Reviews
This page is empty. On purpose.
Sırdaş launched recently and we do not yet have user reviews we can verify. We could fill this page with five hundred invented testimonials, like half the internet does. We are not going to, and here is exactly what we will publish here once they exist.
Why there are no reviews yet
Inventing reviews is deceptive advertising. In Colombia it is banned by article 30 of law 1480 of 2011; in the United States, by the Federal Trade Commission rule on consumer reviews and testimonials; in the European Union, by the Omnibus directive.
And there is a more practical reason: this product's only argument is «do not trust us, check it». One fake review destroys that entirely, and it only takes one reader searching for the person and not finding them.
So there is nothing here yet. Below is what you can check today, without asking us for anything.
Our review policy
Written before we had the first one, so it cannot bend to whatever suits us later.
Real, verifiable people
We publish name, role and country. We check the person exists through a work email, LinkedIn or GitHub, and we say which one we used on each review.
Written permission
Nobody appears here without authorising it, and can ask to be removed at any time, without explaining why.
Verbatim
The text is what the person wrote. We fix a typo, not an argument. If we translate, the translation is marked as a translation.
Bad ones too
A negative review is published like a positive one. If the criticism is right, the product gets fixed; the review stays.
No payment, no trade
We do not give money, discounts, licences or prize draws in exchange for a review. If we ever run an ambassador programme, every ambassador review will say so inside the review.
With its context
We say which version they used, on which surface (web, CLI, MCP) and what for. A review without context helps nobody decide.
What you will never see on this page
- Stock photos with invented names.
- A star counter with no reviews behind it, nor aggregate rating markup while there are no ratings.
- «Over 10,000 companies trust us» without being able to name a single one.
- Customer logos from companies that never signed anything.
- Reviews written by us and signed by someone else.
Meanwhile, what you can check today
None of these claims asks for trust. All of them can be checked from your browser in under a minute.
01Your file is not uploaded
Open the inspector (F12), Network tab, Fetch/XHR filter, and process a document. No request carrying your file appears.
02The browser enforces it
In the same Network tab, look at the response headers: the Content-Security-Policy includes connect-src 'self'. The browser blocks the transfer even if the code wanted to send it.
03It works offline
Load the page once, disconnect WiFi and keep working. If the file were uploaded, that would be impossible.
04Zero third-party resources
Not one font, icon or script is loaded from another domain. The OCR model, around 11 MB, is served from this same site.
05No sign-up, no account
No tool asks for an email, a password or a card. There is no registration form anywhere on the site.
06The tools are the ones we claim
Every tool has its own page and opens straight away, with no queue and no daily limit.
And what you cannot check yourself yet
Because a list of proofs without its gaps is marketing again.
- The engine's 272 automated tests are real and run on every change, but the repository is not public yet, so today you have to take our word. It becomes checkable only when we open the core.
- We are not certified against ISO 27001 or SOC 2, and nobody has audited us externally. We apply ISO/IEC 27002 controls and OWASP ASVS requirements, which is a different thing.
- Redaction is rule-based and can miss data. StarCoder's human annotators reached 53% recall on generic identifiers: nobody does it perfectly, and neither do we.
- We have not yet tested the product on fifty real documents from a client. When we do, we will publish what failed.
Use scenarios
These are not testimonials or customer stories: they are the workflows we built the tool for, described so you can see whether yours fits.
- Scenario, not testimonial
Clinic or practice
A scanned medical record goes through local OCR, gets redacted, and the anonymous text goes to the AI to summarise the history. The answer is restored with real names in the same browser. The scan never leaves the machine.
PDF to Markdown for AI - Scenario, not testimonial
Law firm or notary
Two versions of a contract are compared line by line without uploading either. The parties' data becomes placeholders before asking a model about clauses, and the mapping to reverse it stays on the machine.
Compare PDFs - Scenario, not testimonial
Accounting
A scanned batch of fifty invoices is split into documents, each one identified, and the tables come out as CSV for the spreadsheet. No per-page credits.
Split a batch - Scenario, not testimonial
Data or AI team
A folder of PDFs becomes redacted JSONL, deduplicated and split by document, in the format TRL, Axolotl or LlamaIndex expects. In the browser or with the CLI on your own server.
PDF to AI data
Used it? Tell us
If this tool helped you, or if it failed you, write to us. We publish both.
- 1. Write to hola@sirdaspdf.com telling us what you used it for and what happened.
- 2. Tell us which name, role and country should appear, and whether we can name your organisation.
- 3. We reply with the text exactly as it would be published, so you approve it first.