any-agent

A single interface to use and evaluate different agent frameworks.

[!IMPORTANT] any-agent is in soft deprecation. It started as a research project to compare agent frameworks and refine a minimal common surface. That distillation has graduated to its own package: mozilla-ai-tinyagent (PyPI: mozilla-ai-tinyagent). For new projects we recommend using tinyagent directly.

any-agent continues to be published and we'll take security/bug-fix PRs, but no new features are planned. Reach for any-agent when you specifically need to run or evaluate agents across multiple frameworks (Agno, Google ADK, LangChain, LlamaIndex, OpenAI Agents SDK, smolagents) under one API. If you only need the core agent loop, mozilla-ai-tinyagent is the leaner path.

Documentation

• Agents
• Tools
• Tracing
• Serving
• Evaluation

Supported Frameworks

Planned for Support (Contributions Welcome!)

Open Github tickets for new frameworks

Requirements

• Python 3.11 or newer

Quickstart

Refer to pyproject.toml for a list of the options available. Update your pip install command to include the frameworks that you plan on using:

pip install 'any-agent'

To define any agent system you will always use the same imports:

from any_agent import AgentConfig, AnyAgent

For this example we use a model hosted by Mistral, but you may need to set the relevant API key for whichever provider being used. See our Model Configuration docs for more information about configuring models.

export MISTRAL_API_KEY="YOUR_KEY_HERE"  # or OPENAI_API_KEY, etc

from any_agent.tools import search_web, visit_webpage

agent = AnyAgent.create(
    "tinyagent",  # See all options in https://docs.mozilla.ai/any-agent/
    AgentConfig(
        model_id="mistral:mistral-small-latest",
        instructions="Use the tools to find an answer",
        tools=[search_web, visit_webpage]
    )
)

agent_trace = agent.run("Which Agent Framework is the best??")
print(agent_trace)

[!TIP] Multi-agent can be implemented using Agents-As-Tools.

Cookbooks

Get started quickly with these practical examples:

• Creating your first agent - Build a simple agent with web search capabilities.
• Creating your first agent evaluation - Evaluate that simple web search agent using 3 different methods.
• Using Callbacks - Implement and use custom callbacks.
• Creating an agent with MCP - Integrate Model Context Protocol tools.
• Serve an Agent with A2A - Deploy agents with Agent-to-Agent communication.
• Building Multi-Agent Systems with A2A - Using an agent as a tool for another agent to interact with.

Contributions

The AI agent space is moving fast! If you see a new agentic framework that AnyAgent doesn't yet support, we would love for you to create a Github issue. We also welcome your support in development of additional features or functionality.

Running in Jupyter Notebook

If running in Jupyter Notebook you will need to add the following two lines before running AnyAgent, otherwise you may see the error RuntimeError: This event loop is already running. This is a known limitation of Jupyter Notebooks, see Github Issue

import nest_asyncio
nest_asyncio.apply()