AnyActions: Developing Your AI Agents with Seamless Tool Integration

Tool usage empowers your LLM agents to interact with external applications via RESTful APIs calls. They are usually either used to do data retrieval (e.g. search a flight schedule) or take action in another application (e.g. file a JIRA ticket).

AnyActions helps you build LLM agents with external tools (web search, check calendar, etc) and manage tool usage authentication easier and faster.

from anyactions import ActionHub
hub = ActionHub()

# Set up your tool real quick
tools = hub.tools(["serpapi_google_search"])

# Call LLMs
response = openai.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "What is the weather in Tokyo next week?"}],
    tools=tools,
)

Features

• Manage tool usage context, authentication, and invocation for LLMs
• Search and plug in available tools from our public database easily
• Request AnyActions agent to generate tool usage context and tool functions for you!

Setup

python >= 3.10

For the SDK set up, see AnyActions-SDK/README.md

Core Design

ActionHub

• The main component that manages tool usage, orchestrates the tool authentication and invocation process
• Manage authentication for tool usage: Create, Read, Update, and Delete (CRUD). Currently, only API keys are supported. More authentication methods will be supported in the future.
• Manage tool usage context for LLMs, like input schema.

Retriever

• Retrieve available tools from AnyActions database
• For unavailable tools, request our AnyActions agent to generate tool usage context and tool functions for you

Actor

• Parse LLM responses for tool usage
• One-to-one client for executing tool functions based on parsed LLM responses

Notes

We named with the term action based on the OpenAI convention. In Anthropic's documentation, it is equivalent to tool. However, we realized that OpenAI also conflates the term tool with action. They are the same concept in our SDK.

Acknowledgements

• Swarm