Agno¶

What it is¶

Agno is a lightweight Python framework for building multi-modal agents with memory, knowledge, and tools. As the successor to Phidata v2, it emphasizes high performance and scalability. By July 2026, Agno has integrated full support for FastMCP 3.0 and is optimized for the latest frontier models including Gemma 3, Claude 4.8, and GPT-5.5.

What problem it solves¶

Agno simplifies the transition from a single agent prototype to a production-ready system. It provides a stateless, session-scoped runtime that can be served as a FastAPI backend, making it easy to deploy agents as horizontally scalable services while maintaining complex agent state in external databases.

Where it fits in the stack¶

Layer 6: Agents & Orchestration — A performance-oriented framework for building and serving agentic software.

Typical use cases¶

FastMCP Tool Servers: Building and hosting tool-rich servers using the FastMCP 3.0 protocol for seamless discovery by other agents.
High-performance multi-modal agents: Native support for Vision, Audio, and Text models from providers like Anthropic and OpenAI.
Edge Intelligence: Deploying lightweight agents using Gemma 3 for local-first or privacy-preserving workflows.
Production-grade agents: Serving agent logic via FastAPI for integration into web or mobile applications.

Strengths¶

FastMCP 3.0 Integration: Native support for the Model Context Protocol (MCP) 3.0 standard for tool and resource discovery.
Performance: Optimized for low latency and high throughput, critical for the fast inference speeds of 2026 frontier models.
Stateless Runtime: Designed to be horizontally scalable out of the box, delegating state management to robust backends.
Multi-modal: Native support for various model modalities, allowing for seamless integration of multimodal reasoning.

Limitations¶

New Rebrand: As the successor to Phidata, some documentation and legacy links might still refer to the old name.
Python Only: Currently focused exclusively on the Python ecosystem.

When to use it¶

When building agents that need to scale horizontally in production.
For projects requiring strong multi-modal support across different model families.
If you require native FastMCP 3.0 support for building tool-enabled services.

When not to use it¶

If you require a TypeScript-native framework (consider Bee Agent Framework).
For very simple, synchronous scripts where the overhead of a framework isn't justified.

Getting started¶

Installation¶

pip install agno openai duckduckgo-search

Basic Usage (with Gemma 3)¶

from agno.agent import Agent
from agno.models.ollama import Ollama
from agno.tools.duckduckgo import DuckDuckGo

# 1. Create the agent with a tool and Gemma 3 via Ollama
agent = Agent(
    model=Ollama(id="gemma3:27b"),
    tools=[DuckDuckGo()],
    description="You are a helpful AI assistant running locally.",
    markdown=True
)

# 2. Run a query
agent.print_response("What is the latest status of the MCP 3.0 Task Protocol?")

CLI examples¶

# Initialize an Agno project
agno init

# Start the Agno serving environment (FastAPI based)
agno serve

# Manage agent sessions via the CLI
agno sessions list

API examples¶

Creating a FastMCP Server with Agno¶

from agno.agent import Agent
from agno.mcp.server import FastMCPServer

# Define an agent that will act as a tool provider
tool_agent = Agent(
    name="LogAnalyzer",
    instructions="Analyze logs for security patterns",
    tools=[...]
)

# Host the agent via FastMCP 3.0
app = FastMCPServer(agents=[tool_agent])

if __name__ == "__main__":
    app.run(port=8000)

Sources / references¶

Contribution Metadata¶

Last reviewed: 2026-07-21
Confidence: high