The chat metrics bar reported In/Out/Reasoning/Total/cost only, so a
playground user had no signal that provider prompt caching worked. The
cached-token counts were already visible in the Logs drawer, which meant
the answer to "does caching work here" lived on a different page.
Adds cacheReadTokens and cacheCreationTokens to TokenUsage and renders
them as two chips, reusing the prompt-cache tooltip wording already
introduced for the Logs drawer so both surfaces say the same thing.
A single helper, extractPromptCacheTokens, normalizes the three usage
shapes the playground consumes: Anthropic Messages
(cache_read_input_tokens / cache_creation_input_tokens), chat
completions (prompt_tokens_details) and the Responses API
(input_tokens_details). All three producers call it instead of parsing
per surface. Counts that are absent, zero or non-finite are dropped, so
providers without prompt caching render exactly what they render today.
CLI session tokens minted by /sso/cli/poll set team_id and team_alias but
never team_models or team_model_aliases, so the token carried a team with
none of that team's grants. /v1/models bails out to "unrestricted" when both
key_models and team_models are empty and listed the whole proxy, and team
model aliases never resolved because both can_team_access_model and the
pre-call rewrite read team_model_aliases off the token.
The team data was not close at hand: _fetch_cli_sso_team_details projected
full team rows down to team_id and team_alias before they reached the mint.
Widen that projection to include the team's models and its joined alias
table, and populate both fields at mint time.
Also stop writing the user's personal allowlist into the key models slot
when a team is bound, matching virtual-key semantics where a team-bound
credential is governed by the team grant.
Because an empty team grant is itself a real value meaning unrestricted, a
team whose grants cannot be resolved must not be minted as empty: that is
the same "unrestricted" bail-out this fix exists to close. The poll now
refuses to mint when the selected team has no complete cached detail.
That refusal is only safe because a login can no longer be pinned to a team
whose grants will never resolve. Deleting an organization drops its team
rows but leaves the memberships behind, so the login now offers only teams
whose rows still exist, and a lookup that fails outright fails the login
rather than caching a session that silently drops every team.
The repository's comment policy keeps reasoning out of the source, so the
notes on the delete dialog, the tooltip wrapper, and the test query choices
move to the pull request instead.
The assertion pinned the Tailwind track string, which jsdom can never
evaluate: it does no layout, so the test could not fail for the reason
that matters, and its pattern accepted any minmax minimum, staying green
if that minimum changed enough to break the layout outright.
Only a viewport-resize browser test can observe this, which belongs in
tests/e2e/ui rather than a route's unit tests. The responsive behaviour
itself is unchanged; the widths it reproduces are recorded in b341a22339
Replaces antd and Tremor with shadcn primitives across the six route-owned,
form-free playground components: the compare view and its panel, message input
and unified selector, plus the realtime playground and the agent builder.
Markup only, no behaviour change. The characterisation tests added in the
previous commit are untouched here and stay green through the swap.
Adds ui/slider.tsx via the shadcn CLI and retires the six antd
no-restricted-imports suppressions the migration made obsolete.
The antd Row and Col the migration replaced never squeezed the summary
cards below their content width; they wrapped onto a second line instead.
Measured on a live dashboard, antd laid out five per row at 1280 and 1024,
then four plus one at 900 and three plus two at 820, never narrower than
about 126px. A fixed grid-cols-5 kept all five on one line and compressed
them to 86px at 820, so the metric values overflowed their cards.
An auto-fit track with a 7rem minimum reproduces antd's wrap points and
card widths exactly at all four measured viewports, and is identical at
1280, so the route's visual baseline is unchanged.
The added test fails against grid-cols-5 and passes against the track
Rewrite the playground tests that reached for antd class names so they
locate controls by role, text, placeholder or lucide icon instead. Add
characterisation suites for AgentBuilderView and RealtimePlayground,
which had none, including tab state that must survive a round trip
through another tab.
Every assertion here passes against the current antd components.
Replaces the antd Button, Row, Col, Spin and Typography usage in
GuardrailsOverview with the shadcn button, a CSS grid and the house
loading wrapper, and swaps the ant-design icons for lucide equivalents
that the already-migrated GuardrailDetail sibling uses. The busy state
keeps the aria-busy attribute the antd Spin exposed, so the loading
contract is unchanged for assistive tech.
The route's other analyzer-listed file, GuardrailsMonitorView, keeps its
DateRangePickerValue type import because the shared AdvancedDatePicker
prop contract requires it, so there is nothing to migrate there.
Retires the file's now unused no-restricted-imports suppression
Ports the 21 route-owned guardrails components off antd and Tremor onto the
installed shadcn base-vega primitives. Forms, tables and shared components stay
where they are: the route analyzer buckets add_guardrail_form, guardrail_info
and six others as DEFERRED, and eleven components under src/components as
SHARED, so none of them are touched here.
Three behaviours needed explicit handling because Base UI does not match antd by
default. Base UI Tabs mounts only the active panel, so the playground draft and
every panel's local state would reset on a tab switch; TabsContent now carries
keepMounted. ComboboxList takes a function child to render the filtered items,
and mapping over the array instead renders everything and silently disables the
type to filter behaviour antd gave for free. SelectContent gets
alignItemWithTrigger={false} so popups anchor below the trigger the way antd's
did, which also removes a focus race that made the template picker
intermittently unclickable.
The three dialogs reached from inside the still-antd wizard need a layer above
its z-index of 1000 to be visible at all.
Adds three tests rather than editing the characterisation tests from the
previous commit: a guard that every tab panel stays mounted, and one filtering
regression each for the pattern and mode dropdowns.
Prunes the guardrails entries from eslint-suppressions.json.
Replaces antd and Tremor markup on the usage route with the installed shadcn
base-vega primitives. Behaviour is unchanged: the route's characterisation
tests were rewritten to role and text queries in the previous commit, proven
green against the antd components, and pass through this commit unedited.
Tremor tab panels stayed mounted once rendered, so every migrated TabsContent
and the collapsible model sections in activity_metrics carry keepMounted to
keep view-mode and expansion state alive across tab switches.
* chore(ui): remove Agent Platform announcement bell from navbar
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* feat(ui): highlight Auto Router in the navbar announcement
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(ui): open Auto Router docs link in a new tab
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* Revert "fix(ui): open Auto Router docs link in a new tab"
This reverts commit 3be861bafe.
* chore(ui): title the navbar announcement LiteLLM Auto Router
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: Mubashir Osmani <mubashir@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Covers the header and export action, all five summary metric cards, the
table toolbar heading, the evaluation settings modal wiring, the busy
state and the request failure message. Every assertion is role, title or
text based so it holds against both the antd markup and its shadcn
replacement, letting the migration commit land without editing this file
create_view_tolerating_race took the module's _db = Any. It now takes a
Protocol naming the single operation it calls, so the contract is checkable
at its call sites without retyping the rest of the module.
Kept free of Any deliberately: an earlier version typed the Protocol's
parameters as Any and pushed create_views.py from 26 basedpyright errors to
30 by adding reportExplicitAny. This version measures identical to the
baseline on both create_views.py (26) and utils.py (1355).
Against a real Postgres the previous commit still died on MonthlyGlobalSpend:
only 2 of the 8 creation sites went through the tolerant helper, so the losing
replica re-raised on the first unguarded one and skipped the rest.
The regression test now makes every CREATE lose the race and asserts all 8 are
still attempted, which fails on the partial fix.
Rewrites the usage route's library-coupled assertions to role and text
queries so they characterise behaviour rather than the widget library that
happens to render it. Every test here is green against the current antd and
Tremor components and is meant to survive a migration unedited.
Drops the wholesale vi.mock of antd, @ant-design/icons and @tremor/react in
UsagePageView, UsageViewSelect and activity_metrics, and drives the real
controls instead. Two of those mocks were hiding behaviour: activity_metrics
asserted panel order through a heading role that only existed because the
mock faked an h2, and the Tremor tab stubs flattened panel selection away
entirely.
Replaces DOM-shape lookups with anchors that do not move: a chart is found
from its own heading rather than a fixed wrapper depth, an active panel is
identified by the inactive markers both tab libraries set, and select options
are matched by text since antd's real options carry no option role.
Adds the missing characterisation test for team_multi_select, and covers the
mount contracts the route depends on: Tremor keeps every tab panel mounted,
and antd Collapse keeps a section mounted once it has been expanded, which is
what preserves the view-mode state a model section owns.
Rewrite the PatternModal test off antd class selectors onto role and text
queries, and add characterisation tests for the guardrails components that
had none: the keyword modal, the content filter display and configuration,
the guardrail garden and the custom code modal.
Also covers two behaviours the migration must preserve: the action and
severity dropdowns in the content filter tables, and test playground state
surviving a tab switch away and back.
Every assertion here passes against the current antd and Tremor components
so the same file can prove the shadcn versions unedited.
Every replica booting against the same fresh database sees each view as
absent and issues the CREATE. Postgres fails all but one with a
duplicate-object error, and that exception propagated out of
create_missing_views, so every view after the first was never created and
/global/spend* 500'd for the life of the deployment.
Losing that race reaches the desired end state, so treat it as success.
Genuine DDL errors still propagate.
Every case in TestAccessControl asserted that something was refused. A gateway
that denied the allow-listed model too would have passed all of them, so the
suite could not tell "denied correctly" from "broken outright".
Adds the positive half: a key allow-listed for gemini-2.5-flash can call it and
gets back a real completion rather than a 200-wrapped error.
Also tightens the unknown-model case. It accepted any valid JSON, so a bare
"{}" or even "null" satisfied it. It now requires the OpenAI-shaped error
envelope with a message a client can actually surface, parsed through a typed
model instead of json.loads.
Completed batches that contain only failed requests do not generate an
output file, leaving output_file_id unset while the failures are recorded
through error_file_id instead.
The completion handler attempted to read the output payload regardless of
whether an output file actually existed. During retrieve polling this caused
the logging pipeline to fail with "Output file id is None cannot retrieve
file content", preventing normal completion bookkeeping from running.
Skip output retrieval when no output file is available and return an empty
batch summary (zero usage, zero cost, no model entries). The lower-level
file retrieval helper still reports an error if it is called directly with
an invalid or missing file identifier.
Closes#33987
Prettier flagged the two test files added while fixing review findings.
Migrating these components off Ant Design also retired the lint suppressions
they carried, so prune those 15 entries and leave the unrelated ones for the
PRs that made them stale.
