Localization · checked 2026-10-19

Can I vibecode DeepL Pro?

KINDA· weekend project

Text translation is the easy half: a current LLM or an open translation model, a glossary list injected into the prompt, a formal or informal switch and a small web page. Word files are the harder half. Keeping styles, tables and tracked changes while swapping text run by run takes a day of fiddling, and PowerPoint and PDF are worse. The quality ceiling is also someone else's model: DeepL's own engine is the product, and a self-built one is only as good as the model you point it at, notably weaker for rarer language pairs on local models. Fine for a few hundred thousand characters a month, not for a team's compliance review.

$8.74/moprice per seat

$105you'd save a year

weekendbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyIntermediate · Some system setup or keys.

TimeA weekend; Word formatting is the slow part

Claude Code, Codex or CursorAny of them can run the prompt.
Node.js 22better-sqlite3 compiles a native module, so build tools must be installed.
A model endpointAn API key for a hosted model, or Ollama running locally if text mustn't leave your machine.
Your DeepL glossariesDownload each as CSV from the web translator or the desktop app.

Before you buildDeepL Free covers 50,000 characters and one file a month. If you translate a few emails a week, you may not need to pay or build.

1. Install one agent: Claude Code, Codex, Cursor.
2. Make an empty folder and open it in the agent.
3. Paste the prompt below, press enter, and answer its questions as it builds.

The prompt

Build me a personal translator to replace DeepL Pro. Requirements:

- Node.js 22 + TypeScript: a Fastify server and a React page built with Vite. `npm start`
  serves http://localhost:4830 with two panes, a language picker and a formal/informal switch.
- Engine: any OpenAI-compatible chat endpoint (`openai` SDK; OPENAI_BASE_URL, OPENAI_API_KEY and
  MODEL in .env, so Ollama works too). Detect the source language with `franc-min`. The
  system prompt must return only the translation, keep markup and placeholders untouched,
  and apply the tone setting.
- Storage: better-sqlite3 in ./data/translator.db: glossaries (source term, target term,
  language pair, notes) and a translation memory of finished sentence pairs.
- Glossary: CSV import and export in DeepL's format (`term,translation,SRC,TGT`). Send the
  terms found in the text with every request, then check that each target term appears and
  list the ones that don't.
- Memory: split text into sentences (`Intl.Segmenter`); reuse an exact match from memory
  and only send new sentences to the model; show which parts came from memory.
- Documents: translate .docx by unzipping with `jszip`, editing the w:t text of each
  paragraph with `fast-xml-parser` and writing a new file with styles, tables and footnotes
  untouched. Check it with three sample files and open them in Word or LibreOffice.
- A "history" page keeping the last 7 days of translations, with a delete button.
- Single user. No accounts, no telemetry, local-first: only the configured endpoint sees your
  text, and the README says so in one sentence.
- Out of scope: PDF, PowerPoint and Excel files, browser extensions, Office add-ins, voice.
- README: .env keys, the cost per 100,000 characters for your model, and a warning that
  quality for rare language pairs depends on the model you choose.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.

Share on X

Max MiskowSponsoredAI agents, automations and MVPs, built for your company.Visit site ›

Then keep going

Next prompts, same folder.

The first prompt builds the core. Run these one at a time in the same folder to grow it into your own DeepL Pro.

Add a folder mode to the translator in this folder. Requirements:

- A command `translate-dir <src> <dest> --to de,fr` that reads .md and .txt files,
  keeps front matter keys and code blocks untouched, and writes one folder per language.
- Use the glossary and memory; skip files whose content hash hasn't changed since the last run.
- Print a table of files, sentences from memory and sentences sent to the model.
- Update the README.

Add a check mode to the translator in this folder. Requirements:

- After translating, translate the result back to the source language and show both texts
  side by side with differences highlighted (`diff` on words).
- Flag numbers, dates and names that differ between the source and the back-translation.
- It's off by default, because it doubles the cost; a checkbox turns it on.
- Update the README.

Add a small desktop helper to the translator in this folder. Requirements:

- A Node script using `clipboardy` and a global shortcut library for the OS you run on:
  press the shortcut, send the clipboard text to the local server, and show the translation
  in a notification.
- The target language comes from a setting; formal or informal tone too.
- Keep nothing on disk except the normal history entry, and show errors in the notification.
- Update the README with the autostart steps for your OS.

What you lose

DeepL's own trained translation engine, which is what the subscription mainly buys
PowerPoint, Excel and PDF translation that keeps the layout
Browser extensions and the Word, Outlook and Google Docs add-ins
Phone apps and desktop apps for Windows and Mac
A written promise that your text is deleted after processing and never used for training

Why people still pay

People pay for the quality of the engine, and for the documents: a 30-page Word file comes back with fonts, tables and footnotes intact. The no-training promise and ISO and SOC 2 certificates also matter to anyone sending client text through it.

Moatproprietary models integrations execution polish

Do the math

DeepL Pro or your own copy?

Seats

Over

Your usage

Count my time12h ×$/h

DeepL Pro, 1 yr$105

Your own copy, 1 yr$42running + build session

You keep$63pays for itself in 2 months

Estimate · A local model costs nothing to run; a hosted model costs a few cents per 100,000 characters. The build is token usage on your agent plan.

Leaving DeepL Pro

Take your data, then cancel.

1 · Take your data out of DeepL Pro

1Glossaries: open the glossary window in the web translator or a desktop app and download the glossary; you get a CSV file (comma-separated).
2Translated documents are files you already downloaded; DeepL doesn't keep a library of them for you.
3Translation history is kept for 7 days unless you delete it sooner, so copy out anything you need now.
4A glossary CSV can be uploaded again later, one or several language pairs per file.

2 · Cancel DeepL Pro

1Open your DeepL account and go to the Subscription tab.
2Click Cancel subscription at the bottom of the page.
3The cancellation takes effect at the end of your current billing cycle; Pro features keep working until then.
4To check a refund, log in and use the chat on the DeepL support page.

Cancel before the free trial ends if you don't want to pay: the full year is charged up front after the trial on annual billing.
A refund is possible if you cancel within the trial or inside your statutory right of withdrawal (Section 11 of the terms); the support chat checks eligibility.
Downgrades, for example from Team to Individual, take effect at the end of the current billing period.
Free and Individual hold only 5 glossary entries per language pair, while Team holds 1,000: a larger glossary should be exported before you downgrade.

Checked 2026-10-19 against DeepL Pro's docs: 1, 2, 3, 4, 5

The escape hatch

Don't feel like building it? These are already free.

LibreTranslateOpen sourceAGPL-3.0 translation server with a web page and an API, offline capable and built on Argos models.★ 17k · active Oct 2026 · docker to self-host
WebVisit ↗Argos TranslateOpen sourceMIT Python library and command line for offline translation, with a package per language pair.★ 6.6k · active Aug 2026 · self-host, real ops
mac, Windows, LinuxVisit ↗Firefox TranslationsFreeMozilla's built-in page translation in Firefox; models run on your device, so text doesn't leave the browser.★ 364 · active Sep 2026 · they host it
WebVisit ↗

All 4 free alternatives to DeepL Pro ›

The numbers

DeepL Pro pricing

Plan	Monthly	Annual, per month	What you get
Free	$0/user	$0/user	50,000 characters a month, 1 file translation a month up to 5 MB, 1 glossary with 5 entries per language pair, 1 saved translation.
Individual	—	$8.74/user	300,000 characters a month, 3 file translations a month up to 30 MB, 1 glossary with 5 entries per language pair, 1 saved translation, formal and informal tone, no data training.
Team	—	$28.74/user	1,000,000 characters per user a month, 20 file translations, 5 glossaries with 1,000 entries per language pair, 5 shared style rule lists, bulk file translation, SSO for 50 or more users.
Business	—	$57.49/user	Unlimited characters under fair use, 100 file translations up to 100 MB, unlimited glossaries, translation memory, SCIM, Write Pro included.
Enterprise	custom	—	Custom character and file limits, bring your own key, 99.9% availability, dedicated account team.

Free tier: 50,000 characters and one file a month; the Free glossary keeps only 5 entries per language pair.
Billing: Monthly and annual billing both exist, with a 30-day free trial on Individual and Team. The pricing page lists $8.74 a month billed annually (about $105 a year, paid up front after the trial); the monthly-billing price wasn't readable in the page text we got, but it's higher, since the page advertises a 16% annual saving. Billed in USD outside the euro countries.
Hidden costs: Write Pro, the rewriting assistant, is an add-on at $7.49 per user a month billed annually; Business includes it. The API is a separate product: Growth is $26 a month plus $27.50 per extra million characters, and the free Developer tier now gives a one-time 1 million characters, not a monthly allowance.
Source: www.deepl.com

Also on the list

LanguageTool$24.90/mo · AI writingYES

Crowdin$59/mo · LocalizationYES

Lokalise$169/mo · LocalizationYES

Questions

DeepL Pro, briefly

Not answered here? Tell us and we'll add it.

Ask a question ›

Free translates 50,000 characters a month and one file of up to 5 MB, with one glossary of 5 entries per language pair.

Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: DeepL's own trained translation engine, which is what the subscription mainly buys, PowerPoint, Excel and PDF translation that keeps the layout.

About $8.74 a month on the Individual plan per seat, $104.88 a year (checked 2026-10-19).

DeepL's own trained translation engine, which is what the subscription mainly buys; PowerPoint, Excel and PDF translation that keeps the layout; Browser extensions and the Word, Outlook and Google Docs add-ins; Phone apps and desktop apps for Windows and Mac; A written promise that your text is deleted after processing and never used for training. If any of those matter to you, keep paying.

Yes: LibreTranslate (open source), Argos Translate (open source), Firefox Translations (free).