CSV to Markdown
Runs locally in your browserConvert CSV into Markdown tables or compact AI-ready context
CSV input
Paste, drop or upload a CSV file. Ctrl/Cmd + Enter converts.
Paste CSV, upload a file or load the example. Output updates automatically.
Options
Trims cells and headers in every output mode. Your source text is not changed.
Writes | as \| in Markdown tables so cells cannot break the table.
Lowercase snake_case names, column_N for empty headers and unique suffixes for duplicates.
What CSV to Markdown does
It turns CSV into three representations: a conventional Markdown table for documentation, AI-Ready Markdown with a dataset heading, row and column counts and an inferred schema, and Compact AI Context that keeps the data as a fenced CSV block under a short schema. A Markdown table is not automatically smaller than CSV, so the tool shows estimated token counts for the original CSV and every output and lets you pick the representation that fits your documentation or AI agent.
How to use it
- 1.Paste CSV, drop or upload a .csv file, or load the example. The first row is the header.
- 2.Choose Markdown Table, AI Context or Compact. Output updates automatically; Convert re-runs it and Stop cancels long work.
- 3.Open Options for columns, an inclusive row range, delimiter, alignment, trimming, pipe escaping and header normalization.
- 4.Copy the result or download it as a .md file.
Parsing, schema and options
- •Parsing uses Papa Parse: quoted values, escaped quotes (""), multiline cells, a UTF-8 BOM and comma, semicolon or tab delimiters (auto-detected or forced).
- •Schema types (integer, decimal, boolean, date, datetime, string, empty/null) are advisory and never change your values. Dates must be real calendar dates in ISO form, so 2026-02-31 stays a string.
- •Markdown tables escape | as \| and write line breaks as <br>. Compact context re-serializes the data as comma-separated CSV inside a fence longer than any backtick run in the data.
- •Statistics cover rows, columns, empty cells, duplicate rows, unique values and low-cardinality columns (at most 10 distinct values that repeat).
Limits, token estimates and privacy
Everything runs in your browser. Large inputs are parsed in a Web Worker that you can stop; the preview shows the first 200 rows while copy and download include every row. Inputs over 10 MB show a notice, the hard limit is 50 MB for pasted text and files, and at most 100,000 data rows are processed (the output says so when truncated). Token counts come from the o200k_base tokenizer (js-tiktoken) loaded inside the Worker; other models tokenize differently. If the tokenizer or Worker is unavailable, a labeled heuristic is used instead, and main-thread fallback is limited to 1,000,000 characters. Generated headings such as Dataset, Rows and Schema stay in English so prompts look the same in every language.
