Diff Ledger

Upload two comparable CSVs. Diff Ledger joins rows by key column(s) and sorts every pair into Match, Mismatch, A only, or B only — instantly with local rules, optionally double-checked by AI for wording-only differences.

1Upload files
How this works (click to expand)
File A Uploaded CSV File B Uploaded CSV Fast Compare Joins rows by Key Confident results Match, mismatch, unique Low-confidence mismatches Wording differs, needs a look AI semantic review Optional — your choice of engine Cloud model Your API key Local model Runs in-browser Diff Table Every result lands here

Both files are joined by their Key column(s) and every row pair is judged by fast, deterministic rules — free, instant, and always run first. Most rows are resolved with full confidence right there. The rows left over — wording differs but nothing is clearly right or wrong — are the only ones eligible for optional AI review, using whichever engine you choose: your own Cloud API key, or a small model running locally in your browser. Either path's output rejoins the same Diff Table as everything else.

Performance expectations & disclaimers (click to expand)

Fast mode (always runs first, free, instant)

RowsExperience
Up to ~60kInstant, fully interactive
~60k–200kA few seconds, still comfortable
200k+Works, but consider splitting the file

AI review (optional, only re-checks Fast's mismatches)

Rows queued for AIExpect
Up to ~5005–15 min
~500–2,00015–60 min, run in background
2,000+1+ hours — treat as unattended/overnight
AI review speed depends heavily on your device and what else is running. Local AI mode needs a browser with WebGPU (recent Chrome/Edge) — it may be disabled by organization IT policy on locked-down machines. These figures are estimates, not guarantees; the live progress bar during AI review will show a measured, updating ETA once it starts.