https://txtfetch.com/changelog/
Every entry below is dated to the commit that shipped it, and links to the page that proves it. Three entries are corrections, not features — we'd rather list those than hide them.
Subscribe: /changelog.xml
The site said a merged cell always survives, and a PDF table comes back with its structure. Production says otherwise. HTML and Word lose the merge, and the cells after it move one column left. A PDF table comes back as lines of text. The new page shows real output for seven formats, with and without a merged cell.
The site said a PDF with only an owner password extracts. It does not: the API refuses every PDF that has encryption. A locked Word or Excel file fails with extraction_failed, not encrypted. A ZIP with an encrypted member returns success and the member name only. The new page shows real output and unlock scripts.
Deleted words come back from a Word or OpenDocument file, and nothing marks them. Moved text comes back twice in Word. An ODT deletion comes back at the start of the document. The new page shows real output and an accept-all script. The Word to text tool page no longer says the API leaves deletions out.
Headers, footers, footnotes, endnotes and comments all come back as plain text. A Word comment carries the reviewer's name, so the personal-data page no longer says names are not a leak path. A Word header comes back once for each section, not once for each page. The new page shows real output for Word, ODT, PDF, PowerPoint and Excel.
The site said email headers come back. For .eml and .mbox they do not. Only the body and the attachment text come back. An Outlook .msg does return its subject, sender name, To and Cc. A ZIP returns every file, with each path on its own line. The new page shows real output for each case.
The site never said, and about ten pages said it wrong. A PDF page is read by OCR when it has 10 or fewer typed characters. Pictures in Word, PowerPoint and Excel are read too, and their text lands at the end. The new page shows real output for each. The old whole-document claim and the premium advice are gone.
The site never said. The words of a link come back. The address behind them does not, for HTML, Word, PowerPoint and email. PDF and Excel return the address as a separate line. The new page shows real output for each and short snippets to read the URLs yourself.
The site never said. txtfetch does not decode barcodes or QR codes. A printed caption under a barcode comes back as ordinary text. The new page shows real output for each case and a short way to decode the codes yourself.
The site never said what txtfetch returns for a filled-in form. PDF form values come back after the last page, keyed by field name. XFA values come back too. Word form values come back inline. The new page shows real output for each.
The site never said what the page field means. PDFs return a page number on format=json. Excel, PowerPoint, Word, web pages, email and images return null. The new page shows each result and a chunker that keeps the page range.
The site never said what txtfetch does with equations. Word equations vanish and the response still says success. PDF math loses its layout. The new page shows real output for each input and the fix for each.
The site never said what txtfetch does about injected instructions. It does nothing: every character comes back. The new page shows where hidden text hides and adds a paste-in checker for invisible characters.
txtfetch ships no MCP server, so the site never answered whether an agent like Claude Code or Cursor could call it as a tool. The new page gives a tested one, in Python and Node.js, that you run yourself.
The site never said whether txtfetch reads handwriting. It doesn't, on the standard tier. The new page states that plainly, and shows what comes back for each kind of input.
The homepage said every request 'always answers with text.' It does not: a bad request returns an error code. The format wall's '+ ~976 more' was also hard-coded and stale; it is now derived from the checked Tika coverage count.
/docs/re-indexing answers the daily-churn question. txtfetch keeps no content cache, so a full nightly run pays for every file. The page gives four steps to pay for changed files only. It adds a monthly cost table and a stdlib Python manifest.
/ingest now covers keyword search engines as well as vector stores. The new pages are Elasticsearch, OpenSearch, Meilisearch, Typesense, and Postgres full-text search. Each page names the point where a long document stops counting. On Postgres, that is word position 16,383.
A new page traces one document hop by hop inside txtfetch, from the edge to the S3 buffer to usage records. The premium-tier row states plainly that it is switched off by default today.
This page answers it: the full API surface, the six response fields, and why there is nothing to export. It also names three replacements — self-hosted Tika, an open-source parser, another vendor — and what each one costs you to take back.
GPT, Claude, and Gemini all read PDFs natively. This page names the documented token cost per vendor. It also names the formats none of them accept, and where txtfetch's own vision-language tier fits in.
The /sources page names every place a document can live before txtfetch reads it. It states which ones hand over a URL, and which need a file upload.
Paste text that came out of any extractor and see the damage signals. Each one links to the fix that explains it, and nothing is uploaded.
txtfetch stops at extracted text. /ingest covers the chunk, embed, and upsert step for five vector stores, cited and dated.
/compare/libraries checks txtfetch against Docling, MarkItDown, PyMuPDF4LLM, Marker, and MinerU on licence, GPU need, format coverage, and OCR.
/languages covers nine document languages: legacy encodings, chunking risk, and where OCR support actually stops.
/solutions claimed txtfetch always kept multi-column PDFs in reading order. It does not: the engine gets most PDFs right, and /fixes/columns-out-of-order covers the rest. The page now says so.
No plugin ships for any of them. /integrations shows the HTTP step each platform already has, pointed at txtfetch.
Apache Tika is free. /compare/self-hosted-tika shows the pinned versions, the from-source Tesseract build, and the runtime limits behind the question.
/glossary defines PDF text layers, OCR, mojibake, OOXML, and chunking in plain language. Each term links to its full page.
/diff shows ten committed documents with the raw parser text next to the corrected text, with every difference marked.
/extract/captions states exactly what comes back from an .srt, .vtt, .mp4, .mp3, or .mkv file. A file with no caption track and no tag data returns an error, not an empty success.
That figure counted every media type Tika can detect, not every one it can parse. /formats/coverage now lists the real, checked count, measured against the exact jar this build ships.
/tools/file-to-text detects a file's real type from its bytes and extracts its text in the browser, with no upload.
/tools/doc-to-text reads the 97-2003 binary Office formats (.doc, .xls, .ppt) in the browser, alongside the existing DOCX/XLSX/PPTX tools.
/tools/image-ocr-check measures resolution, focus, skew, and inversion from a photo's actual pixels, in the browser.
Copy on /tools/subtitles-to-text said the API returns a title, artist, duration, and codec for media files. It has no such field. A success carries the extracted text, the content type, the byte and character counts, and whether OCR ran.
/tools/email-to-text extracts a message and its attachments recursively, in the browser.
/tools/docx-to-text and its XLSX and PPTX siblings read modern Office files in the browser. No upload, no API call.
Append .md to any URL for a plain-text twin. /llms.txt indexes the whole site by section, and /llms-full.txt is the whole corpus in one file.
/fixes covers symptoms like mojibake, missing spaces, and scrambled columns: what causes each one, and how to fix it.
/solutions shows the same API applied to different jobs, with the parts of the response each job actually uses.
/docs covers authentication and both ways to call the API. It also covers response formats, async jobs and webhooks, idempotency keys, and the full error reference.
/status shows current status and 90-day uptime history for the extraction API. No login is required.
/extract lists format-specific pages covering PDF, Office, email, HTML, images, and more, each with its own working curl.
what this page is
This page has no roadmap section and no deprecation policy. The one fact worth stating about versioning: every endpoint lives under /v1. Everything else here already shipped.
Create an account and get an API key in minutes. The free Hobby plan needs no card.
Get started →