aisuite is a lightweight Python library that provides a unified, OpenAI-style Chat Completions API and an Agents API for working across multiple LLM providers. It standardizes interactions with providers such as OpenAI, Anthropic, and Google, allowing developers to change models by altering a single string.
Project overview
The library addresses the problem of integrating disparate LLM SDKs by offering a standardized interface that normalizes provider differences, supporting automated tool calling, an agents framework with prebuilt toolkits, and native Model Context Protocol integration.
Deployment
Refer to project documentation
License
MIT
Best for
Developers who need to integrate multiple LLM providers into a single Python application without managing separate SDKs.
AI engineers building structured agent workflows that require tool calling, prebuilt toolkits, and MCP server integration.
Key capabilities
Provides an OpenAI-style interface for multiple LLM providers, including OpenAI, Anthropic, Google, and Ollama, supporting standard parameters like temperature, max_tokens, and tools.
Allows passing plain Python functions as tools; automatically generates schemas, executes calls, and feeds results back to the model up to a specified max_turns limit.
A first-class API for longer-running, structured agent work, allowing users to declare an agent, run it with a Runner, and attach prebuilt toolkits including files, git, and shell.
Natively supports Model Context Protocol so any MCP server's tools can be handed to a model without boilerplate. Also provides an explicit MCPClient for reusable connections.
Allows adding new providers by implementing a lightweight adapter using a naming convention for automatic discovery.
Limitations and risks
Streaming cannot be combined with max_turns for automated tool calling, requiring manual assembly of tool calls in those scenarios.
Getting started
Install via pip, configure API keys for your chosen providers, import aisuite, and call client.chat.completions.create to make a request. The setup uses a standard OpenAI-style interface.