FileGizmo

Changelog

What changed, why it matters, and the details we refused to leave rough.

Eight new tools, and a round of hardening

TXT, HEIC and WebP to PDF, grayscale images, zip and unzip, Split Excel and PDF bookmarks, plus fixes for damaged files, phones and very large inputs.

  • Added TXT to PDF, Image to grayscale, HEIC to PDF, WebP to PDF, Zip files, Unzip, Split Excel and PDF bookmarks.
  • Every PDF tool now refuses a damaged or cut-short file when it is chosen, with one plain message, instead of failing at the last step.
  • Auto-redact finds addresses with accented letters, addresses and phone numbers broken across lines, and addresses split between table cells.
  • Fill form works on dense forms on a phone, and Unzip can save every file into a folder you choose in Chrome and Edge.

The eight new tools follow the same rule as the rest: the file is read and written in your browser, and nothing is uploaded. Unzip refuses archives built to expand into far more than they hold, and entries whose names point outside the archive or hide characters that change how a name reads. Split Excel turns each sheet of a workbook into its own file.

Most of this release is less visible. Damaged PDFs used to reach the options form and fail only when you pressed the last button; they are now named the moment they are chosen. HTML table to CSV was rewritten to read tags in one pass, so a large or malformed table is read in milliseconds and matches what a browser makes of it. A CSV over the million-row limit is refused at once rather than after being read to the end, and long pastes on HTML table to CSV and Wrap text run off the page’s thread so the page stays usable.

On phones, the boxes on a dense form no longer cover each other, so a tap types into the field it lands on. Pages with a saved draft hold the room the restore notice needs, so the tool no longer jumps down when it appears.

Bring your own key, without a proxy

Starter and Lifetime accounts can send reviewed document text directly from the browser to OpenAI for four focused jobs.

  • Added summary, table extraction, redaction suggestions, and document description.
  • Kept provider mode off until the visitor agrees to send text off the device.
  • Sent the key and reviewed text directly to OpenAI, never through FileGizmo.
  • Added local PDF and text extraction, friendly provider errors, and token-use reporting.

Local language models were measured first and did not clear the product’s quality and delivery bar. The smaller models failed structured extraction and redaction checks; the first plausible tier was more than one gigabyte compressed and still made factual mistakes. FileGizmo did not ship a weaker result to fill a roadmap box.

Bring your own key is the useful lane that remains. The original file opens in the browser. The visitor sees and can edit the extracted text, explicitly enables provider mode, then sends that text and their own key directly to OpenAI. FileGizmo does not proxy the request, receive the key, meter usage, or add a credit charge. OpenAI’s current API terms and charges apply to the visitor’s project.

FileGizmo is ready for launch

The first public release brings private, local-first tools for PDF, image, CSV, text, and developer work.

  • Shipped 63 focused tools across five practical collections.
  • Standardized worker outputs so downloads, filenames, and file types stay predictable.
  • Fixed CSV smart types while preserving identifiers such as 007.
  • Added keyboard flows, result focus, responsive layouts, and light and dark themes.

FileGizmo began with a simple constraint: ordinary file work should not require handing a file to a stranger’s server. This release turns that constraint into a complete local-first toolkit. The interface stays quiet, files stay on the device, and every shipped workflow is covered by build, conversion, responsive, and accessibility checks.