Share a CSV as a link, not an attachment

curl -T data.csv https://rows.page answers in plain text: the public link, an owner link with the token, and one line with rows, columns, size and the expiry date. Add -H "Accept: application/json" for JSON. No key, no account, nothing to install besides curl, and requests.put("https://rows.page/data.csv", data=open("data.csv", "rb")) does the same from Python.

Without an account you get 50 pushes and 1 GB per day per IP, 10 pushes per minute, 100 MB per file. The browser route is the same thing: open the file here, press Share.

A table, not a download prompt. Whoever opens the link can sort, filter (region:EMEA amount:>1000), run SQL, see column stats, and export CSV, Parquet, JSON, JSONL or Markdown, on a phone too. Nobody needs an account to view; viewing is never paywalled on any plan.

The original bytes stay available at /<id>/raw.csv with CORS headers, so a script, a notebook or a web page can fetch the file directly.

The owner link carries a token. Push the next export to the same link, curl -T data.csv -H "Authorization: Bearer rpt_..." https://rows.page/<id>, and it becomes version 2. The page shows how many rows are new, gone or changed against the previous version, matched by a key column you name with ?id_column=sku, and which columns were added or removed.

Identical bytes answer unchanged: true and add nothing. Anonymous links keep up to 20 versions; a free account opens the last 3, Pro keeps 90 days of versions, Max 365.

Anonymous and free links are public but unguessable: 10 random characters, and the page is marked noindex so search engines leave it alone. Pro ($12/month) adds private datasets, 100 datasets, files up to 500 MB; Max ($39/month) goes to 1,000 datasets and 2 GB files. Files over 100 MB upload in 64 MB parts through the API.

Scripts on a plan use an API key (rpk_...) and can upsert by name: curl --data-binary @report.csv -H "Authorization: Bearer rpk_..." "https://rows.page/api/v1/push?key=nightly-report" updates the same dataset every night instead of minting a new link.

A GitHub Actions step is the curl line; the link lands in the job log or a PR comment. Agents use the MCP server at rows.page/mcp, which exposes push, get_dataset, list_datasets, delete_dataset and pin_finding, so a model can publish a result table and read it back with filters instead of pasting rows into chat.

The bin viewer on webhooks.sh has an Open in rows.page button that pushes captured requests as a table. Anything that can send an HTTP PUT can publish a dataset.