PraisonAI

PraisonAI is a Python framework for developers building AI agents that research, write code, or carry out tasks across apps. It supports single agents and coordinated teams for work that needs several steps or specialist roles.

• Agent workflows: Combine task handoffs, conditional routing, parallel work, and repeated review in workflows defined with Python or YAML.
• Memory and knowledge: Agents can retain information across runs, retrieve relevant material from your documents, and summarize context as conversations grow.
• Research and coding: Agents can search and fetch web pages, write and debug code, or coordinate Claude Code, Gemini CLI, and Codex.
• Tools and controls: Connect agents to tools through MCP, require human approval for actions, validate inputs and outputs, and stop runs that hit budget limits or repeat without progress.
• Support bots: Build bots for Telegram, Discord, or Slack that use memory and retrieved knowledge to answer questions.

PraisonAI is open source under the MIT license. It works with local models through Ollama and cloud providers including OpenAI, Anthropic, and Google Gemini. Using a cloud model sends model requests to that provider; running the framework locally doesn't make those requests local.

Agents can run on your machine or in self-hosted Docker containers. You can also keep the agent loop local while moving shell, file, and code tools to a shared Docker sandbox or cloud services such as E2B, Modal, and Daytona. Custom tools remain on the local machine in that arrangement. Whole-agent hosting through Anthropic moves model calls, the agent loop, and tools to its hosted runtime.