Phidata - Building Advanced AI Agents

Phidata is a Python framework for building advanced AI agents with built-in persistent memory, knowledge base integration (RAG), and tool-use capabilities. It empowers developers to create autonomous, context-aware AI applications that retain information and interact with external systems.

beginner15 min7 steps

The play

Install Phidata
Set up your Python environment by installing the Phidata framework and any necessary LLM providers. Replace `ollama` with your desired provider (e.g., `openai`, `anthropic`).
Prepare Your LLM
Ensure your chosen Large Language Model (LLM) is accessible. If using Ollama, make sure the Ollama server is running and your desired model (e.g., `llama3`) is pulled locally. For cloud LLMs, configure your API keys.
Initialize a Basic Phidata Assistant
Create an `Assistant` instance, linking it to your chosen LLM. This forms the core of your AI agent and defines its initial persona.
Understand Agent Memory
Phidata agents come with built-in memory management. Explore how to configure different memory types or let the framework handle conversational history automatically to ensure context retention across interactions.
Integrate Knowledge Bases (RAG)
Leverage Phidata's Retrieval Augmented Generation (RAG) capabilities to link your agent to external data sources, documents, or databases. This allows the agent to retrieve and incorporate factual information beyond its training data.
Empower with Tools
Define and register custom Python functions as tools. These tools allow your agent to perform specific actions like calculating, searching the web, or interacting with external APIs, giving it autonomous capabilities.
Run and Test Your Agent
Engage with your agent through its `print_response` or `run` methods. Observe how it leverages its components to provide intelligent and context-aware answers, and debug its behavior as needed.

Starter code

```python
from phi.assistant import Assistant
from phi.llm.ollama import Ollama

# --- Prerequisites --- 
# 1. Install Phidata: pip install 'phidata[ollama]'
# 2. Start Ollama server: ollama serve
# 3. Pull a model: ollama pull llama3

assistant = Assistant(
    llm=Ollama(model="llama3"),
    description="You are a helpful assistant.",
    show_tool_calls=True,
)

# Interact with the assistant
assistant.print_response("Hello, how are you?", stream=False)
assistant.print_response("What is the capital of France?", stream=False)
assistant.print_response("Can you tell me more about Phidata?", stream=False)
```