PraisonAI
PraisonAI is a framework for building autonomous AI agents that can research, plan, and execute tasks. It solves the problem of writing boilerplate code for agent orchestration by providing a five-layer stack covering prompts, context, tools, loops, and graphs. It is for developers who want to deploy AI agents in their applications quickly.
🎬 Self Reflecting AI Agents BEATS CrewAI and AutoGen in Accuracy! PraisonAI 2.0 · Mervin Praison
✨ Key features
- Five-layer agent stack: prompt, context, harness, loop, graph
- Managed agents: run tools or entire agents on remote sandboxes
- Support for 100+ LLMs including OpenAI, Anthropic, Gemini, local models
- AgentFlow for graph-based workflows with routing, parallel, and loops
- Built-in doom-loop detection and execution controls
- CLI and UI options for managing agents and workflows
🎯 Use cases
- Automate research and analysis by gathering data from multiple sources
- Generate, debug, and refactor code with AI agents
- Create content like blog posts and documentation with multi-agent teams
- Build data pipelines that extract, transform, and analyze data
- Deploy 24/7 customer support bots on messaging platforms
📦 Installation
🧰 Requirements: Python 3.x or Node.js for JavaScript SDK; requires an OpenAI API key for the quickstart, but supports many providers.
pip install praisonaiagents
For the full CLI and UI, use:
pip install praisonai
For JavaScript:
npm install praisonai
🚀 Usage
from praisonaiagents import Agent
agent = Agent(instructions="You are a senior data analyst.")
agent.start("Analyze the top 3 tech trends of 2026 and format as a markdown table.")
❓ FAQ
How do I install PraisonAI?
Install the core SDK with pip install praisonaiagents. For the full CLI and UI, use pip install praisonai. For JavaScript, use npm install praisonai.
What LLMs are supported?
PraisonAI supports 100+ LLMs including OpenAI, Anthropic, Gemini, and local models like Ollama. Examples for each provider are in the repository.
Can I run agents on remote sandboxes?
Yes, you can use tools_run_on to run tools on Docker, E2B, Modal, Daytona, or FlyIO, or run_on to run the whole agent on a hosted runtime like Anthropic.
What is AgentFlow?
AgentFlow is a way to define multi-agent workflows as a graph, with steps like routing, parallel execution, and loops. It can be expressed in Python or YAML.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.