Recalc is a tool that helps you use spreadsheet cells as individual AI terminals, allowing multiple agents to share data and handle complex workflows.
Imagine this: You open Excel or Google Sheets as usual. But instead of just entering numbers into each cell, you command an AI, “Analyze this data and write a draft report.” In the next cell, another AI agent (an AI that executes tasks autonomously based on user instructions) is running code in real-time to find the necessary information. It’s as if several smart AI assistants have gathered in one ledger to hold a meeting together.
Recently, on November 8, 2026, the open-source spreadsheet tool ‘Recalc’ was released on GitHub, and it is bringing this future into reality [Source 11].
Why is this important?
Until now, working with AI agents was usually confined to the narrow space of a chat window. If an agent needed to exchange data with other sessions while performing complex tasks, it was cumbersome to save files one by one or manually link complex programming interfaces (APIs).
Recalc turns the tool we are most familiar with—the ‘spreadsheet’—into a hub for AI collaboration. By allowing individual agent sessions to share data and communicate within a common grid, it has become much more intuitive to manage AI evaluations or complex business processes (workflows) [Source 1, Source 11].
Understanding it simply
In short, while existing AI agent tasks were like talking to a ‘researcher thinking alone in a solitary room,’ Recalc provides an environment where ‘several researchers draw together on one large blueprint.’
- A passage for data: Spreadsheet cells are no longer just containers for numbers. Each cell becomes a small terminal window (a command input box) and an independent room where AI agents like Claude Code (Anthropic’s AI coding agent) or Codex reside [Source 10, Source 13].
- Data sharing: The cells are connected. The result generated by an AI in cell A can be immediately utilized by another AI in cell B. This is possible through a technology called MCP (Model Context Protocol, a standard for AI models to communicate with external data or environments), and this protocol acts as an adhesive that organically glues the entire app together [Source 1, Source 11].
There is no need to explain things at length in a chat window; you can modify an AI agent’s behavior instantly just by editing a cell in the table [Source 10].
Current status
Recalc is currently being developed as an open-source project that anyone can participate in. Its main features are as follows:
- Field testing: The developer created this tool to perform actual AI performance evaluations (benchmarks) at their own startup. For example, there is a demo case where they test how accurately an AI predicts whether a document submitted to the U.S. Food and Drug Administration (FDA) will pass [Source 1].
- Operating environment: It is currently designed to work smoothly in a macOS environment. It can be run on Linux, but it has not yet undergone sufficient testing, and it does not yet support Windows [Source 1].
- Security: When linked with tools like Claude Code, they typically run in the local environment of the user’s computer. It is designed with security in mind, as it communicates directly with the model without a separate backend server [Source 3].
What will happen in the future?
Recalc shows the potential to completely change the way humans and AI collaborate. In the future, a time may come when not only data analysts but also developers and planners use spreadsheets to coordinate dozens of AI agents. If it continues to grow, including support for Windows and becoming easier to use in a wider variety of environments, it is expected to grow from a tool for simple repetitive tasks into a complex enterprise business management tool [Source 1].
MindTickleBytes AI Reporter’s View
This is a case that well illustrates the fact that ‘the more familiar the tool, the faster an AI’s capabilities expand.’ The most powerful weapon of Recalc is that handling AI through a table-editing method we use in our daily lives is far more intuitive than learning complex new commands. The era of complex AI agents is closer to us than we think.
References
- Show HN: Recalc – Spreadsheet where cells can be Claude Code or Codex terminals
-
[ClaudeCode by Anthropic AI Coding Agent, Terminal, IDE](https://claude.com/product/claude-code) -
[Recalc - Spreadsheet with Claude Code · Hacker News Zeli](https://zeli.app/story/50413970) - Recalc Spreadsheet Lets Cells Run Claude Code or Codex
- recalc/demos/done-means-green/repo/.codex at main - GitHub
- recalc/CLAUDE.md at main · sshah03/recalc · GitHub
- Store values
- Calculate formulas
- Execute terminal commands and run AI agents
- All of the above
- Controlling agents by editing cells without interrupting the chat flow
- Waiting for the agent to choose randomly
- Learning a dedicated programming language
- Blockchain
- MCP (Model Context Protocol) server
- Dedicated cloud API