Repository navigation
Conversation
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 2, 2026 13:37
97e1edd to
e24a39d
Compare
allozaur
force-pushed
the
allozaur/feat/sidebar-sections
branch
from
November 2, 2026 13:37
cd054b7 to
2fe53cb
Compare
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 3, 2026 10:02
e24a39d to
ab65713
Compare
allozaur
force-pushed
the
allozaur/feat/sidebar-sections
branch
from
November 3, 2026 10:02
2fe53cb to
2286d80
Compare
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 9, 2026 10:05
ab65713 to
4f55550
Compare
allozaur
force-pushed
the
allozaur/feat/sidebar-sections
branch
from
November 9, 2026 10:05
2286d80 to
441d36f
Compare
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 9, 2026 11:29
4f55550 to
3853ab3
Compare
allozaur
force-pushed
the
allozaur/feat/sidebar-sections
branch
from
November 9, 2026 11:56
441d36f to
002e45f
Compare
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
2 times, most recently
from
November 9, 2026 12:44
3265809 to
f11e13f
Compare
allozaur
changed the base branch from
allozaur/feat/sidebar-sections
to
master
November 9, 2026 12:45
allozaur
force-pushed
the
allozaur/feat/model-configuration
branch
from
November 10, 2026 15:00
eb7c0b4 to
03936d0
Compare
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 10, 2026 15:01
5325f6f to
c979bf7
Compare
allozaur
force-pushed
the
allozaur/feat/model-configuration
branch
from
November 10, 2026 15:01
03936d0 to
18e8a0c
Compare
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 10, 2026 15:56
c979bf7 to
d41ca69
Compare
allozaur
force-pushed
the
allozaur/feat/model-configuration
branch
from
November 10, 2026 15:57
18e8a0c to
06d2547
Compare
allozaur
force-pushed
the
allozaur/feat/model-configuration
branch
from
November 10, 2026 16:22
06d2547 to
9818bed
Compare
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 10, 2026 16:22
d41ca69 to
8da8b1e
Compare
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 10, 2026 17:23
8da8b1e to
cc1caa6
Compare
allozaur
force-pushed
the
allozaur/feat/model-configuration
branch
from
November 10, 2026 17:23
9818bed to
f43fa70
Compare
allozaur
changed the base branch from
allozaur/feat/model-configuration
to
allozaur/feat/prompt-library
November 10, 2026 17:23
allozaur
force-pushed
the
allozaur/style/glass-surfaces
branch
from
November 10, 2026 18:04
cc1caa6 to
455ed4b
Compare
allozaur
force-pushed
the
allozaur/feat/prompt-library
branch
from
November 10, 2026 18:04
da46e19 to
2d4c882
Compare
allozaur
force-pushed
the
allozaur/feat/prompt-library
branch
from
November 12, 2026 01:29
2d4c882 to
12fb4f6
Compare
This was referenced Nov 12, 2026
Draft
llama-server refuses browser requests whose Origin does not match its own host unless --cors-origins allows it, and the Vite proxy forwarded the page's Origin (localhost:5173, or a LAN or Tailscale address). Loading a model, saving tools and /v1 calls failed with 403 under `vite dev`. The proxy now calls the server with its own Host and no Origin, WebSocket upgrades included. Sec-Fetch-Site still passes through, so requests from other sites stay refused. Assisted-by: Claude (Opus 5.5 Medium)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Overview
Llama UI, the design system and component library published as
@ggml-org/llama-ui, comes into llama.cpp asui/lib/ui, with three tiers of tokens, matte glass surfaces and Llama Sans and Llama Mono. Llama Chat is built on it, taking its components, tokens, fonts and shared code from the library instead of keeping its own copy, and the library takes the brand's colours and logo from@ggml-org/llama-brand.Demo
Recorded at e24a39d; click the preview for the full video.
Additional information
Part 1 of 13 of a work-in-progress stack on the fork, based on
master, which has SvelteKit 3 since ggml-org#32463 was merged. Its first commit is fork-only and was #70's first: the dev proxy calls llama-server without the page's Origin, sovite devpasses the server's cross-origin check from a LAN or Tailscale address. Upstream, this PR is the other 11 commits onmaster(ggml-org#32468).The Storybook is left alone here, apart from one import that the move of the base components needs. Llama UI's own Storybook, Llama Chat's stories on the new design system and the Submenus story that tests the submenu fixes are in the next PR, #67. The brand and communications pages moved to
ggml-org/llama-brand#10.Read it commit by commit: the one after the dev proxy brings in the library as it was in the temporary
ggml-org/llama-uirepository atedaf518, without its Storybook (119 new files).@ggml-org/llama-ui0.3.0 on npm is published from thisui/lib/ui.Known gaps: the Docker and Nix builds were not run locally. The fonts are still the library's own WOFF2 copies: llama-brand ships them as TTF only, three times the size, so they move to llama-brand's once it has WOFF2 builds.
@ggml-org/llama-brandis a dependency of the published package by its GitHub tarball, as Llama Chat already installs it, now under the same name.tw-animate-cssis now a peer dependency of the library, so llama.pages needs to import it beforetheme.csswhen it moves to 0.3.0.DESIGN.mddescribes the UI of the whole stack, so it also names pages and packages from later PRs. The demo was recorded on an earlier version of this layer and needs recording again.What changed
ui/lib/ui): the tokens, fonts, surfaces, base components and Model UI, built with@sveltejs/package. A CI job type-checks and packs it, andui/README.mddescribes the layout ofui/. The npm scope is@ggml-orgfor every package.DESIGN.mdholds the tokens, roles and rules for agents.ui/lib/uigets the 17 base components it lacked (alert, alert-dialog, button-group, checkbox, collapsible, context-menu, dialog, drawer, dropdown-menu, empty, hover-card, radio-group, scroll-area, skeleton, table, tabs, textarea) and the inset-group and selected-row tokens and surfaces. It now holds all 29.$lib, which in an app points at the app's ownsrc/lib.ui/apps/chatimports base components as@ggml-org/llama-ui/ui/<name>andcn/tvas@ggml-org/llama-ui/utils, andapp.cssimports the library'stheme.css. Itssrc/lib/components/ui,styles/tokens,styles/surfaces.cssandassets/fontsare gone.scripts/vite-plugin-llama-ui.tsresolves the library's entry points and what its sources import from the app's ownnode_modules, and the SvelteKit config gives TypeScript the same paths (a unit test checks they match), so there is one copy of Svelte and bits-ui in the bundle and in the types.ui/lib/uinext to the app and rebuilds when a library source changes. The Dockerfiles copy the library's sources, and the Nix derivation takesui/as its source.--bits-menu-*ones now. The chat group submenu takes the actions menu'sz-[999999]layer, as it no longer renders inside that menu.ModelIdandModelOrgAvataras thin wrappers for its settings. New entry points:@ggml-org/llama-ui/app/<group>and@ggml-org/llama-ui/models.ColorMode,TooltipSide, the model enums,formatParameters,minMemoryTierGb,highlightCode), from@ggml-org/llama-ui/constants,/enums,/codeand/models. It keeps its own where they differ (UI_DATA_ATTRS,formatFileSize).@ggml-org/llama-brand: the library's primitives build on itscss/colors.css, which apps import beforetheme.css, instead of keeping copies of its 25 colours, andLogodraws llama-brand's icon.--elevation-key, through--chat-form-key-shadow): a 0.5px hairline and a 1px drop instead of the control elevation's soft spread. Chosen from the options in Scrapbook › UI › Composer keys; dark keeps the control elevation.updateCurrentNodeawaited the database write and then wrote the node to whichever conversation was active by then. When the conversation closed during the write the reply ended in a streaming error dialog, and when another one opened it got the wrong node. It now sets the node in memory only while the same conversation is still active.Testing
ui/lib/ui:npm ci,npm run check,npm run prepack(svelte-package and publint) and its 11 unit tests pass.ui/apps/chat:npm run check,npm run lintandnpm run buildpass, with 794 unit tests, 270 client tests and the 49 story tests this layer has.@font-facerules, the:where(code, pre, kbd, samp)rule and the:rootblocks move, and thebodybase rule is split in two. Two tokens are new. 21 utilities come from library components the app does not use (Tabs, ModelCard, ModelFamilyCard).scripts/ui-assets.cmake, and the Docker and Nix builds.Demo script
Requirements
Claude (Opus 5.5 Medium)