Praison AI

Introduction

PraisonAI runs AI agents on your own computers, even fully offline, so private data never has to leave the building. It is built for highly regulated teams, and free and open source for everyone.

PraisonAI runs AI agents on your own computers, even fully offline, so private data never has to leave the building. It is built for highly regulated teams, and free and open source for everyone.

Private AI that earns trust one task at a time.

python
from praisonaiagents import Agent

agent = Agent(instructions="You are a helpful assistant")
agent.start("Write a haiku about AI")

What is PraisonAI?

PraisonAI is an open-source Python framework for building AI agents. It is still a general-purpose framework: build any kind of agent, with any model, in the cloud or on your own computers. Our company focus is private AI for regulated teams, and this framework is the foundation for that work.

Building something else?

Everything works as before. Nothing has been removed, and every feature page is where it was.

Agents

One agent with instructions and tools

 Multi-agent teams

Several agents working together

 AgentFlow

Step-based pipelines: route, parallel, loop

 Memory

Agents that remember across sessions

 Knowledge

Give agents your documents

 RAG

Retrieve before you answer

 Tools

Let agents call your functions

 MCP

Connect Model Context Protocol servers

 Messaging bots

Telegram, Discord, Slack, WhatsApp

 CLI

Run agents from the command line

 Training

Train agents and fine-tune models

 Browser automation

Agents that use a web browser

 Sandbox

Run agent code in isolation

 Deploy

Docker, cloud and local servers

 UI

Chat interfaces for your agents

 YAML

Define agents without Python

 JavaScript SDK

The same agents in TypeScript and JavaScript

Run fully offline

Agents can run on a model on your own computer, with no internet connection and no cloud API key.

Run fully offline

Install, pull a local model, and run an agent with the network turned off.

Early access: task handover

We are building a way to move each task to your own AI only after it passes a test, and move it back if quality slips. It is in early access, and we are taking on our first design partners.

Task handover (early access)

The five steps, and the parts of the open-source framework it builds on.

Core Components

Diagram

🚀 Production

🛠️ Development

🤖 Agent
Single AI worker

👥 AgentTeam
Multi-agent orchestration

🔄 AgentFlow
Step-based pipelines

⚡ AgentOS
Production deployment

🤖 Agent

Single AI worker with tools and instructions

 👥 AgentTeam

Multi-agent orchestration with sequential/hierarchical process

 🔄 AgentFlow

Step-based pipelines with route, parallel, loop patterns

 ⚡ AgentOS

Production deployment with API, webhooks, scheduler

When to Use What

Diagram

One agent, one task

Multiple agents collaborate

Complex pipelines

Production API

What do you need?

🤖 Agent

👥 AgentTeam

🔄 AgentFlow

⚡ AgentOS

ScenarioUse This
Chat with one AIAgent
Research → Analyze → WriteAgentTeam or AgentFlow
Route to specialistsAgentFlow + route()
Parallel processingAgentFlow + parallel()
Production APIAgentOS

Use Cases

Customer Service

Build intelligent support agents that can handle customer inquiries and resolve issues autonomously.

 Data Analysis

Create agents that can process, analyze, and derive insights from complex datasets.

 Content Creation

Deploy agents that can generate, edit, and optimize content across various formats.

 Process Automation

Automate complex workflows with intelligent agents that can coordinate and execute tasks.

Getting Started

Python

  1. Install
    bash
    pip install praisonaiagents
  2. Set API Key
    bash
    export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
  3. Create Agent
    Single Agent
    from praisonaiagents import Agent

agent = Agent(instructions="You are a helpful assistant")
agent.start("Write a haiku about AI")
    Multi-Agent Team
    from praisonaiagents import Agent, AgentTeam

researcher = Agent(instructions="Research topics")
writer = Agent(instructions="Write content")

team = AgentTeam(agents=[researcher, writer])
team.start()
    AgentFlow Pipeline
    from praisonaiagents import Agent, AgentFlow

researcher = Agent(instructions="Research topics")
writer = Agent(instructions="Write content")

flow = AgentFlow(steps=[researcher, writer])
flow.start("Research AI trends")

TypeScript

  1. Install
    bash
    npm install praisonai
  2. Set API Key
    bash
    export OPENAI_API_KEY="${OPENAI_API_KEY:?Set OPENAI_API_KEY in your shell}"
  3. Create Agent
    Single Agent
    import { Agent } from 'praisonai';

const agent = new Agent({ instructions: 'You are a helpful assistant' });
agent.start('Write a haiku about AI');
    Multi-Agent Team
    import { Agent, AgentTeam } from 'praisonai';

const researcher = new Agent({ instructions: 'Research topics' });
const writer = new Agent({ instructions: 'Write content' });

const team = new AgentTeam({ agents: [researcher, writer] });
team.start();

No Code (YAML)

  1. Install
    bash
    pip install praisonai
  2. Create agents.yaml
    yaml
    agents:
  researcher:
    instructions: Research AI trends
  writer:
    instructions: Write content based on research
  3. Run
    bash
    praisonai agents.yaml

Key Features

Autonomous, Self-Improving Agents
  • Understand natural language
  • Make decisions
  • Execute tasks
  • Learn and persist new skills
 Flexible Architecture
  • Modular components
  • Extensible tools
  • Custom workflows
 Advanced Capabilities
  • Multi-agent collaboration
  • Memory management
  • Tool integration

Why PraisonAI?

Developer First

Modern SDK designed to be intuitive and powerful

 Production Ready

Enterprise-grade with built-in security and scale

 Open Source

Available on GitHub

Low Code Friendly

Easy for non-technical users

How It Works

An Agent receives your instructions, calls the LLM, and returns a result — teams and flows chain multiple agents on top of that.

Diagram
LLMAgentUserLLMAgentUserAgent(instructions=”…”).start(“…”)Send prompt to the selected modelModel responseFinal answer

Best Practices

Pick the right building block

Use Agent for one task, AgentTeam when agents collaborate, and AgentFlow for step-based pipelines with routing, parallel, or loop patterns.

Keep the first example tiny

Start with Agent(instructions="…").start("…"). Add name, role, tools, and Task only as your needs grow.

Set provider keys in the environment

Export OPENAI_API_KEY (or another provider’s key) in your shell. PraisonAI auto-selects a matching default model.

Move to AgentOS for production

Wrap agents in AgentOS when you need an API, webhooks, or a scheduler for production deployment.

Related

Quick Start

Build your first AI agent in minutes.

 Workflows

Learn workflow patterns for multi-step pipelines.