Recalc Spreadsheet Lets Cells Run Claude Code or Codex

A new open-source spreadsheet tool transforms how developers run AI-assisted evaluations by turning individual cells into active Claude Code or Codex terminal sessions. Recalc, officially launched on November 8, 2026, bridges traditional spreadsheet workflows with agentic coding environments, letting users orchestrate multiple AI sessions from a single grid interface.

What Recalc Does

Recalc reimagines the spreadsheet as a control plane for AI development work. Each cell can host a full terminal session running Claude Code or Codex, while a side panel maintains a lead agent session. Cells reference each other by name or coordinate, enabling data flow between independent AI sessions. The tool also exposes an MCP server that lets sessions communicate across the grid, share context, and hand off tasks without manual copy-paste.

Developer sshah03 built Recalc to solve a personal workflow problem: running evals and benchmarks required constant context switching between terminal windows and tracking which agent had which information. By embedding terminal sessions directly in cells, Recalc keeps all work visible in one interface. If a cell-based session needs more room, users can pop it out into a standalone terminal window while maintaining state continuity.

Release Date and Availability

Recalc was officially released on November 8, 2026 as a free, open-source project on GitHub. The repository includes setup instructions and examples for connecting Claude Code and Codex sessions to the grid. The tool runs locally and does not require cloud dependencies beyond the AI model APIs themselves.

Use Cases for Evaluation Workflows

The primary design target is AI evaluation and benchmark construction. Developers can dedicate cells to different test scenarios, run parallel agent sessions against the same dataset, and compare outputs side by side. Named cells act as shared variables, letting one agent's output become another's input without exporting CSVs or switching tabs.

• Run multiple Codex sessions testing different code generation approaches in adjacent cells
• Use a lead Claude Code session in the side panel to orchestrate cell-based agents
• Reference cell outputs by name to chain agent workflows across the grid

Beyond evals, the tool suits any workflow that benefits from parallel AI sessions with structured data relationships. Research prototyping, dataset annotation, and multi-agent simulations all fit the model.

What This Means

Recalc represents a shift in how developers interface with agentic coding tools. Instead of treating AI assistants as isolated chat sessions, the spreadsheet paradigm makes agent orchestration explicit and visual. By combining the familiar grid structure with terminal-level access to Claude Code and Codex, the tool lowers friction for complex evaluation workflows while keeping all session state in one place. Its open-source release on November 8, 2026 makes the approach immediately accessible to teams building AI benchmarks and multi-agent systems.