLM

A Python library that converts OpenAPI specifications into Large Language Model (LLM) tool/function definitions, enabling OpenAPI invocations through LLM generated tool calls.

Table of Contents

• Overview
• Features
• Installation
  • Supported Python Versions
  • LLM Provider Dependencies
• Quick Start
  • Synchronous Example
  • Asynchronous Example
• Customization: from_spec
• Library Scope
  • OpenAPI Specification Validation
• Requirements
• Development Setup
• Testing
• License
• Security
• Contributing

Features

• Convert OpenAPI specifications into LLM-compatible tool/function definitions
• Support for multiple LLM providers (OpenAI, Anthropic, Cohere)
• Handle complex request bodies and parameter types
• Support for multiple authentication mechanisms
• Support for OpenAPI 3.0.x and 3.1.x specifications
• Handles both YAML and JSON OpenAPI specifications

Installation

pip install openapi-llm

Supported Python Versions

• Python >= 3.8

LLM Provider Dependencies

By default, OpenAPI-LLM does not install any particular LLM provider. You can install exactly the ones you need:

pip install openai     # For OpenAI
pip install anthropic  # For Anthropic
pip install cohere     # For Cohere

Quick Start

Below are minimal working examples for synchronous and asynchronous usage.

Synchronous Example

import os
from openai import OpenAI
from openapi_llm.client.openapi import OpenAPIClient

# Create the client from a spec URL (or file path, or raw string)
service_api = OpenAPIClient.from_spec(
    openapi_spec="https://bit.ly/serperdev_openapi",
    credentials=os.getenv("SERPERDEV_API_KEY")
)

# Initialize your chosen LLM provider (e.g., OpenAI)
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

# Ask the LLM to call the SerperDev API
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Do a serperdev google search: Who was Nikola Tesla?"}],
    tools=service_api.tool_definitions,  # LLM tool definitions from the client
)

# Now actually invoke the OpenAPI call based on the LLM's generated tool call
service_response = service_api.invoke(response)
assert "inventions" in str(service_response)

Asynchronous Example

import os
import asyncio
from openapi_llm.client.openapi_async import AsyncOpenAPIClient
from openai import AsyncOpenAI

async def main():
    # Firecrawl openapi spec
    openapi_spec_url = "https://raw.githubusercontent.com/mendableai/firecrawl/main/apps/api/v1-openapi.json"

    # Create the async client
    service_api = AsyncOpenAPIClient.from_spec(
        openapi_spec=openapi_spec_url,
        credentials=os.getenv("FIRECRAWL_API_KEY")
    )

    # Initialize an async LLM (OpenAI)
    client = AsyncOpenAI(api_key=os.getenv("OPENAI_API_KEY"))

    # Ask the LLM to call Firecrawl's scraping endpoint
    response = await client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Scrape URL: https://news.ycombinator.com/"}],
        tools=service_api.tool_definitions,
    )

    # Use context manager to manage aiohttp sessions
    async with service_api as api:
        service_response = await api.invoke(response)
        assert isinstance(service_response, dict)
        assert service_response.get("success", False), "Firecrawl scrape API call failed"

asyncio.run(main())

Customization: from_spec

Both OpenAPIClient and AsyncOpenAPIClient provide a classmethod called from_spec, which automatically:

• Loads the OpenAPI specification from a file path, URL, or raw string.
• Builds a ClientConfig for you.
• Constructs the client instance.

For example, you can validate the spec before using it by supplying a custom config_factory:

from openapi_llm.client.config import ClientConfig, create_client_config
from openapi_llm.client.openapi import OpenAPIClient
from openapi_spec_validator import validate_spec

def my_custom_config_factory(openapi_spec: str, **kwargs) -> ClientConfig:
    config = create_client_config(openapi_spec, **kwargs)
    validate_spec(config.openapi_spec.spec_dict)
    return config

# Usage:
client = OpenAPIClient.from_spec(
    openapi_spec="path/to/local_spec.yaml",
    config_factory=my_custom_config_factory,
    credentials="secret_token"
)

This design gives you full control over the spec-loading and configuration-building process while still offering simple defaults.

Library Scope

OpenAPI-LLM focuses on the core of bridging LLM function calls with OpenAPI specifications. It does not perform advanced validation or impose a high-level framework. You can integrate it into your existing app or build additional logic on top.

OpenAPI Specification Validation

This library does not automatically validate your specs. If your OpenAPI file is invalid, you might see errors during usage. Tools like openapi-spec-validator or prance can help ensure correctness before you load your spec here.

Requirements

• Python >= 3.8
• Dependencies:
  • jsonref
  • requests
  • PyYAML

Development Setup

1. Clone the repository:

   git clone https://github.com/vblagoje/openapi-llm.git

2. Install Hatch (if you haven’t already):

   pip install hatch

3. Install Pre-Commit Hooks:

   pip install pre-commit

4. Install Desired LLM Provider Dependencies (e.g., openai, anthropic, cohere):

   pip install openai anthropic cohere

Testing

Run tests using hatch:

# Unit tests
hatch run test:unit

# Integration tests
hatch run test:integration

# Type checking
hatch run test:typing

# Linting
hatch run test:lint

License

This project is licensed under the MIT License. See the LICENSE file for more details.

Security

For security concerns, please see our Security Policy.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Author

Vladimir Blagojevic ([email protected])

Early reviews and guidance by Madeesh Kannan