local-llm-translator by moccot-zheng

Overview
Local LLM Web Translator is a Firefox extension that uses local or cloud LLMs to translate entire web pages or selected text. Version 2.0.0 introduces bilingual UI, encrypted API Key storage, translation testing, configurable system prompts, and more.
Installation and First Setup
Install the extension in Firefox.
On first use, you will be asked for data collection consent.
Local model: data never leaves your device.
Cloud API: data is used only for real-time translation.
In the options page, select a translation backend:
Local model: enter the local server address, usually http://127.0.0.1:8080.
Cloud service: choose a provider and enter the API Key and model name.
How to Translate
Translate a page: click the toolbar icon and select “Translate Page”, or use the context menu.
Translate selected text: select text and choose “Translate Selection”.
Stop translation: click “Stop Translation” or navigate away from the page.
Retranslate: click “Retranslate” after a completed translation.
Options
Language: switch between Simplified Chinese and English.
API Key management: encrypted automatically.
System prompt: customize translation behavior.
Thinking mode: disable to reduce latency.
Code block handling: choose whether to skip code/pre tags.
Frequently Asked Questions
Why does nothing happen when I translate?
Check if the model is running or if the API Key is correct.
Does it support Chrome or Edge?
Currently Firefox only.
Will it break page layouts?
No, it replaces text in place.
What happens if translation fails?
It retries twice, then keeps the original text and shows an error.

Feature Description
Local LLM Web Translator v2.0.0 Feature Description
Core Features
Multi-provider translation backend: Local, DeepSeek, Qwen, GLM, Moonshot, Baichuan, and custom OpenAI-compatible APIs.
Bilingual UI: automatic language detection with manual override.
Encrypted API Key storage: AES-GCM encryption, keys exist only in background memory.
Translation test panel: preset examples, custom input, duration and character count.
Configurable system prompt: fine-grained control over translation style.
Thinking mode control: disable chain-of-thought to cut latency by 60%–80%.
Two-phase visible-first translation: viewport content first, hidden content later.
Data consent mechanism: blocks all outbound requests without consent.
Intelligent text processing: hybrid chunking, smart skipping of code, links, numbers.
Multiple translation triggers: full page, selection, stop, retry, connection check.
Dynamic content support: IntersectionObserver and MutationObserver.
Progress indication: floating badge showing real-time status.
Error handling and reliability: timeout protection, retries, graceful fallback.
Technical Architecture
Manifest V2 with persistent background page.
Pure vanilla JavaScript, no external dependencies.
Dual-process design: Content Script and Background Page.
WebCrypto API for encryption.