# Phidata Assistants — Memory, Knowledge & Tools in One Class

> Phidata Assistant glues memory, knowledge bases, tool use into one Python class. Add knowledge with PgVector or Lance, free UI playground included.

## Install

Copy the content below into your project:

## Quick Use

1. `pip install phidata duckduckgo-search`
2. Define an Assistant with `description`, `tools`, optional `memory` and `knowledge_base`
3. `assistant.print_response("<your question>")` — or serve the playground at localhost:7777

---

## Intro

Phidata's `Assistant` class is the Python-first agent abstraction — memory, knowledge bases, structured outputs, and tool use plugged in as keyword arguments. No graph, no nodes, no decorators outside the tool registration. Best for: builders who want a Python class they can grow into a production agent, not a framework that takes over their whole codebase. Works with: any LLM via LiteLLM, PgVector / Lance / SingleStore for knowledge. Setup time: 5 minutes.

---

### Hello assistant

```python
from phi.assistant import Assistant
from phi.tools.duckduckgo import DuckDuckGo

assistant = Assistant(
    description="Research assistant. Use web search aggressively, cite sources.",
    tools=[DuckDuckGo()],
    show_tool_calls=True,
    markdown=True,
)

assistant.print_response("Compare Pinecone, Weaviate, Qdrant in 2026")
```

### Add memory

```python
from phi.memory import AssistantMemory
from phi.memory.db.postgres import PgMemoryDb

assistant = Assistant(
    memory=AssistantMemory(db=PgMemoryDb(table_name="assistant_memory")),
    ...
)
```

Memory persists across runs. The next session starts with a summary of past conversations.

### Add knowledge (RAG built in)

```python
from phi.knowledge.pdf import PDFUrlKnowledgeBase
from phi.vectordb.pgvector import PgVector

kb = PDFUrlKnowledgeBase(
    urls=["https://example.com/paper.pdf"],
    vector_db=PgVector(table_name="kb"),
)
kb.load(recreate=False)

assistant = Assistant(
    knowledge_base=kb,
    add_references_to_prompt=True,
    ...
)
```

The Assistant retrieves relevant chunks per query and injects them into the prompt.

### Built-in playground

```python
from phi.playground import Playground, serve_playground_app

app = Playground(assistants=[assistant]).get_app()
serve_playground_app("main:app", reload=True)
```

`localhost:7777` gives you a chat UI to test the Assistant — useful in dev, not for prod.

---

### FAQ

**Q: Phidata vs Agno — what's the relationship?**
A: Same team. Phidata is the older brand; Agno is the newer name + cleaner runtime. Phidata still maintained but Agno is where new features land. For new projects, prefer Agno.

**Q: Is Phidata free?**
A: Yes — open-source under Mozilla Public License 2.0. The hosted Phidata Cloud (monitoring + auth) is paid, but everything you run on your own infra is free.

**Q: Can I use Phidata with Claude?**
A: Yes — Phidata uses LiteLLM under the hood. Set `llm=Claude(model='claude-3-5-sonnet-20241022')` or any of 100+ providers. Tool use, structured outputs, and memory all work the same.

---

## Source & Thanks

> Built by [Phidata](https://github.com/phidatahq). Licensed under MPL-2.0.
>
> [phidatahq/phidata](https://github.com/phidatahq/phidata) — ⭐ 13,000+

---

<!-- ZH -->

## 快速使用

1. `pip install phidata duckduckgo-search`
2. 定义 Assistant，写 `description`、`tools`、可选 `memory` 和 `knowledge_base`
3. `assistant.print_response("<你的问题>")` 或在 localhost:7777 起 playground

---

## 简介

Phidata 的 `Assistant` 类是 Python 优先的 agent 抽象 —— 记忆、知识库、结构化输出、工具使用都作为关键字参数接进来。没有图、没有节点、除了工具注册没有装饰器。适合想要一个能长成生产 agent 的 Python 类、而不是一个接管整个代码库的框架的开发者。兼容任何走 LiteLLM 的 LLM 和 PgVector / Lance / SingleStore 知识后端。装机时间 5 分钟。

---

### Hello assistant

```python
from phi.assistant import Assistant
from phi.tools.duckduckgo import DuckDuckGo

assistant = Assistant(
    description="Research assistant. Use web search aggressively, cite sources.",
    tools=[DuckDuckGo()],
    show_tool_calls=True,
    markdown=True,
)

assistant.print_response("Compare Pinecone, Weaviate, Qdrant in 2026")
```

### 加记忆

```python
from phi.memory import AssistantMemory
from phi.memory.db.postgres import PgMemoryDb

assistant = Assistant(
    memory=AssistantMemory(db=PgMemoryDb(table_name="assistant_memory")),
    ...
)
```

记忆跨运行保存。下次会话以过去对话的摘要起手。

### 加知识（内置 RAG）

```python
from phi.knowledge.pdf import PDFUrlKnowledgeBase
from phi.vectordb.pgvector import PgVector

kb = PDFUrlKnowledgeBase(
    urls=["https://example.com/paper.pdf"],
    vector_db=PgVector(table_name="kb"),
)
kb.load(recreate=False)

assistant = Assistant(
    knowledge_base=kb,
    add_references_to_prompt=True,
    ...
)
```

Assistant 按 query 取相关 chunk 注入 prompt。

### 自带 playground

```python
from phi.playground import Playground, serve_playground_app

app = Playground(assistants=[assistant]).get_app()
serve_playground_app("main:app", reload=True)
```

`localhost:7777` 给你一个聊天 UI 测 Assistant —— 开发时用，生产别用。

---

### FAQ

**Q: Phidata 跟 Agno 啥关系？**
A: 同一个团队。Phidata 是旧品牌，Agno 是新名字加干净的运行时。Phidata 还维护但新功能都进 Agno。新项目优先选 Agno。

**Q: Phidata 免费吗？**
A: 免费 —— Mozilla Public License 2.0 开源。托管的 Phidata Cloud（监控 + 认证）付费，但自己跑的部分全免费。

**Q: Phidata 能配 Claude 吗？**
A: 能 —— Phidata 底层走 LiteLLM。设 `llm=Claude(model='claude-3-5-sonnet-20241022')` 或 100+ provider 任意一个。工具、结构化输出、记忆都一样工作。

---

## 来源与感谢

> Built by [Phidata](https://github.com/phidatahq). Licensed under MPL-2.0.
>
> [phidatahq/phidata](https://github.com/phidatahq/phidata) — ⭐ 13,000+


---
Source: https://tokrepo.com/en/workflows/phidata-assistants-memory-knowledge-tools-in-one-class
Author: Phidata