Sheet Slim Gateway

Compress messy real-world spreadsheets into agent-readable structured summaries

Problem

When AI agents process real business spreadsheets, a medium-sized sheet can blow past a million tokens if read fully, immediately exceeding context windows. Most agents fall back to chunk sampling, which misses rows and misreads structure, a wall spreadsheet-tool builders keep hitting.

Solution

A layer between agents and sheets: parse xlsx/csv structure, formulas, merged cells and dirty data, emit a semantically faithful compact schema (columns, stats, anomaly flags), and support on-demand drill-down into a few raw rows so agents get usable signal with small token budgets.

Users

Teams building spreadsheet AI tools and analysts who need agents to read Excel

MVP

Upload xlsx/csv and parse structure and formulas; output compact schema summary JSON; per-column stats and anomaly flags; drill-down by row/column range; API plus a simple playground

Revenue

Usage-based billing per API parse call; self-hosted one-time license for data-compliance-sensitive teams

Why now

Enterprise spreadsheets are one of the stickiest LLM use cases, and context limits plus dirty data aren't going away soon