In a Realtime call, a short pause mid-sentence can make the Realtime provider split the sentence into two turns. If you kept talking before the first half was transcribed, Open WebUI asked the voice model for one reply per half. The voice model hears the whole conversation each time, so both replies answered the full sentence and you heard the same answer twice, the second one about 15 seconds later. Now Open WebUI asks the voice model only once, after the last half, and that one reply covers the whole sentence.
Fixes#32764
Connecting Google Drive asked every user for the drive.readonly scope, read access to their whole Drive. Google classes that scope as restricted, so each self-hosted install needed Google's restricted-scope verification and security assessment, and until then users got the "Google hasn't verified this app" warning. The picker now only requests drive.file, access to the files the user picks, which Google classes as non-sensitive and which needs no such verification. With that scope Google requires the picker to name the Cloud project, so it sends the project number again, taken from the leading digits of the Google Drive client ID. It also opens in list view, as Google recommends for this scope, since the picker gets no thumbnail access with it.
Fixes#29537
On the first day of a month, chats from the day before were listed under "Previous 7 days" in the sidebar instead of "Yesterday". The same happened on January 1st for chats from December 31st. Any chat from the previous calendar day is now listed under "Yesterday", whatever the month or year.
Fixes#31964
Read Aloud and response auto-playback always played at normal speed, regardless of the Speech Playback Speed setting. The speech is played one sentence at a time, and the browser put the speed back to normal every time a new sentence started. The chosen speed now stays in effect for every sentence. Voice mode already respected the setting.
Fixes#31870
In chat replies, inline math written right next to an em dash or en dash, with either $...$ or \(...\), was left unrendered, as in "constants—\(e\), \(i\) and \(0\)—in one equation" where the 0 showed up as plain "(0)". Both dashes are now allowed directly before or after a formula, like spaces, commas and brackets already are, so the math renders whenever a dash touches it.
iPhone photos were only converted to JPEG when the browser reported their type as exactly image/heic. Firefox reports image/heif and some browsers report no type at all, so those photos were uploaded as HEIC and vision models failed with an error. When conversion did run, the JPEG was still uploaded with the HEIC file name and type, so the model was told it was HEIC anyway. HEIC and HEIF photos are now converted whatever type the browser reports and are uploaded as a .jpg with the JPEG type, in chats, channels and notes.
Fixes#28411
On iOS and iPadOS, auto-playback of a finished reply is never heard and the speaker button stays stuck in "speaking", so the first tap only stops a playback that never started. WebKit rejects play() started from a network event, and the audio queue ignored that rejection, so its state never reset. Call mode on the same devices was silent too.
A rejected play() now resets the queue, returns the button to idle and shows a toast (an aborted play from stop or a message switch is ignored). The first user tap or keypress plays a 10 ms silent clip on the shared audio element, which WebKit then allows to play later without a gesture; if that attempt fails it retries on the next gesture. Call mode plays unmuted: WebKit pauses an element that is unmuted after play() outside a gesture, even once unlocked.
The unlock and call mode change follow the reporter's on-device tests (iPhone iOS 27, iPad iPadOS 26.6.2). Verified in Chromium with autoplay restricted: the base queue wedges and drops later chunks; with the fix it reports the error, recovers, the unlock plays once and never interrupts audio already playing, and chunks queued during the unlock still play.
Fixes#30262
Mermaid diagrams and the shared SVG sanitizer now accept only same-origin and data: references. Image URLs, class styles and directive config are checked on the parsed diagram before it renders, and the sanitizer drops attribute values and stylesheet rules that point at another origin. use elements keep local #id references only.
SVGPanZoom and the SVG file preview call the shared sanitizer instead of keeping their own configs, so SVG artifacts and uploaded SVG files follow the same rule. Uploaded SVGs that reference external sprites now render those parts blank, which is the point of the change.
A diagram that references an external resource now reports an error instead of rendering, and an SVG artifact keeps everything except the rules that reference one.
Uploading an .svg to a chat attached it as a vision image input, so the model received a data URI it could not decode. PIL-backed servers answered "cannot identify image file" and OpenAI answered "The image data you provided does not represent a valid image". No setting made it work.
SVG now takes the ordinary file upload path, so its XML source is extracted and indexed and the model can answer questions about it. Rasterizing was the alternative and it would have discarded the part of an SVG a model reads best, the source itself. Raster formats are untouched and still go up as image inputs.
A shared helper replaces the ad hoc image/ prefix checks at the points that decide image input versus document, on both ends. It normalises the content type first, because a stored "image/SVG+xml" or a trailing charset parameter slipped past a plain comparison.
One behaviour change worth knowing: an SVG now needs the model to have the file upload capability, where before it rode in as an image.
Fixes#30100
The branch navigation handlers each carried their own copy of the loop that walks a message's childrenIds down to the deepest child, eleven copies in total across the single-response view, the multi-response view and Chat.svelte. They now call one getDeepestChildId helper in the frontend utils, which tracks the ids it has already visited and stops when one repeats, the same way the message list build already does.
The helper also absorbs the "start from the last root message" fallback that two of the call sites repeated inline, so every site is now a single call.
Link URLs rendered from markdown, citations and web search results now pass a scheme check before they reach an anchor. A new safeLinkUrl helper sits beside isValidHttpUrl and keeps http, https, mailto, tel and relative URLs, returning undefined for anything else so the label renders without a link. Markdown links with an unusual scheme (ftp, sms, and application deep links such as obsidian or vscode) render as plain text from now on.
The citation checks move off a substring test for "http" onto isValidHttpUrl, which is what Citations.svelte already uses for the same question. That also drops two long-standing quirks: an uppercase HTTP:// source never rendered as a link, and a filename merely containing "http" rendered as a dead external one.
* fix: render channel mentions whose ID contains spaces or parentheses, such as workspace model IDs
* fix: accept any non-whitespace mention ID, matching the model ID rule on dev
