Summary
MailBilingual AI adds immersive, paragraph-by-paragraph bilingual translation to the Thunderbird message reader. Translations appear directly under each original paragraph with SSE streaming, powered by your own AI provider — OpenAI-compatible APIs (including self-hosted gateways and Zhipu/BigModel), DeepSeek, Anthropic Claude, or a fully local Ollama. Your API key is stored encrypted (AES-GCM) and message content is only ever sent to the provider address you configure.

Detailed description
What it does

When you open a message, a small floating toolbar appears above the message body and the translation starts automatically. Each translated paragraph is inserted right under the original one, preserving the original layout, images and formatting. Translations stream in as they are generated (SSE), with per-paragraph retry buttons for any failures.

Where to find the entry points

In-message floating toolbar — appears at the top of every opened message: switch between Bilingual / Translation only / Original only views, re-translate, or toggle translation on/off.
Toolbar popup — click the MailBilingual AI icon in Thunderbird's main toolbar for a quick panel: enable/disable, view mode, translation style, target language, today's token usage, diagnostic logs, and a shortcut to the settings.
Settings page — Add-ons Manager → MailBilingual AI → Options (opens in a tab): provider credentials, prompt style, glossary, cost controls and privacy options.
Keyboard shortcuts — Ctrl+Shift+Y toggles translation on/off, Ctrl+Shift+U cycles the view mode (both configurable in Thunderbird's shortcut settings).
How to use it (quick start)

Open the settings page (toolbar popup → ⚙ Settings).
Choose a provider and enter its Base URL — the add-on ships with no preset endpoints, so you type your own, e.g. https://api.openai.com/v1, https://api.deepseek.com, or http://localhost:11434/v1 for a local Ollama. OpenAI-compatible endpoints ending in /chat/completions are also accepted as-is. Encrypted https:// addresses are required; plain http:// is only accepted for localhost addresses on your own machine.
Enter your API key (not needed for local/intranet addresses such as Ollama) and pick a model, then click Save settings. On the first save, Thunderbird will ask you to grant network access to that provider address — this add-on requests no host permissions at install time, so you stay in control of where your data goes. Use Test connection to verify everything works.
Open any message — the bilingual view appears automatically. Pick a target language and style in the popup or settings (Business / Academic / Casual / Literal; the Custom style is available in the settings page).
Cost-saving & email-aware features

Content-hashed local paragraph cache, reused across emails (30-day TTL).
Smart batching: on-screen paragraphs are merged into a single LLM call.
Filters for signatures, quoted replies, code blocks and legal disclaimers.
Rate limiting, request spacing and a daily token limit.
Optional glossary (CSV/JSON import & export) and sensitive-data masking (phone numbers, emails, bank cards, ID numbers) before anything leaves your machine.
Privacy

Message bodies are sent only to the provider address you configure, solely for translation. The add-on ships with no preset endpoints — every provider address is typed in by you. No telemetry, no analytics, no data collection.
All outgoing requests are enforced over encrypted https:// connections; plain http:// is only accepted for localhost (loopback) addresses, where data never leaves your machine, and the add-on declares no http:// permissions at all.
Until you enter credentials, a banner makes clear that nothing is sent anywhere, and a background gate enforces the same rule.
API keys are encrypted at rest (PBKDF2 + AES-GCM) and never shown again after saving.
Local-only mode restricts all requests to localhost/intranet addresses.
See the add-on's privacy policy (Manage Authors & License) for full details.