Tool Limitations — AtoZTools

Every tool has limits. This page collects them in one place — one section for each of the site's 42 tools — so you can check whether a tool fits your job before you start. The same limits are printed on each tool's own page.

On-Page SEO and Redirect Checker

This tool fetches a public URL from our server and reports what it finds. How it behaves, exactly:

It identifies itself as AtoZToolsBot/1.0 (+https://atoztools.pro/tool-limitations/) and fetches only http and https URLs on ports 80 and 443.
For your security and ours, it refuses private, loopback and link-local addresses — including IP addresses written in decimal, hex or octal form — and it re-checks the destination address at every redirect hop. Internal network addresses cannot be checked with this tool.
It follows at most 10 redirects, waits at most 10 seconds in total, and reads at most 2 MB of the page. Very large pages are analysed from their first 2 MB, and the report says so when that happens.
It renders nothing: JavaScript is not executed, so content injected by scripts is not visible to the checker. What it reports is what a crawler receives in the raw HTML.
It publishes no score. Checks are reported as evidence with a severity (High, Medium, Low) and a suggested fix, because a single number would hide which problem actually matters.
Results are cached for 15 minutes per URL so repeated checks do not hammer the target site. The cache key is a hash of the URL; your IP address is not stored with results. Checks are limited to 10 per visitor per hour, with a global hourly cap to protect target sites.

Security Header and DNS Checker

Only public addresses on the standard web ports can be fetched; private, loopback and reserved addresses are refused for safety, requests are timed, and large responses are capped.
The cookie check can only grade cookies visible in the final response's headers; a page that sets cookies later through scripts will show none.
The TLS probe reads the certificate presented to a standard verified connection; servers that demand unusual connection behaviour may decline the probe, in which case the report says the certificate could not be read rather than guessing.
DKIM can only be checked under the selectors the tool probes — DNS cannot be enumerated, so "not found" means "not found among the selectors tried", not "this domain has no DKIM". Enter your provider's selector to check it directly.
DNS answers are a moving target: records propagate at different speeds to different resolvers and results are cached briefly, so a record published a minute ago may not appear yet; the report stamps every DNS section with the time it was queried.
Checks are limited to 10 per visitor per hour, plus a site-wide cap shared across all visitors — the same arrangement as the SEO Checker.

JSON Studio

Runs entirely in your browser; nothing is uploaded. Inputs over 5 MB are refused with an explanation, because browser memory — not our policy — is the real ceiling for very large documents.
Duplicate object keys are preserved as written (last value wins in queries and tree view), because silently dropping data would be worse than showing it.
The query tab implements a practical subset of JSONPath ($, child, index, wildcard, recursive descent and simple filters). It is not a full JSONPath implementation.

Regex Tester

It speaks JavaScript (ECMAScript) regex only: no PCRE-only constructs, no recursion, no conditionals, and no atomic groups or possessive quantifiers — restructuring a slow pattern is the available remedy here.
Testing runs in a background worker with a one-second limit. Very old browsers without Web Worker support run the same code on the main thread, where the timeout cannot interrupt a pattern mid-run.
Match collection stops at 1,000 matches per test (the run tells you when the cap is hit), and the capture-groups table prints the first 200 rows.
Invalid-pattern errors show the engine's own message with, where a position can be derived, an approximate character position — labelled approximate, because the JavaScript engine reports no position of its own.

CSV Cleaning Studio

Runs entirely in your browser. Files over 20 MB or 200,000 rows are refused with an explanation. Previews show the first 200 rows; operations always apply to the full file.
Type detection is a heuristic based on the values in each column. A column of numbers stored as text with mixed separators may be detected as text; renaming the type is not forced on you, and the summary reports what was changed.
Mean, median and mode fills are calculated from numeric values only. If a column has no numeric values, those fill options explain themselves instead of inventing a number.
The formula-safe export (on by default) prefixes cells that begin with =, +, - or @ so spreadsheet programs cannot execute them as formulas. Turn it off only if you know the file's destination.

Chart Builder

Runs entirely in your browser. Data over 5,000 cells (rows × columns) is rejected with an explanation, so aggregation and rendering stay fast on ordinary hardware.
Cells that are not numbers are excluded from charted values: the warning names the excluded cells, and a group with no usable numbers is shown as a gap rather than a misleading zero. Empty cells are skipped and counted separately.
Pie and doughnut charts use the first ticked value column only, and only positive values can be slices; negative or zero groups are omitted from the slices.
A scatter point needs both coordinates to be numbers; rows where either cell is empty or non-numeric are skipped and reported in the warning count.
Average aggregation says nothing about variation, and with several series ticked, value sorting orders labels by the first ticked column only.

Image Optimizer

Runs entirely in your browser. At most 20 images per batch and 25 MB per file; processing is sequential in your browser tab, so a large batch is bounded by your device's memory.
Re-encoding can make a file bigger — an already well-compressed WebP or JPEG re-encoded as PNG, or at a higher quality than the original — and the result card shows that as a negative saving rather than hiding it.
Cropping is centre-based: the rectangle is always taken from the middle of the frame. The before/after preview exists so you can catch a cut-off subject before downloading.
AVIF support is detected, not assumed: some browsers can show AVIF but cannot create it through canvas, and a few cannot open it at all; the note under the format selector states what your browser can do. WebP is the fallback.
Compress-to-target has a floor: a noisy, detailed photograph cannot shrink below a certain size at given dimensions — the search lowers quality first, then steps the dimensions down, then stops and reports the smallest honest result.
JPEG has no transparency channel in this pipeline, so transparent pixels are flattened onto white when converting to JPEG. Animated images are treated as a single still frame, and HEIC files cannot be decoded by most browsers — convert those to JPEG first.

PDF Toolkit

Runs entirely in your browser. Limits: 10 files and 50 MB in total, enforced before anything is read. Very long documents take time to thumbnail, because the page manager renders pages one by one.
Password-protected PDFs cannot be processed: there is no password prompt and no workaround — remove the protection in the application that created the file, then add the unprotected copy. Corrupt or non-PDF files are refused as corrupt or invalid.
No compression: merging and extracting copy pages without shrinking them, so a merged file is roughly the sum of its sources.
No OCR: scanned pages are pictures — the toolkit can rotate, move and extract them, but cannot read their text or make them searchable. No form filling, signing or editing of page text either; existing content is carried across as it is, and editing a signed PDF in any tool can invalidate its signatures.
Rotation is visual and permanent in the output: it sets the page rotation in the saved file and does not rotate scanned content inside the page image. Flattened or unusual structures (for example portfolio PDFs) are outside launch scope and may be refused as unreadable.

Structured Data Generator

It covers nine types — Article, FAQPage, HowTo, BreadcrumbList, Organization, Product, LocalBusiness, WebSite and WebApplication — with the properties most pages need. Schema.org's hundreds of other types, specialist markup (events, courses, job postings, recipes) and advanced patterns such as a multi-type @graph are outside its scope.
The validator checks the rules encoded in the tool — required properties, formats, ranges and recommendations drawn from Google's documentation at build time. It is not Google's Rich Results Test and cannot see your live page, your template's other markup, or policy changes published after the page was written: a clean result means "ready to test on the real page", never approval.
Markup must match the visible page: the tool cannot stop you describing things the page does not contain, giving relative instead of absolute URLs, or choosing the wrong type — keep the two in step and update the markup whenever the visible content changes.

Robots.txt and Sitemap Toolkit

Robots.txt is not a security mechanism: it is a public, voluntary request that well-behaved crawlers choose to honour. It cannot stop a browser, script or hostile bot from fetching a blocked path, blocked URLs can still appear in search results as bare links, and listing a sensitive path advertises that the path exists. Protect private content with logins and access controls.
Googlebot ignores the Crawl-delay line entirely (use Search Console's crawl-rate settings for Google); Bing and Yandex do honour it. Blocking Google-Extended affects AI training use only — Google Search follows the separate Googlebot token.
Paths are case-sensitive, and a trailing $ anchors a pattern so strictly that a query string stops it matching; the simulator demonstrates both behaviours.
The URL tester implements the RFC 9309 core — group selection, longest match, wildcards and anchors — against one file at a time. Individual crawlers add their own extensions and size limits, and a crawler holding a stale cached copy of your file will not see today's edit.
The sitemap generator validates form, not truth: it cannot know whether a listed page exists, whether a date is accurate, or whether a page deserves indexing.

UTM Campaign Builder

The builder constructs and checks URLs; it cannot verify that the destination page exists, that your analytics property is configured to read the parameters, or that a link shortener in the middle preserves them — test the final shared link end to end, and confirm the parameters survive any redirect.
Presets are stored only in this browser's local storage, so they do not follow you between devices or teammates; for a shared convention, export a bulk CSV and keep it as the team's reference copy.
The bulk builder holds 50 rows per batch by design; larger campaigns should be split by channel. Rows past the cap in an import are reported, not silently dropped.
QR codes encode the built URL exactly, but a very long URL makes a dense code that older phone cameras struggle with — the tool warns past roughly a thousand characters.
Existing utm_ parameters on a pasted destination are removed and replaced by the builder's values, and the builder tells you when it does so, because two values for one parameter are ambiguous.

Pricing and Break-Even Simulator

Some targets are impossible: a target margin that leaves nothing for costs after percentage fees cannot be solved, and a product whose contribution per sale is zero or less never breaks even — the tool reports both situations instead of dividing by a loss.
Margin and markup are not interchangeable: a 25% margin needs a 33.33% markup on the same product. Fees are modelled on the discounted, tax-inclusive amount the customer pays; if your platform charges on a different basis, adjust the percentages so the fee rows equal a real payout.
The currency selector changes the display symbol only — no live exchange rates are used, and mixing inputs from two currencies in one simulation produces meaningless results.
Returns, refunds and chargebacks are not a separate input: fold their average cost into the unit or ad cost per sale. Tier discounts stack on the base discount; if your store replaces the base discount at tier quantities, set the base discount to 0 and let the tiers carry the reduction.
It is a planning model, not accounting: costs are treated as constant per unit, ignoring stock financing, payment timing, tax deductibility of fees and volume pricing from suppliers.

Content Analyzer

Scores are estimates from published formulas, not measurements of quality: the formulas count syllables and sentence lengths and cannot detect a wrong fact, a broken argument or a dull paragraph. Necessary technical vocabulary lowers a score without lowering the writing.
This tool does not detect AI-written text. Readability statistics look the same whoever or whatever produced the words, and no threshold separates human from machine writing reliably.
English only at launch: the sentence splitter, syllable heuristic, stopword list and participle patterns are all English-specific, so numbers produced for another language will not mean what the labels say.
The syllable counter is a heuristic (vowel groups with a silent-"e" adjustment) and the passive-voice indicator is a pattern match that both misses and over-flags; comparing two drafts with the same counter is fair, treating one score as absolute is not.
Keyword-density thresholds (3% single word, 2% multi-word phrase) are rules of thumb for prompting a second look, not limits published by any search engine.
Pasted and uploaded text is processed entirely in your browser and never transmitted. The one exception is the optional URL import: our server fetches the public address you submit through the SSRF-safe client described in the Privacy Policy, extracts the page's readable text and caches it for 15 minutes under a SHA-256 key; imports are limited to 10 per visitor per hour with a shared site-wide cap. Pages built entirely by scripts, or behind logins, may import incompletely or not at all, and the heading check needs HTML or Markdown input.

Text and Document Comparison

Each input is capped at 1,000,000 characters; an input over the cap is refused with a message naming its size, and nothing is compared — trim the text or split it at a natural boundary and compare in parts.
Files are read as text in the browser: .txt, .md, .html, .csv and .json work; Word .docx files are not extracted (save as .txt first), and binary files fail to read with an error saying so.
"Ignore whitespace" is implemented by pre-normalising the text before the diff runs (in line mode, each line is trimmed and its internal whitespace collapsed for comparison while the display shows your original lines) — it is documented this way because the underlying diff library's own option does not do what its name promises in character mode.
"Ignore case" folds case one Unicode code point at a time, so it cannot reproduce context-sensitive whole-word rules (for example the Greek capital sigma at a word's end may still be reported as a difference).
Line endings are normalised (CRLF and CR become LF) before comparing. Very large inputs near the cap, in character mode especially, can take a few seconds, and there is no progress bar.

QR Code Generator

Runs entirely in your browser; the code encodes exactly the text you type — no short link, redirect or tracking is inserted, so the destination can never be changed later without reprinting the code.
Capacity is the QR standard's: at most 2,953 bytes at error-correction level L (1,273 at H) in a version-40 code. Codes near the ceiling are extremely dense and many cameras will fail to read them; keep the encoded text short.
Dark modules on a light background is the only safe arrangement; inverted colours or a code set over a photograph may be technically valid and practically unscannable. Keep the quiet zone (blank margin) intact.

Password Generator

Length is capped at 128 characters and floored at 4; the destination site's own maximum governs what you can actually use.
The entropy figure is an estimate built on uniform random selection from the chosen sets. It describes this generator's output only — it says nothing about how a site stores passwords or whether it has been breached. Turn on a second factor wherever it is offered.
History is the last 20 generated passwords, held in this tab's memory only and gone when the tab closes. It is a safety net against losing a password you just made, not a password manager — store passwords in one of those.

Base64 Studio

Text mode accepts up to 1,000,000 characters and File mode up to 10 MB, because the input, the encoded form and the output must all fit in the tab's memory at once. Larger files belong in a desktop or command-line tool.
Text mode decodes as UTF-8 only; binary data belongs in File mode. Base64 carries no filename, extension or type — a decoded download may need renaming to the extension the data actually is.
The decoder validates characters, length and padding, but it cannot know what the bytes were supposed to be: valid Base64 encoding the wrong file decodes perfectly, into the wrong file. Base64 is an encoding, not encryption, and verifies nothing about identity or safety.

Word Counter

Counting rules differ between tools (hyphenated words, numbers, punctuation), so counts here will not always match a word processor or a submission system. Where a system gives its own count, that count wins for that submission.
Sentence counting is approximate because abbreviations and initials imitate sentence endings. Reading and speaking times are fixed-rate averages (200 and 130 words per minute) and say nothing about a particular audience.
Keyword density is descriptive, not a target: no percentage on the page makes a text rank or read well. Density works on exact tokens — it does not stem (tool vs tools differ) or merge synonyms — and text inside images or tracked changes is invisible until extracted as plain text.

Merge PDF

At most 10 documents and 50 MB combined per pass; larger sets are bound in stages, feeding one merged result back in with the next group.
Merging copies pages unaltered: the output weighs roughly the sum of its inputs. Compress afterwards if size matters. Form fields and annotations are transplanted, not reworked, and folding a digitally signed document into a larger file can break its signature's validity.
Password-protected files are refused by name — the tool never asks for a password. Portfolio PDFs (bundles wearing one cover) may be refused as unreadable rather than dismantled.

Split PDF

One source per session, up to 50 MB. The pieces together weigh about what the source weighed — splitting helps when a limit applies per file, not when the total must fall.
The cutter knows page positions only: chapter headings, bookmarks and scan boundaries are invisible to it, so breaks land exactly where your ranges dictate. Multiple pieces arrive in a single ZIP archive.
Password-protected documents cannot be opened at all; split an unlocked copy made in the application that produced the file.

Extract Pages

One document at a time, up to 50 MB. The unit of extraction is a whole page — half a page, or text lifted off a page on its own, cannot be harvested.
Output order follows your typing exactly (a selection starting "10, 2" opens on page 10), repeated pages are deliberately collapsed, and a number beyond the document's end halts the job with the offending number named.
No compression happens along the way, and a page that began as a photograph stays a photograph — legible, but with no text to search or copy.

Remove Pages

One document per run, up to 50 MB, and at least one page must survive — a list accounting for the whole document is refused rather than producing an empty file.
Selection is by file position counted from the first sheet, never by the page number printed on the page itself, and never by content: the tool will not hunt down blank pages for you. Bookmarks and links on surviving pages that pointed at removed pages are carried across as they were and may lead nowhere.
Password-protected documents cannot be opened; trim an unguarded copy instead.

Organize PDF

One document per session, up to 50 MB, and at least one page must remain or there is nothing to save. Whole pages are the unit of work — part of a page cannot be moved, and two pages cannot be squeezed onto one sheet.
Rendering a thumbnail for every page takes time on long documents and modest hardware; a stored crookedness inside a page's picture cannot be levelled by the Rotate button, which adjusts stored rotation only.
Password-protected documents are refused by name.

Rotate PDF

Quarter turns only (90, 180, 270). A page scanned a few degrees crooked needs deskewing in an image editor — stored rotation cannot level it — and a picture that is sideways inside the page content cannot be repaired by changing how the page is presented.
One document per run, up to 50 MB. A minority of producers record orientation inside the page content instead of the standard rotation value; pages built that way may appear unaffected.
Password-protected documents cannot be opened and are refused by name.

Compress PDF

"Smallest size" destroys the text layer. Pages become JPEG images: text cannot be selected, searched, copied or read aloud; links, bookmarks and form fields stop working. This is stated in the mode picker, in the result line and here because it is the mode's most important fact.
"Keep text" cannot shrink image data — a photo-heavy PDF may come back the same size or slightly larger, which is the honest result of a lossless pass. Neither mode performs OCR. High DPI and quality settings can make the output larger than the input.
One document at a time, up to 50 MB; long documents rasterise page by page in your tab and can be slow. Password-protected PDFs cannot be opened in either mode.

Repair PDF

Repair rebuilds structure only. Content that is itself missing or corrupt inside the file — a truncated image stream, a half-written page — is carried into the rebuilt document as the same damage; nothing is invented to fill gaps.
A file that is not really a PDF is refused as corrupt, and an encrypted file is reported as password-protected and refused — encryption is not damage. Incremental-update history in the original is not preserved; the output is a clean single revision.
One file at a time, up to 50 MB. A repaired file is not a verified file: compare the recovered page count with what you expect and skim the result before relying on it.

JPG to PDF

At most 10 images and 50 MB in total per batch. Page order is exactly the on-screen list order — there is no drag-to-reorder.
WebP images are re-encoded as PNG inside the PDF, which can make the result larger than a JPEG-based one. There is no text recognition: a photographed document stays a picture, unsearchable and uncopyable.
A4 mode never enlarges a small image; in Fit-image mode, very large images are scaled down by a 1,440-point cap, which reduces print resolution at very large sizes.

PDF to JPG

One PDF at a time, up to 50 MB, and output is JPEG only — there is no PNG or WebP export, so line art and screenshots carry JPEG artefacts a lossless format would avoid.
The JPG is a picture of the page: text is not selectable, form fields flatten to their appearance, and links do not survive. Very long documents at 300 DPI are slow and memory-hungry because every page is rasterised at full size in your tab.
Password-protected PDFs are reported by name and the conversion stops.

Scan to PDF

There is no camera view inside the page and no OCR: cleaned pages are still pictures whose text cannot be searched, selected or copied. Rotation is quarter turns only — there is no deskewing — so photograph pages as square-on as you can.
Enhancement cannot rescue everything: deep shadows, glare and very faint pencil can survive cleaning or be pushed too far by Document mode; Greyscale is the safer choice for delicate originals.
At most 10 images and 50 MB per batch; pages follow list order, and each page keeps its own image's size, so mixed-resolution batches produce mixed page sizes.

HTML to PDF

The output is a raster picture of your HTML: text is not selectable or searchable, links are not clickable, and screen readers cannot read the pages. Scripts never run — by design, for safety — so content that appears only after JavaScript executes will be missing.
Page breaks are blind slices at A4 height; a heading, table row or image straddling a boundary is cut in two. Print stylesheets are ignored and some modern CSS renders differently from a full browser tab; external images that fail to load appear as fallback text or empty space.
Very long snippets produce many pages and large files, because every page is a full-resolution image.

PDF to Markdown

The conversion is heuristics, not understanding: a large line is assumed to be a heading and a wide gap a paragraph break, so display type, pull quotes and captions can be misclassified. There is no table structure — cells come out as plain lines in reading order — and multi-column documents may be read across the columns.
Scanned PDFs have no text layer and produce empty output; this tool does not perform OCR. Images and charts are dropped; only real text items are converted. The output is a first draft that expects a cleanup pass.
Password-protected PDFs cannot be opened and are reported by name.

OCR PDF

The OCR engine is a real download — the Tesseract engine, worker, WASM core and English language data total 6.05 MB — loaded only on this tool's page and cached by your browser afterwards. English only; handwriting in any language is not recognised.
Recognition quality is bounded by the scan: skewed, blurred, low-contrast or very small type will be misread in places. OCR produces an approximation, not a transcript — the page image is preserved underneath the text layer, and important documents should be checked against it.
Recognised pages are stored as page images with a text layer, so file size reflects the render DPI; layout is approximated by word positions, and complex tables and columns may extract imperfectly. One file at a time, up to 50 MB; long documents are slow because every page is rendered and recognised on your device.

Edit PDF

This is an overlay tool: existing text cannot be edited, moved or deleted, and anything you cover is still present underneath the covering shape — in the file and in its text layer. Never use it to hide personal data for sharing; that is what Redact PDF is for.
Text stamps are drawn in Helvetica only, on a single line: no wrapping, no rich text, no font upload. Lines are straight; there are no freehand drawings, arrows or curves in this version.
One PDF at a time, up to 50 MB. Password-protected PDFs cannot be opened at all, and adding content to a signed PDF can invalidate its signatures in readers that verify them.

Add Page Numbers

Numbers are drawn in one standard typeface (Helvetica) in dark grey — no font choice, colour picker or bold option. One numbering scheme applies per run: documents needing Roman numerals for front matter and Arabic for the body need the parts numbered separately.
The stamp sits on top of the page; if a page already has a footer where the number lands, the two will overlap. Existing page numbers in the document are neither detected nor removed.
One document at a time, up to 50 MB. Locked documents cannot be numbered at all.

Add Watermark

Text only — no image or logo watermark and no font choice. The stamp is drawn on top of the content as a translucent layer: it marks a document's status and discourages casual copying, but it is not a security feature, because anyone with a PDF editor can remove or cover it.
One wording and one style per run; tiling density and angle derive from the text size and the single rotation setting. Light grey at low opacity can vanish in print or photocopying.
One document per run, up to 50 MB. Encrypted documents must be exported without their guard before stamping.

Crop PDF

A crop is not a redaction. Content outside the CropBox remains in the file, hidden from normal view but fully present and revealable. Never crop to remove names or figures that must actually be gone — and because nothing is deleted, file size barely changes.
Margins are uniform for every selected page (the same percentages in percent mode); there is no automatic white-space detection, so documents whose pages are offset differently need separate runs. A crop must leave at least 36 points on each axis.
One document at a time, up to 50 MB. Password-protected PDFs cannot be opened.

Sign PDF

This is not a digital signature in the cryptographic sense: no certificate, no identity verification, no tamper evidence, and no entry in a reader's signature panel. Do not use it where a certified digital signature is required.
One signature, on one page, per run — signing several pages means feeding each output back in as the next input. Typed signatures use one typeface (Times-Italic); a JPEG signature has a solid background that will cover whatever is beneath it, where a transparent PNG does not.
Flattening is permanent in the output file, and very wide signatures are refused rather than silently shrunk (an A4 page fits at most 523 points at the standard margins). Password-protected PDFs cannot be opened or signed here.

Fill PDF Forms

Only real AcroForm fields can be filled. A scanned or photographed form contains no fields and produces the no-fields message; XFA forms (the older Adobe LiveCycle format) are likewise out of scope. Push buttons and signature fields are listed by type but cannot be filled with a typed value here.
Calculation and formatting scripts inside a form are not run, so totals a reader would normally compute must be checked yourself before flattening. Flattening is one-way: a flattened copy cannot be un-flattened, so keep the unflattened version.
One file at a time, up to 50 MB. Protected forms stay closed here.

Compare PDF

Text only. Differences in fonts, layout, images, colours and anything visual are invisible to the comparison — two visually different documents can score 100%, and the score says nothing about appearance.
Pages are matched by position, so a page inserted near the front of one document can make every later page report as changed. Scanned pages have no text to compare: OCR both sides first, because a scan compared against a text PDF reports every word as added or removed.
Similarity is a word-count measure, not a judgement — changing "shall" to "shall not" scores in the high nineties and reverses the meaning. Two files at a time, each up to 50 MB; password-protected PDFs cannot be read here.

Protect PDF

Permissions are requests, not walls: the open password is enforced by encryption, but the permission flags are instructions compliant readers honour, and a determined recipient can often bypass print or copy restrictions elsewhere. There is no password recovery — nothing about your passwords leaves your device and no copy of the file is kept, so store the password safely and test the protected copy before sharing it.
AES-256 needs a reasonably current reader; choose AES-128 only for genuinely old recipient software. A file that already carries encryption — including quietly encrypted bank or scanner files — is detected and refused: unlock first, then re-protect.
One file at a time, up to 50 MB. Protection changes nothing about content, though modifying a signed PDF in any tool can affect its signatures.

Unlock PDF

You must know a password that opens the file. This tool removes protection; it does not recover, guess or bypass forgotten passwords, and exactly one password is attempted — the one you type, or the empty password if the field is blank — so a typo is reported as a wrong password.
The output is rebuilt page by page, which guarantees no leftover encryption data; page content, order, size and rotation are copied exactly, but catalogue-level extras are not guaranteed to survive — form fields may stop being editable and the document outline (bookmarks) can be lost.
One file at a time, up to 50 MB. Damaged files cannot be unlocked, and unlocking a signed document can affect its signatures.

Redact PDF

Marked pages are rebuilt as images with the original content discarded — that is what makes the redaction real — and the price is that a rebuilt page's text, including text you did not cover, can no longer be selected, searched or copied. Rebuilt pages are JPEG images, so fine text is slightly softer than the original vector text, more so at 150 DPI than at 200.
Redaction here covers rectangular regions on pages. It does not hunt for patterns, and attachments, embedded files, layers, comments, form-field values and document metadata are outside its scope — check a sensitive file's attachments and properties separately before sharing. The black is only where you drew it: a region that misses the last digit of a number leaves that digit visible.
Redaction is permanent in the output; keep the original somewhere safe and private. One PDF per run, up to 50 MB; password-protected PDFs cannot be opened.

General

No tool on this site requires an account. Browser tools do not store your input on our servers. The two server-side checkers fetch public URLs only, under the rules above, and the Content Analyzer's optional URL import fetches a public page the same way, as described in its section and in the Privacy Policy.
All 42 tools are live — the 14 launch tools plus the 28 added since: 4 flagship tools and the 24-tool PDF suite. Category pages and the homepage link only to tools that exist.