🔍 Read the full analysis: What I Learned Making My Vibe-Coded PDF Editor Reliable on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer reports that six days of testing and revisions turned Docestra from a quickly built PDF editor into a more thoroughly tested personal tool. A 930-file test exposed a save bug, while encryption and redaction tests led to design changes; the author says time and cost savings have not been measured.
Thorsten Meyer says testing and review turned Docestra, his AI-built PDF editor, from a one-day working prototype into a more extensively tested personal tool over six days. In an account published November 13, he describes a save defect found across 930 real PDFs, limits on editing encrypted files, and a redaction redesign; he says the tool remains private and its time and cost savings have not been measured.
Meyer says he gave an AI coding agent a one-sentence brief on November 6, 2026, to build a PDF editor he could use. The first version was committed at 21:47 Central European Time that evening. It could open and search documents, add annotations and signatures, fill forms, reorder pages, and save PDFs. The local TypeScript web app runs in a browser on his Mac, and he says files stay on the machine.
The first broad test covered 930 PDFs and found that saving some files could produce a broken cross-reference table, causing markups to disappear. Meyer says he changed the save process to avoid compressed object streams. In a separate set of 69 encrypted PDFs, 52 did not permit annotations under restrictions set by their authors. Because the PDF library could not re-encrypt edited files, Docestra was changed to request confirmation before unlocking and to save an unprotected copy without overwriting the original.
Meyer also distinguishes the first version’s “white-out” feature from actual redaction: it covered text visually but left the underlying content in the PDF. After an independent AI review identified 14 issues and then 13 in successive reviews of an initial redaction design, Meyer says he replaced in-place edits with a method that rasterizes pages containing redactions and adds a searchable text layer. He reports that a test on 917 PDFs found zero leaks. These are results from his own testing, not an independently published evaluation.
Docestra · Nov 6–13, 2026
From one prompt to a tool I can rely on
What the first version gave me, what real files broke, and what it took to fix it, from my project log.
What the tool rests on (schematic)
Built bottom up. The top layer is a plan, not a finished feature.
What happened, by date
- Nov 6One-sentence brief; first working version committed the same evening
- Nov 6Test over 930 real PDFs finds a save bug that dropped markups
- Nov 752 of 69 encrypted PDFs forbid annotations: unlock only into a new copy
- Nov 8Real redaction: review finds 14, then 13 problems; simpler design, 0 leaks in 917 PDFs
- Nov 12Version 1.5.1: 745 unit tests, 32 browser test suites
- Nov 13Version 2.0 starts: one shared PDF engine for our AI agents (in development)
Source: Docestra project records (development log, plans for versions 1 and 2.0, commit history), Nov 6–13, 2026. The layer stack is a schematic, not a measurement.
Graphic: Powered by Thorsten Meyer AI
Testing Exposed Prototype Limits
Meyer’s account illustrates why a software feature appearing to work on a few sample files is not the same as being dependable across varied documents. A save defect can silently discard annotations, while a visual cover-up that leaves text in place can create a serious disclosure risk. Those issues matter to anyone using an AI-generated tool for contracts, forms, or other documents where edits and sensitive information must be handled correctly.
The account also shows the work involved in making an AI-built application safer to use: testing representative files, documenting defects, reviewing changes, and deciding when a design needs replacing rather than another patch. Meyer says the project’s automated unit tests rose from 17 in the first version to 745 in version 1.5.1. That count indicates a larger test suite, but does not by itself establish the software’s overall reliability or security.
There is no reported productivity result yet. Meyer explicitly says he has not measured time or cost savings, so the account supports conclusions about his development process, not a quantified claim that vibe coding is cheaper or faster overall.
Products Worth Considering
Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black
Jackery Explorer 1000 v2 Portable Power Station
As an affiliate, we earn on qualifying purchases.
From One-Day Build to Version 1.5.1
Meyer defines “vibe coding” as describing desired software in plain language and having an AI model produce the code. He says AI coding agents using Claude Code and Claude Opus wrote Docestra, while OpenAI’s Codex reviewed the code from the start. The first version already included viewing, annotations, signatures, form filling, page operations, combining and extracting pages, printing, undo, and saving.
By November 12, the project had reached version 1.5.1. Meyer says features and fixes came from his own use as well as testing. He also describes checking software licences before adding libraries, including declining to ship fonts from a viewer library because they were under GPL-2.0 and did not meet his project’s rules. These details frame reliability as including legal and packaging choices, not only whether the interface works.
The account is a first-person report from the developer, rather than a third-party audit or a public product release. Docestra is not available to download, according to Meyer.
“Vibe coding gets you a working tool quickly. Turning it into a tool you can rely on is the real work.”
— Thorsten Meyer, describing the project’s main lesson
Limits of the Reported Tests
The results are reported by Docestra’s developer, and the source account does not provide a public test dataset, test procedures, or independent validation of the reported zero leaks across 917 PDFs. It is also unclear how closely those files represent documents users would encounter in other workflows, or how the editor performs beyond Meyer’s own Mac and browser setup.
Docestra is private, so readers cannot independently try the current version. Meyer does not report measured time or cost savings, a formal security audit, or a release plan. His account describes the application as a personal tool and the testing as project work, not as a guarantee that every PDF or redaction will be handled correctly.
Version 2.0 Remains in Development
Meyer says work on Docestra 2.0 began November 13, 2026, the day the account was published, and that it is not finished. He does not give a release date or say whether the application will become available to others. The next stated milestone is therefore continued development, with no public launch confirmed.
For now, the account leaves readers with a record of changes through version 1.5.1 and the limits of the evidence presented: a growing test suite and reported results on real files, but no public product, independent validation, or measured savings.
Key Questions
What is Docestra?
Docestra is Thorsten Meyer’s AI-built PDF editor, described as a local TypeScript web app that runs in a browser on his Mac. He says files stay on the machine and that the tool is private, not available to download.
What defect did testing find?
Meyer says a test run over 930 real PDFs found that saving some files could break a cross-reference table, making markups disappear from the saved document. He says he changed the save method to avoid compressed object streams.
Did the original white-out tool remove text from a PDF?
No. Meyer says the initial tool covered text visually while leaving the underlying content in the PDF. He later built a separate redaction approach and reports zero leaks in a test of 917 PDFs, a result that has not been independently validated in the account.
Is Docestra available to the public?
No public download is confirmed. Meyer describes Docestra as a private tool, and says version 2.0 is still in development.
Did Meyer measure savings from using AI to build it?
No. Meyer says he has not measured time or cost savings. His account describes the development process and testing, but does not establish a quantified productivity benefit.
Source: ThorstenMeyerAI.com
