Pi, a minimalistic open-source harness compatible with more than 1,500 models and variants, has recently reached version 1.0.0, signaling its maturity according to its developers. Designed as an alternative to platforms like Claude Code or Codex, Pi does not rely on models from companies such as Anthropic or OpenAI. Alongside its release, a more robust version called Pi Durable was introduced, capable of withstanding "catastrophic" failures, ensuring greater reliability in its operations. In the world of artificial intelligence, chatbots are often the most visible applications—systems that answer questions when asked. However, the potential of generative AI extends far beyond simple conversations. When paired with the right tools, these systems can generate code, read and analyze files, and even execute commands, making them valuable for a wide range of tasks. The Pi harness is a minimalist yet extensible AI agent, designed to be both simple in structure and flexible in use. In AI systems, two main components are typically distinguished: the model itself, which functions like the core of a reactor by generating text or multimedia content, and the harness, which surrounds the model to enable broader functionality. The harness allows the model to access files, search the internet, and interact with external tools, expanding its capabilities beyond text generation alone. Unlike some existing platforms that limit users to specific models, such as Claude Code, which is restricted to Anthropic models, Pi offers greater flexibility. Users can choose from a wide range of models—commercial, open-source, or local—and can even switch or replace models as needed. This adaptability, combined with its minimalist design, positions Pi as a versatile tool for developers and researchers looking for an open and customizable AI harness.