Phidata¶
What it is¶
Phidata is a framework for building AI assistants with memory, knowledge, and tools. It allows you to turn any LLM into an "Assistant" that can store data in a database, search across local files, and take actions using Python functions.
What problem it solves¶
It bridges the gap between raw LLMs and functional agents by providing out-of-the-box support for RAG (Retrieval Augmented Generation), persistent session storage, and a structured way to define tools.
Where it fits in the stack¶
[Framework / Agent / Knowledge] - A comprehensive framework for building robust, knowledge-enabled agents.
Typical use cases¶
- Research agents that can search the web and read PDF files
- Coding assistants with access to a local codebase
- Customer support agents with persistent memory of past conversations
Strengths¶
- Simple API: Very easy to get started with a few lines of code.
- Database Integration: Built-in support for PostgreSQL, Pinecone, and more for memory/knowledge.
- Built-in Tools: Includes many ready-to-use tools like DuckDuckGo, Shell, and SQL.
Limitations¶
- Ecosystem: Smaller than LangChain or LlamaIndex.
- Transition: Recently underwent a major rebranding/v2 transition to Agno.
When to use it¶
- When you need to build a single agent with RAG and memory capabilities quickly.
- If you want a Python-native, lightweight framework.
When not to use it¶
- For extremely complex multi-agent graphs (consider LangGraph).
CLI examples¶
# Initialize a new phidata project
phi init
# Start a phidata environment (e.g., with PostgreSQL)
phi start
# Stop the phidata environment
phi stop
Getting started¶
Installation¶
pip install phidata openai duckduckgo-search
Working Example¶
from phi.agent import Agent
from phi.model.openai import OpenAIChat
from phi.tools.duckduckgo import DuckDuckGo
# 1. Create the assistant with a tool
agent = Agent(
model=OpenAIChat(id="gpt-4o"),
tools=[DuckDuckGo()],
description="You are a helpful assistant that can search the web.",
show_tool_calls=True,
markdown=True,
)
# 2. Run a query
agent.print_response("What is the latest news about AI agents?", stream=True)
Licensing and cost¶
- Open Source: Yes (MIT License)
- Cost: Free
- Self-hostable: Yes
Related tools / concepts¶
- Agno (Successor to Phidata v2)
- LlamaIndex
- Agent Protocols (MCP)
Sources / References¶
Contribution Metadata¶
- Last reviewed: 2026-03-02
- Confidence: high