Your document has code in it.
This colours it like your editor.

What it actually does

Most highlighters expect code you have already separated from everything else, and expect you to say which language it is. A document does not arrive that way. Code sits in the middle of paragraphs, headings, captions and tables, and usually nothing labels it — especially after a trip through a PDF export or somebody else's template.

So the first job is not colouring. It is working out which lines are code at all, which lines are terminal output, which are equations, and which are ordinary prose that happens to contain a bracket. Highlit Docs does that with a deterministic detector built for exactly this problem, then colours what it found and puts it back where it came from.

You are not stuck with its answer. Every line in the result carries a confidence figure, a control jumps you straight to the least certain ones, and selecting any passage lets you mark it as code, terminal output or an equation — or unmark it entirely. Your corrections flow through to every download.

Why an algorithm and not a chatbot

A language model can do this job, and it does it worse. It has a context window, so a long document has to be fed to it in pieces and the pieces disagree. It paraphrases: you asked for highlighting and got a rewrite. It invents lines that were never in your file. It costs money per page, and that price only goes one way. And every run gives a slightly different answer, so a document you correct once does not stay corrected.

A deterministic detector has none of those problems. It reads the whole file at once however long it is, it never changes a character of your prose, it produces the same result every time, and it is free to run because running it costs almost nothing. There is a fuller argument, point by point, on why not just use AI.

If you do want a second opinion, one button asks for it — and only for the handful of lines the detector itself flagged as doubtful. Nothing leaves until you press it.

Who this is for

• Students and lecturers handling assignment briefs, lab handouts and lecture notes that arrive as PDFs with unreadable grey listings.
• Technical writers reviewing a specification whose code samples lost their formatting somewhere between Word, Confluence and a PDF export.
• Engineers reading a design document, RFC or postmortem written by somebody who pasted code straight into Word.
• Anyone reviewing a CV or portfolio PDF that includes code samples and is painful to read as-is.

What you get back

On screen, the document is rebuilt as a reading view inside an editor window: prose stays prose, and every detected block becomes a syntax-highlighted pane with line numbers and a language tag, in whichever of the eleven IDE themes you pick. Terminal output gets its own darker surface so you can tell a program from its output at a glance. Equations are typeset on white, the way they are printed.

You can download the result as PDF, Word or HTML, whichever format you started from, or copy every code block at once.

Asking for the format you came in with edits your original file. Give it a .docx and take a .docx back and only the code is touched — every other paragraph keeps its exact font, size, spacing, colour and style. That guarantee is structural rather than best-effort: only the parts of the file belonging to code are rewritten, so the rest cannot drift even slightly. The same holds for a .pdf in and a .pdf out.

Asking for the other format rebuilds the document — but it carries the formatting over. Page size and margins, point size per run, bold and italic, serif versus sans versus monospace, text colour, indentation, alignment and line spacing all come across, so a text-and-code document lands very close to the original. Images and tables do not, and the exact typeface cannot: a PDF can only use fonts embedded in it, so Calibri arrives as Helvetica. The tool tells you which of the two you are about to get. What survives a conversion.

Accuracy, honestly

A Word file that used a monospaced font or a code style for its listings is close to a solved case — the formatting says "this is code" and the tool believes it. Plain text with no formatting at all is harder, and short one-line snippets are the weakest case by design: a lone ambiguous line is treated as prose rather than risking a false positive in the middle of a sentence.

Things it will get wrong: heavily tabulated data can look like code, and a PDF that stores its text as scanned images has no text to read at all. That is why the confidence figures and the low-confidence jump control exist — the doubtful lines are shown to you as doubtful instead of being presented as certain. The limitations page lists every failure mode found so far, with examples.

Common questions

What happens to my file?

It is sent to our server, held in memory, and deleted. Reading a .docx or .pdf and writing a highlighted one back out happens there, not in your browser. The file travels over HTTPS, is never written to disk, is never sent anywhere else, and is discarded within 20 minutes of your last action — or the moment you close the tab, whichever comes first. It is never used to train anything.

One button goes further afield. Refine output with AI sends the lines the detector was unsure about — not the file, and not the whole document — to Google's Gemini API for a second opinion. Nothing is sent until you press it. If the document is confidential — client work, unreleased source, anything under an NDA — see the privacy page for exactly what goes where.

Which languages are supported?

Python, C, C++, Java, JavaScript and TypeScript are identified by name. Code that clearly is code but matches none of them is still highlighted and labelled "Unknown" rather than guessed at. See supported languages.

Does it work on scanned PDFs?

No. If a PDF is a photograph of a page, there is no text layer to extract, and this tool does no OCR. If you can select the text in your PDF reader, it will work; if you cannot, it will not.

Can I get my Word document back with the highlighting in it?

Yes. Open a .docx, press Download Word, and you get your file back with the code blocks coloured and everything else identical. PDF and self-contained HTML are available too.

Will it change any of my normal text?

Not when you download the format you uploaded. For Word files, non-code paragraphs are never rewritten — their font, size, spacing and styles are the original bytes rather than a reconstruction. For PDFs, the code regions are painted over and redrawn while the rest of the page is left exactly as it was.

Is there a file size limit?

20 MB. Reading happens on the server, so the page stays responsive whatever the size, but a very long document takes proportionally longer to scan. The server sleeps when idle, so the first request after a quiet spell can take up to a minute to wake it.

Is it really free?

Yes, and it is meant to stay that way. There is no account, no trial, no per-page charge and no queue you can pay to skip. The detection costs a fraction of a cent of compute per document, which is the whole reason a promise like that can be made at all.