OnsetLab

OnsetLab is a Python framework for building tool-calling AI agents that run entirely on your local machine. It combines REWOO planning with ReAct fallback reasoning to create self-correcting agents that can connect to MCP servers and use any Ollama-compatible model without requiring API keys or cloud dependencies.

• REWOO + ReAct Planning - Plans all execution steps upfront and runs them in sequence, automatically falling back to step-by-step reasoning when tool calls fail
• MCP Server Integration - Connect to GitHub, Slack, Notion, filesystem, web search, and any other MCP server with a single line of configuration
• Self-Correcting Execution - Automatically retries with corrected parameters when tool calls fail, with ReAct fallback kicking in when plans need adjustment
• Any Ollama Model Support - Works with Qwen, Mistral, Hermes, Gemma, and any other Ollama-compatible model to fit your hardware requirements
• Local-First Architecture - Run everything on your own machine with your own models and tools, maintaining full control over your data and infrastructure
• Simple Installation - Get started with pip install onsetlab and configure your agent with just a few lines of Python code
• Flexible Deployment - Export your agents as Docker containers, YAML configurations, or standalone scripts for production deployment
• Built-in Tools - Includes Calculator, DateTime, and other utility tools out of the box, with easy extensibility for custom tools
• Interactive Playground - Test and experiment with agents directly in the browser-based playground before deploying to production