AiTrans is a Windows desktop app for screenshot translation and AI-assisted follow-up chat. This README is written for end users first: installation, first-time setup, daily usage, and update behavior.
- Capture a screen region and extract the source text
- Translate the captured content through OpenAI official API, DeepSeek API, Zhipu AI API, or another OpenAI-compatible service
- Continue the same task in a lightweight AI chat window
- Keep the current conversation in local cache until you start a new chat or clear history
Choose one package:
- Installer:
desktop-screenshot-translate-<version>-x64.exe - Portable:
desktop-screenshot-translate-<version>-x64-portable.exe
The installer build is recommended for most users.
On first launch, AiTrans will:
- show a floating anchor icon on the desktop
- create a local runtime config file
- open the setup window if translation settings are missing
Please configure at least:
- translation service
- model name
base_urlapi_key
AiTrans currently supports:
- OpenAI official API
- DeepSeek API
- Zhipu AI API
- other services that expose an OpenAI-compatible API
This does not mean every vendor-native API works out of the box. If a provider only offers its own private protocol rather than an OpenAI-compatible API, an extra adapter is still required.
In Settings & Connection, check these first:
- translation service preset
- translation model name
- translation service
base_url - translation service
api_key - source language: auto / Chinese / English / Japanese
- capture shortcut
- send shortcut
- Click the floating anchor or use the capture shortcut
- Drag to select a region
- On multi-display setups, move the pointer to another screen to switch the active capture target
- After capture completes, the conversation window should reopen with the result
- A screenshot becomes a user message in the current thread
- The assistant reply includes source text and translated text
- You can keep typing in the bottom composer for follow-up translation or discussion
- When unpinned, the window collapses on blur
- When pinned, it stays visible
- Left click: expand or collapse the conversation window
- Drag: move the floating anchor
- Right click: open actions such as new chat, clear history, settings, and update check
- The current conversation is stored locally
- Choosing New Chat or Clear History asks for confirmation before removing the cached session
- OCR now prefers a local PaddleOCR runtime, then falls back to bundled Tesseract if PaddleOCR is not configured or fails locally.
- Translation supports OpenAI official endpoints and OpenAI-compatible vendor endpoints.
- For the best OCR accuracy, try setting the source language to English for clear single-line English text, or configure a local PaddleOCR runtime.
If you want better local OCR accuracy for Chinese, English, and Japanese screenshots, prepare a Python runtime with PaddleOCR on the same machine and point AiTrans to it from Settings & Connection.
Suggested steps:
- Prepare a local Python runtime. A system Python or a dedicated virtual environment are both fine.
- Install PaddleOCR and the matching Paddle runtime into that Python environment.
- Open Settings & Connection.
- Fill Local PaddleOCR Python with the Python executable path. If your shell
PATHalready resolves the right Python, you can leave it blank. - Keep the device on
CPUunless you already know your Paddle runtime is ready for GPU. - Save the settings and capture again. AiTrans will try PaddleOCR first and fall back to bundled Tesseract only when the local PaddleOCR path fails.
- The installer build supports update checking
- The portable build does not auto-update
- If an update source is configured, you can trigger an update check from the right-click menu
For development, packaging, and release workflows, use the monorepo docs instead of this user guide.