Discover Pi, the minimalist yet powerful AI agent harness that adapts to your workflow, and its integration with Ollama to run language models locally. With support for multiple providers and full customization, this tool is set to change programming forever.

What is Pi and Why is the Developer Community Talking About It?

In the fast-paced world of artificial intelligence, open-source tools continue to gain ground. According to its official website, Pi defines itself as a minimalist agent harness. But what does this mean in practical terms? An agent harness is a framework or environment that allows AI models to interact with your operating system, execute commands, read files, and modify code autonomously.

The core philosophy of Pi is adaptability: instead of forcing you to use a preset workflow, the tool allows you to customize it with extensions, skills, prompt templates, and themes. All of this can be packaged and shared via npm or git.

How to Install Ollama

For those looking to run language models locally and privately, Ollama is one of the most popular solutions. As detailed on its download page, it is available for macOS, Linux, and Windows.

For Windows: Requires Windows 10 or later. The installation in PowerShell is done with the following command:

irm https://ollama.com/install.ps1 | iex

How to Install Pi

Pi can be perfectly integrated with Ollama, allowing you to use local AI models within its interface. Its installation is just as simple and is done from the terminal:

curl -fsSL https://pi.dev/install.sh | sh

It is also available through Node package managers like npm, pnpm, or bun.

Main Features of Pi

  • Multiple providers: Supports more than 15 AI providers, including Anthropic, OpenAI, Google, and of course, Ollama for local models.
  • Tree-based history: Sessions are saved in a tree structure, allowing you to go back to any previous point, create branches, and share the entire session in HTML or as a GitHub Gist.
  • Four modes of operation: Interactive (full TUI), Print/JSON for scripts, RPC for non-Node integrations, and SDK to embed in your own apps.
  • Context engineering: Uses files like AGENTS.md and SYSTEM.md to load project instructions and customize the system prompt, controlling what enters the context window.

Unlike other sealed products, Pi does not include sub-agents or planning modes by default; instead, it invites you to build them yourself through extensions or install third-party packages that do the work your way. A true revolution for customization lovers!