OnsetLab

OnsetLab is a Python library that lets you build and run tool-calling AI agents directly on your local machine. It connects to various AI models through Ollama (including Qwen, Mistral, Hermes, and Gemma) and enables these agents to interact with tools and services through Model Context Protocol (MCP) servers. This means you can create AI assistants that work with GitHub, Slack, Notion, and other services without relying on cloud APIs or internet connectivity.

The tool combines REWOO planning with ReAct fallback mechanisms, allowing agents to plan all steps upfront and execute them sequentially. When a tool call fails, the system automatically switches to step-by-step reasoning to correct errors and retry with improved inputs. This self-correcting capability makes the agents more reliable in real-world scenarios where API calls might fail or require parameter adjustments.

OnsetLab is particularly valuable for developers who want to build AI applications that maintain privacy, reduce cloud costs, and work offline. It's ideal for creating personalized AI assistants that can access your local tools and data while giving you full control over the underlying models and infrastructure. The ability to export agents as Docker containers, YAML, or scripts makes deployment straightforward for production environments.