Pi for Excel is an open-source, experimental AI sidebar add-in for Microsoft Excel, designed to empower users with AI-driven data analysis, research, and workbook manipulation. It targets Excel power users, researchers, and anyone seeking to automate complex tasks within spreadsheets, offering a significant productivity boost by integrating advanced AI capabilities directly into the familiar Excel environment.

How It Works

This project is built as an Office taskpane add-in using Vite for bundling and Lit for its web component UI. The core functionality is powered by pi-agent-core, which manages the agent's tool-use loop and state, while pi-ai provides multi-provider LLM abstraction (supporting Anthropic, OpenAI, Google Gemini, and GitHub Copilot). A key design pattern is "auto-context injection," where the AI automatically receives the workbook's structural blueprint, current selection, and recent cell changes before each turn, eliminating the need for manual context setup. The add-in uses Office.js to interact with the Excel workbook, with a "workbook coordinator" serializing mutating calls to prevent conflicts between session tabs.

Quick Start & Requirements

To install, download the manifest.prod.xml file and upload it via Excel's "My Add-ins" -> "Upload My Add-in" feature. For development, Node.js (version 20 or higher) and mkcert for local HTTPS are required. After cloning the repository and running npm install, generate local certificates and start the development server with npm run dev. Sideloading the development manifest into Excel requires specific steps for macOS or Windows.

Highlighted Details

• Multi-Model Support: Seamlessly switch between AI models from Anthropic, OpenAI, Google Gemini, and GitHub Copilot, or configure custom OpenAI-compatible gateways.
• Extensive Toolset: Features 16 core tools for reading/writing cells, modifying workbook structure, applying formatting, tracing dependencies, explaining formulas, managing comments, and more.
• Workbook Recovery: Automatic checkpoints are created before any data mutation, allowing for one-click rollback from the sidebar if an operation goes wrong.
• AI-Generated Extensions: Users can install sidebar extensions (mini-apps) generated directly by the AI via the extensions_manager tool.
• Integrated Web Search: Includes opt-in web search capabilities (Jina, Serper, Tavily, Brave) to fetch and incorporate external information directly into Excel.
• Session Management: Supports multiple session tabs per workbook with auto-save/restore functionality and a /resume command.

Maintenance & Community

The project is maintained by Thomas Mustier and leverages core agent framework components developed by @badlogic (Mario Zechner). No specific community channels (like Discord or Slack) or sponsorship details are provided in the README.

Licensing & Compatibility

The project is released under the MIT license, which permits broad usage, including commercial applications and integration into closed-source projects, with minimal restrictions.

Limitations & Caveats

This add-in is described as "experimental." Development requires setting up local HTTPS, and certain integrations like OAuth may necessitate a CORS proxy. Advanced features such as Python, LibreOffice, or tmux bridges have specific external software prerequisites (Python, LibreOffice, tmux) that must be installed separately.