Comparison
PDF tools that do not upload your file
When a PDF site does its work on a server, the shape is always the same: you choose a file, it travels to a computer you have never seen, something happens there, and a result comes back. That is not a criticism of any particular site — it is what running software on a server means. A program cannot operate on bytes it has not received.
OpenTools is built the other way round. The program is sent to your file instead: the code arrives as part of the page, and the work happens inside the tab that is already open. Which model a tool uses is a fact about its architecture, not a promise in its privacy policy — and it is the only difference on this page worth caring about.
What is enforced, and by whom
“We don’t keep your files” is a policy. A policy is a description of an intention. These are controls, and none of them depends on our good behaviour:
- The browser refuses the connection. PDF tool pages are served
connect-src 'none', which switches off fetch, XHR, WebSocket, EventSource and sendBeacon together. Your browser enforces it, so it holds even against a bug in our own code. - The build refuses the code. The test suite scans every engine, worker, component and route source for network primitives and literal remote addresses and fails when one appears, so no tool can quietly acquire an upload path between releases.
- Each release is attacked on purpose. An executable protocol tries five exfiltration routes from the page’s own context, then runs a real file through a real tool and measures what left. Most recent run: six of six checks passed in Chromium and WebKit on 21 September 2026, with 0 bytes recorded to any off-origin host.
The full protocol, the result and how to reproduce it in thirty seconds — and the threat model if you are deciding this on behalf of other people.
The tools, and what each one will not do
Every one of these is live now, free, and needs no account. The right-hand column is the part a comparison table usually hides.
| Tool | What it does | Where it stops |
|---|---|---|
| Merge PDF | Combine PDF files in the order you choose. | Capped at 20 files and 150 MB per run while the tool is canary — a safety limit of ours, not a browser one. |
| Extract PDF pages | Choose pages or ranges and save them as a new PDF. | Splits by page selection, not by bookmark or by detected chapter. |
| Page tools | Reorder, remove, rotate, number, watermark and label PDF pages. | Edits the page graph. It is not a content editor: body text stays as it is. |
| Compress PDF | Rewrite a PDF more compactly and re-encode the photos inside it. | Only images it can rewrite without changing how the page renders are touched. If the result is not smaller, the original bytes come back and the receipt says so. |
| PDF to Word | Extract the text into an editable .docx: reading order, paragraphs, page breaks, headings. | Layout, columns, tables-as-tables, images and fonts are not reproduced. A scanned PDF has no text layer and is refused by name rather than returned empty. |
| PDF to Excel | Convert bank statements and PDF tables into .xlsx and CSV, with column detection. | Works on tables the PDF describes as text. A photograph of a table is a different job. |
| OCR PDF | Add an invisible searchable text layer to a scanned PDF, keeping its visible pages. | English only. Recognition quality depends on the scan, and it is not proofread. |
| Redact & black out PDF | Black out text or rectangles, rasterise the redacted pages, and purge metadata and annotations. | A password-protected or encrypted PDF is refused by name rather than partially processed; passwords are not bypassed. |
| Sign and fill PDF | Complete a PDF form and draw a signature onto the page. | A visible signature drawn on the page. Not a cryptographic signature, and not identity verification. |
| PDF metadata viewer and remover | See the author, authoring program, dates and XMP packet inside a PDF, then remove them. | Document-level metadata. It does not rewrite information printed inside the page content. |
| Bates numbering | Stamp sequential legal reference numbers across a set of PDFs. | A password-protected or encrypted document is refused by name rather than stamped. |
| Compare PDFs | Whole-document comparison across two drafts: insertions, deletions, moved clauses, formatting changes. | Compares the documents’ text streams. It is not a visual pixel diff of the rendered pages. |
| Images to PDF | Arrange JPEG and PNG images into one PDF. | Capped at 40 images per run. |
The exceptions to the policy above are deliberate, and none of them is a PDF tool. /image/background-remover and /image/editor are served connect-src 'self' instead, because the background remover has to download its ONNX model and WebAssembly runtime before it can run — from this same site, never from a third party. Those requests travel towards your browser, like the scripts and the stylesheet. They do not carry your image anywhere. Every PDF route on this page is connect-src 'none'.
What a server-side PDF service does better
A comparison that finds nothing good on the other side is an advertisement. These are the cases where uploading is the right answer, and they are consequences of the same architecture that makes the rest of this page true.
- Files bigger than your device can hold
- The work happens in your tab, so it is bounded by your machine’s memory. A service running on hardware larger than your laptop can open documents this cannot, and it will not be close on an old phone.
- The same result for everyone
- A server gives every customer the same CPU, the same library versions and the same fonts. Here, the browser and the device are part of the pipeline, so two people can get results that differ slightly — and a rare browser can fail outright.
- Recognition and conversion at a quality a tab cannot reach
- Optical character recognition, layout reconstruction and language coverage scale with model size. A server can run a model of any size. Anything running here has to be downloaded into your browser first, which puts a hard ceiling on it: our OCR is English and our PDF-to-Word keeps the text, not the layout.
- Exact-layout conversion
- Turning a PDF back into a Word document that preserves columns, tables, images and fonts is what commercial conversion engines are for. It is not honestly achievable in a browser tab at a quality worth charging for, so we do not attempt it.
- Everything that needs an account to exist
- Cloud storage connectors, shared team folders, saved templates, audit trails, retention policies, e-signature workflows with identity verification. None of these can work without a server holding state, and we hold none.
- Someone to be accountable to you
- A paid service has a company behind it, a support desk, a contract and a service level. This is an open-source project. If a tool is wrong at two in the morning, there is no one on call.
The short version: if the document is routine and you would rather it stayed on your machine, use this. If you need exact layout reconstruction, a 2 GB file, an audit trail, or a contract with somebody, you need a server — and you should use one.
What this page does not claim
- Nothing about any other company’s product, pricing, limits or handling of your data. We have no verified source for those facts, so they are not here. Where a comparison was unavoidable it is drawn against the server-side model in general, which is a matter of definition.
- No speed claim. Whether a job finishes faster here depends on your device against their hardware, and we have not measured theirs.
- The measured figures belong to the release dated above and to the tools and two engines the protocol covered. A later release is covered only once the protocol has been re-run against it.
- Egress evidence reports where bytes went. It is not a security guarantee and can never establish that no flaw remains.
Keep reading
- Browser-based vs cloud file tools — the architecture in full, and how to check which one any tool you use actually is.
- Open-source PDF tools you can read and run yourself — the licence, the repository and the container.
- Every PDF tool on the site, with its guide.