The toolbar magnifier dropdown mixed search engines and AI chat engines in one
list, which left the Pick-AI button on the sidebar strip looking redundant and
cluttered the thing people reach for when they want to find something on the
web. The dropdown is now search-only; LLMs are picked through the strip's own
AI picker overlay (which already lists them with real icons).
LLM order follows rough global popularity so a fresh profile defaults to the
one most people reach for: ChatGPT, Gemini, Grok, Perplexity, DuckAI, Alisa,
Qwen. Three additions along with it:
* DuckAI (duck.ai) — anonymous, no sign-up, serves GPT-4o mini + Claude 3
Haiku + Llama 3 + Mistral behind one URL. Also wired as a toolbar search
target with pre-filled answer via &ia=chat.
* Alisa (ya.ru/alisa) — Russian-language assistant from Yandex.
* Qwen (chat.qwen.ai) — Alibaba's Tongyi chat; email sign-up, the strongest
non-mainland-phone access to a Chinese model.
Phind stays frozen (Vercel 404 at the root). A saved defaultLlmId pointing at
a frozen engine is cleared so the fallback resolves to the new list[0] rather
than silently flipping to whatever was second.
Engine home URLs can now opt into an override via a `home` field — needed by
DuckAI (its chat lives at the apex of duck.ai but prefers that URL over the
duckduckgo.com stem) and Alisa (chat UI lives at /alisa, not the apex of
ya.ru).
Phind's web UI returns a Vercel 404 at the root as of 2026-10-27, so
users who clicked Open AI with Phind as the default got a dark error
page. The engine is now marked frozen with a reason, which keeps the
Settings row visible with an explanation but takes it out of the
enabled list; anyone whose defaultLlmId was phind automatically falls
back to the next enabled engine (Perplexity).
AI picker icons move from runtime favicon fetch back to bundled files
(ChatGPT, Grok, Gemini return as engine-icons/*.svg — initial-on-color
tiles). The fetch path still runs as a fallback for custom engines and
anything else that lacks a bundle.
The ChatGPT, Grok, Gemini and Phind rows in the sidebar AI picker now
pull each site's own favicon through the same fetch-and-cache path
custom engines already use. The old placeholder color-letter tiles I
shipped in 0.3.106 are removed; the engine's sym emoji is the pre-fetch
fallback until the real favicon lands (cached per profile after that).
Change: fall back from bundledIcon(id) to customFavicon(https://<fav>/)
for built-in engines when no bundled file exists. Perplexity keeps its
existing bundled perplexity.png.
Right-click or chevron-click on the sidebar strip's AI button now opens
a borderless overlay instead of a plain native menu: enabled engines
are listed with their bundled icons, the current engine is highlighted
with the acid accent + check, and a settings row at the bottom jumps
to Search settings. Clicking an engine selects it as the default and
opens it in the left panel in one click.
Bundled engine icons gained an svg-first fallback: engine-icons/<id>.svg
is picked up before <id>.png so plain color-coded letter tiles ship
alongside the existing bitmap favicons. Added tiles for ChatGPT, Grok,
Gemini and Phind — each a simple initial-on-color mark.
Every engine icon was an <img> pointing at Google's favicon service, fetched
again each time the picker, the toolbar or Settings rendered. Offline the
whole list collapsed to the emoji fallbacks, and each open told Google
which engines the user has configured. The catalog's icons now live in
engine-icons/<id>.png inside the app; a custom engine's icon is fetched
once (its own /favicon.ico first, the favicon service as fallback), cached
under the profile, and removed with the engine. Settings no longer falls
through to DuckDuckGo's icon service either. Phind ships no icon: its site
serves none through the bot wall.
