v5.0.0pre3

• [All APIs] The "Show differences" view of the proofreading and rewriting prompts is now interactive: for every change both versions are shown side by side, the original in red and the suggested one in green, and clicking one keeps it. A counter, a "Words"/"Sentences" toggle to compare the texts word by word or sentence by sentence, "Accept all"/"Reject all" buttons, an "Edit manually" mode to change the text by hand, keyboard navigation and a "Use this answer" button that applies exactly the choices made complete the view, which replaces the previous read-only diff [#829].
• [All APIs] A new option sets whether the changes are compared word by word or sentence by sentence when the comparison view opens [#829].
• [All APIs] Fix: the "Show differences" view no longer loses the line breaks of the answer, so a multi-paragraph answer is no longer compared to the original as a single run-together line [#829].
• [All APIs] Fix: the AI answer no longer shows literal <br> tags in the chat window, and they are no longer inserted in the mail when using "Use this answer". The text of the message in composition is no longer sent to the AI with a <br> on every source line, so the model has no reason to write them back, and any <br> still returned is now shown as a real line break.
• [All APIs] Fix: the line breaks of the answer are now preserved when it is inserted in the mail with "Use this answer". A single line break was previously visible in the chat window but silently lost in the message, joining two lines (a closing greeting and the signature below it, for example) into one. Repeated line breaks now always produce a single empty line, instead of a variable amount of blank space depending on how many the model wrote.
• [All APIs] Fix: no more extra empty lines between the AI answer and the signature or the quoted text when the answer is inserted in the mail. A fixed empty paragraph was added before the signature or the quote, leaving a large gap: now there is a single empty line [#849].
• In the Custom Prompts page the {%placeholder%} tokens are now highlighted while editing a prompt too, and not only when reading it. The prompt text is shown in a monospace font in both modes [#833].
• The placeholder autocomplete list now opens at the text cursor instead of below the whole text box, is no longer cut off inside the prompts table, and shows a description of what each placeholder does. It can be accepted with Tab as well as Enter, closed with Escape, and is keyboard accessible [#833].
• The {%placeholder%} tokens are now highlighted while editing in every prompt text box of the add-on: the Custom Prompts and Data Placeholders pages, and the Translate, Summarize, Add Tags, Spam Filter, Get Task and Get Calendar Event settings pages [#833].
• A misspelled placeholder, one that is not available for the prompt type, or one missing its closing %} is now highlighted as a warning while editing, with a tooltip explaining the problem [#833].

Added in 5.0.0pre2

• [All APIs] The AI chat window has been completely redesigned: a new header bar showing the API and the model in use, conversation turns with the assistant avatar, a floating status pill above the input reporting the current state (waiting for the server, streaming, done, error) and a full light/dark theme that follows Thunderbird [#842].
• [All APIs] In the AI chat window every answer now has its own actions toolbar, so it's possible to use, copy or save any answer of the conversation and not only the last one [#842].
• [All APIs] A "Copy" button has been added to the answers in the AI chat window: it copies the answer as plain text, or only the selected part of it if there is a selection [#842].
• [All APIs] The thinking block is now always shown when the model returns reasoning content, even if the thinking option of the connection is not enabled.
• [Google Gemini API][OpenAI Comp API] Improved the detection of the thinking content in the response stream.
• [OpenAI API] The thinking block is correctly handled and reported on the webchat. The reasoning summary parameter must be set [#773].
• [All APIs] Fix: the reasoning content is no longer included in the result of the special prompts (add tags, spam filter, calendar event, task), where the response is parsed and not displayed.
• Fix: The timezone selector of the calendar event and task settings pages now lists all the timezones known to Thunderbird, and has a search box to quickly find one. The previous list was missing every timezone with a fractional offset, like UTC+05:30 (IST), UTC+05:45, UTC+03:30, UTC+04:30, UTC-03:30, UTC+08:45 and UTC+10:30 [#843].

Added in 5.0.0pre1

• Redesigned the settings page to reduce visual complexity, moving advanced options to a dedicated section to improve the onboarding experience for new users [#739].
• A new guided Setup Wizard has been added: choose your AI provider, enter and test the connection, and enable the AI features you want. It can be launched from the welcome page, from the toolbar popup when no AI is configured, and from the settings page [#830].
• A prompt's placement is now managed entirely from the Menu Order page: the "Show in" selector and the enabled checkbox have been removed from the prompt editor, which now shows a "Menu position" button that opens the Menu Order page and highlights the prompt. New prompts start visible in the popup menu [#824].
• The Custom Prompts page UI has been improved [#833].
• The Data Placeholders page UI has been restyled to match the new Custom Prompts page design [#838].

To try this release download and install the thunderai-v5.0.0pre3.xpi file.