Translate a Foreign Email or PDF on Your PC With ChatRTX
Paste a tricky message from a supplier, landlord or relative and get a plain-English version plus a short summary of what they actually want.
In short: ChatRTX is a free desktop app from NVIDIA that translates and summarizes documents fully on your own computer, with no internet needed. Paste a foreign-language email or drop in a PDF, ask in plain English, and you'll get a translation plus a one-paragraph summary of what the sender is actually asking for.
Have you ever stared at a long email from a supplier overseas or a PDF attachment from a relative abroad, trying to guess what they really want? ChatRTX is built for exactly this moment. It runs an AI model locally on your Windows PC (an LLM — a large language model, the same kind of "brain" that powers ChatGPT), so you can paste text or load a file and ask questions in plain English. Nothing leaves your machine, which is helpful if the email contains private details you'd rather not upload anywhere.
💡 Tip: tap a step’s number when you finish it — a green tick appears and your browser remembers how far you got.
- A Windows PC with a recent NVIDIA graphics card. ChatRTX needs a GeForce RTX 30-series, 40-series or 50-series GPU (the graphics-processing chip inside your computer). You can check by pressing Win + R, typing
dxdiag, and hitting Enter — the "Display" tab will list your card model. - About 15 GB of free disk space. The app and its default AI model together take roughly that much. SSDs work best; old spinning hard drives will feel slow.
- A recent copy of Windows 10 or Windows 11. Anything from the last few years is fine.
- Roughly 20–30 minutes the first time. Installing the app and the default model is the slow part; the actual translating takes seconds.
- No account, no subscription. ChatRTX is free to download from NVIDIA. You will need to create a free NVIDIA developer account to download it, but you do not pay anything.
- A short document to try. Have one foreign-language email (you can paste it as text) or one PDF ready so your first run actually does something useful.
Download and install ChatRTX
Open your web browser and go to the official NVIDIA product page for ChatRTX. Look for a button labeled something like Download or Get Started. Click it, then sign in with your NVIDIA developer account (or create one — it only asks for an email and a password). After the installer file finishes downloading, double-click it and follow the on-screen prompts. The installer may ask you to confirm a folder and agree to the license terms.
You'll know it worked when a new icon named ChatRTX appears in your Start menu and you can launch it without an error message.

Install your first AI model
The first time ChatRTX opens, the main window shows an empty chat area and a sidebar (the narrow panel on the left or right) with a section labeled something like Models or AI Models. That section is empty — the app itself is just a window; the actual "brain" (the LLM) is a separate download. Click the button that says Download next to the model named Mistral or Llama (NVIDIA changes the default model name from time to time, so look for whichever one has a "Download" button beside it). A progress bar appears; let it finish — this is the biggest download, so it may take several minutes on a slow connection. If your hard drive seems too small, try the smaller "Mistral 7B" option first.
You'll know it worked when the model name in the sidebar no longer has a "Download" button next to it and instead shows a small green dot or a check mark.

Paste the foreign-language email into the chat
Click inside the chat box at the bottom of the window (it usually has placeholder text like "Type a message…" or "Ask me anything…"). Then go to the foreign-language email in your browser or mail app, select the whole message with Ctrl + A, copy it with Ctrl + C, and paste it into the chat box with Ctrl + V. The text appears in your language. Press Enter to send. A short pause later — usually a few seconds — the model's answer streams in line by line.
You'll know it worked when text starts appearing word by word in the chat area, and the model ends with a clear English paragraph rather than an error message.

Ask for the translation and the summary in one go
Now send a follow-up message that tells the model exactly what you need. A good instruction names the source language, the target language, and the extra thing you want (a summary of the ask). Think of the chat box as a place where you write the prompt — the instruction you give the AI. The more specific the prompt, the better the answer.
If you don't know the source language, ask the model to detect it first:
You'll know it worked when the reply has two clearly separated parts: a translation and a short summary that names an action, a date, or a question the sender is asking about.

Add a PDF instead of plain text
For longer documents, you don't have to copy and paste by hand. Look near the chat box for a small paperclip icon, an attachment icon, or a button labeled Add file or Upload document. Click it, browse to the PDF on your computer, and select it. The file name appears above the chat box. Now type a prompt that refers to "the attached document" rather than "the text above."
If you don't see a paperclip icon, look inside a + (plus) button, a ⋯ (three dots) menu, or a separate Documents tab in the sidebar — NVIDIA occasionally moves this control between releases. You can also fall back to opening the PDF in another program, selecting all with Ctrl + A, and pasting the text into the chat as in Step 3; PDFs that are scanned images won't paste as text, in which case you'll need an OCR step (a tool that reads text out of images) first.
You'll know it worked when the model writes back using details from the PDF — names, dates, amounts — that you never typed into the chat yourself.

Save the translation for later
Most people want to keep the translation somewhere they can find it again. Click anywhere inside the model's reply in the chat to place your cursor there, then press Ctrl + A to select the whole answer and Ctrl + C to copy it. Open any text editor — Notepad, Word, Google Docs — and paste with Ctrl + V. Give the file a clear name like Supplier-email-translation-2026.txt. ChatRTX itself does not have a built-in "save" button, so this copy-and-paste step is how you keep a record.
You'll know it worked when you can close ChatRTX, reopen the saved file, and read the translation without launching the app again.

- Installing the app but forgetting to download a model. The chat box looks ready, but every reply will be an error. The fix: go back to the Models section in the sidebar and click Download next to the default model. Wait for it to finish before trying again.
- Pasting a scanned PDF (an image of text) instead of a real text PDF. Scanned PDFs look like documents but contain pictures, not selectable words. The fix: run the PDF through an OCR tool first (many free ones exist online), or paste a real text version of the email if you have one.
- Asking for "a translation" without saying the source or target language. The model may guess, and guesses can be wrong for short messages. The fix: always name both languages in your prompt, for example "translate from Japanese to English." If you don't know the source language, ask the model to detect it first (Step 4 has a ready-made prompt for this).
Open ChatRTX, paste the first paragraph of any foreign-language email you've been putting off, and send this single prompt:
In under a minute you'll have a plain-English version and the actual question hidden in the message. That's the whole point of having a translator on your own PC — the hard part is no longer the language; it's just deciding what to reply.
❓ Quick questions
How long does this take?
About 6 minutes — the guide has 6 steps, and you can tick each one off as you go.
Which tool do I need?
This guide uses NVIDIA ChatRTX — but the approach works very similarly in other AI assistants.
Do I need to prepare anything?
- A Windows PC with a recent NVIDIA graphics card. ChatRTX needs a GeForce RTX 30-series, 40-series or 50-series GPU (the graphics-processing chip inside your computer). You can check by pressing Win + R, typing
dxdiag, and hitting Enter — the "Display" tab will list your card model. - About 15 GB of free disk space. The app and its default AI model together take roughly that much. SSDs work best; old spinning hard drives will feel slow.
- A recent copy of Windows 10 or Windows 11. Anything from the last few years is fine.
- Roughly 20–30 minutes the first time. Installing the app and the default model is the slow part; the actual translating takes seconds.
- No account, no subscription. ChatRTX is free to download from NVIDIA. You will need to create a free NVIDIA developer account to download it, but you do not pay anything.
- A short document to try. Have one foreign-language email (you can paste it as text) or one PDF ready so your first run actually does something useful.
What mistakes should I avoid?
- Installing the app but forgetting to download a model. The chat box looks ready, but every reply will be an error. The fix: go back to the Models section in the sidebar and click Download next to the default model. Wait for it to finish before trying again.
- Pasting a scanned PDF (an image of text) instead of a real text PDF. Scanned PDFs look like documents but contain pictures, not selectable words. The fix: run the PDF through an OCR tool first (many free ones exist online), or paste a real text version of the email if you have one.
- Asking for "a translation" without saying the source or target language. The model may guess, and guesses can be wrong for short messages. The fix: always name both languages in your prompt, for example "translate from Japanese to English." If you don't know the source language, ask the model to detect it first (Step 4 has a ready-made prompt for this).
Keep reading
✦ Original step-by-step guide by AI World HQ's AI editorial team. Written in plain language, reviewed for accuracy.
← Back to all stories