The tool test panel drove its argument fields through an antd Form, so the
call payload was whatever rc-field-form happened to have mounted. It now runs
on react-hook-form with shadcn controls, and the payload itself lives in
toolCallArguments.ts as a pure function of the schema fields plus the entered
values.
Fields bind by index rather than by name, because an MCP tool's JSON schema
can name a property anything: a key containing a dot would be one flat key to
antd but a nested path to react-hook-form. Binding to args.0, args.1 and
zipping back to the real keys at submit time keeps the emitted arguments
identical whatever the server calls its properties.
Coercion, the blank filter, the required and JSON rules, and the params
wrapper for nested-object schemas all keep their previous behaviour, and the
neutral colours in the panel move onto tokens so it reads correctly in dark
mode.
Removes one named callback from a team and leaves the team's other callbacks
registered and firing. Before this, the only removal route was
POST /team/{team_id}/disable_logging, which clears every callback at once, so a
tenant sharing a team could not deregister its own integration
The handler filters metadata["logging"], keeps the survivors encrypted, refreshes
the cached team so the removal applies to keys that are already live, and emits a
redacted audit row, matching what the add and disable routes do
Resolves LIT-5161
* refactor(ui): move the add model and credential forms off tremor
TextInput becomes the shadcn Input, Text becomes a sized paragraph, the
advanced settings Accordion becomes a bordered Collapsible, and the
Team-BYOK Switch moves to the Base UI switch with onCheckedChange plus an
aria-label. That switch stays wrapped in a span so the antd Tooltip still
shows on hover while it is disabled for non-premium users, matching what
the tremor wrapper div did. provider_specific_fields keeps antd's
Input.TextArea through an AntdInput alias so its antd import stays a single
statement. Prunes the tremor no-restricted-imports suppressions these files
no longer need.
* fix(ui): keep the reveal toggle on the provider secret fields
tremor's TextInput drew its own show/hide button whenever the type was
password, and the shadcn Input is a plain native input, so every provider
secret this form renders (API keys, client secrets, and the same fields
inside the add credential modal) lost that affordance.
Puts the password branch on antd's Input.Password, which is what the other
dynamic credential forms in the dashboard already use, so the reveal comes
back and the antd Form.Item wiring stays untouched. The control chain moves
into an early-return helper, which keeps the extra branch from pushing the
file past its no-nested-ternary budget and drops that count from 5 to 3.
* refactor(ui): move the virtual key create and edit forms off tremor
Swaps the tremor primitives in the create-key modal, the key edit view and
their two shared field components for the in-repo shadcn layer: Accordion
becomes Collapsible, Grid/Col become grid divs, Text/Title become real
paragraphs and headings, and TextInput becomes the shadcn Input. antd stays
where it already was, so the antd Input keeps rendering the textareas and
hidden fields under the AntdInput alias.
Two behavioural notes. The key edit view's Save Changes button keeps saving
because it carries an explicit type="submit"; Base UI's button otherwise
defaults to type="button". Its Cancel button now really is type="button",
where the tremor one had no type at all and so submitted the form on top of
calling onCancel, and a test pins that.
Base UI's Collapsible panel unmounts while closed exactly like the headless
Disclosure panel tremor wrapped, so the create-key tests now open Optional
Settings before querying inside it instead of relying on a tremor mock that
flattened every accordion.
* refactor(ui): hoist the create-key collapsible header classes and keep optional settings a heading
Names the repeated Collapsible trigger and chevron class strings the way the
cost tracking conversion does, since nine copies of each lived in this one
file, wraps the Optional Settings trigger in an h3 so the section keeps a real
heading next to Key Ownership and Key Details, and drops the placeholder=""
that only ever existed to suppress tremor's default hint.
* feat(vector_stores): add Valkey as a managed vector store provider
Adds a valkey provider for managed vector stores, searchable via the
valkey-search module over RESP. Introduces BaseDirectVectorStoreConfig
for datastores that execute searches directly instead of building an
HTTP request, and refactors the valkey semantic cache to share the new
connection URL helper. Registered in the provider enum, router params,
proxy config registry, Admin UI Add Vector Store modal, and provider
endpoint support matrix.
* fix(vector_stores): join list queries and bound valkey socket timeouts
Review feedback: multi-string queries are now space-joined like every
other embedding-based provider instead of dropping all but the first,
and the request timeout is threaded through the direct vector store
interface into bounded socket_connect_timeout / socket_timeout values
on both redis clients so an unreachable Valkey host cannot pin proxy
workers until the OS TCP timeout.
* chore(ui): regenerate schema.d.ts for valkey vector store fields
* docs(ui): make the Valkey vector store setup note and field tooltips explicit
* feat(ui): pick the Valkey embedding model from the proxy's models like Milvus
* fix(ui): number the setup steps in the vector store provider alerts
An internal user who administers an organization saw an empty
Organization Usage dashboard and had to be promoted to proxy admin to
see any of it.
Two independent gates were closed on them. The route layer rejected
GET /organization/daily/activity with 401 before the handler ran, since
the route belonged to no list a non-proxy-admin can reach, and the
handler's own org-admin scoping was therefore dead code. In the
dashboard, viewOrganizationUsage was granted by session role alone, and
an org admin's session role is internal_user, so the Organization Usage
option never rendered and its data fetch stayed disabled.
The route now sits in self_managed_routes, where the handler restricts
results to organizations the caller is ORG_ADMIN of and 403s on any
other org, and viewOrganizationUsage joins the existing per-capability
org-admin allowance that already covers viewDeletedTeams.
A caller who administers no organization resolves to an empty id list
rather than to None, so the organization-alias lookup is scoped by that
same list instead of reading the whole table.
The Usage page falls back to the global view when org-admin membership
is revoked while it is open, so the selector never keeps a value it no
longer offers.
A model with a long time-to-first-token leaves the proxy's response completely
idle, so any hop with an idle read timeout (AWS ALB and nginx both default to
60s) drops a connection that is perfectly healthy and would have delivered its
tokens shortly after.
The keepalive engines LiteLLM already ships wrap the response object, so they
fill a gap once the upstream has answered and then gone quiet. They cannot fill
the gap before it answers at all, and that is where the whole wait is spent:
measured against api.openai.com/v1/chat/completions with gpt-5.6 at
reasoning_effort high, the response headers and the first body byte both arrive
at 37.90s. Nothing has entered the ASGI response phase by then.
The upstream call is now raced against the keepalive interval, and when it
loses, the SSE response is opened immediately and ": ping" comments, which every
conformant SSE client ignores, fill the wire until the real response is ready to
be replayed onto it. One seam per funnel: base_process_llm_request covers every
native route, create_pass_through_route covers every passthrough route.
Committing the status line that early is the cost. A failure discovered after
the first ping reaches the client as an SSE error frame under a 200 rather than
as an HTTP error status, and LiteLLM's own x-litellm-* response headers are not
yet known. keepalive_ping_has_fired already documents the same trade-off for the
existing engines. Both are why this stays off until an operator sets
litellm_settings.sse_keepalive_ping_interval_seconds.
Separately, the passthrough relay reached neither engine even for mid-stream
gaps, which is the shape of #32491 and #24929, so the relayed bytes get the same
treatment, gated on the upstream declaring text/event-stream and only emitted
between complete frames so a binary transport (AWS event streams on /bedrock)
and a stall halfway through a frame are both left alone.
Fixes#34819
On /v1/messages, AnthropicStreamWrapper synthesizes the content_block_start for
the first content block, queues it, then hits a bare `continue` when that same
upstream chunk's translated delta is empty. The queue is only drained at the top
of the next __next__ / __anext__, so the queued content_block_start waits for a
further upstream chunk to arrive.
An empty delta on the opening chunk is the normal tool-call shape: Bedrock
Converse's contentBlockStart carries the tool id and name with no arguments, and
OpenAI-format streams send arguments: "" on the chunk that names the function.
So a client learns a tool call started one upstream event late, and when the
provider delivers argument fragments as a trailing burst it sees nothing at all
after message_start for the whole generation.
Flush the queued event before continuing, in both the sync and async paths. The
sibling block-transition path already returns from the queue, so only the
first-block-open case changed.
* refactor(ui): migrate the MCP per-user env vars modal to react-hook-form and shadcn
Moves UserEnvVarsModal off the antd Form store onto react-hook-form with a
zod schema built from the server's declared per-user variables, and swaps
antd Input.Password for the shared PasswordInput.
The submit payload is unchanged: every declared variable is still sent as a
key, trimmed, with an untouched field sending an empty string. antd reset
the store from the modal's afterOpenChange; the migrated form reproduces
that by remounting on the same callback, so reopening still starts blank.
Adds UserEnvVarsModal.test.tsx, which was written against the antd original
and proven green before any production change, then re-run unedited against
the migration. Two further cases cover the reveal toggle, which antd
provided through visibilityToggle.
* refactor(ui): migrate the MCP toolset create and edit form to react-hook-form and shadcn
Moves the toolset name and description fields off the antd Form store onto
react-hook-form with a zod schema, and takes the surrounding panel onto
semantic colour tokens so the tab renders in dark mode. The purple selected
tool styling keeps its hue and gains dark variants rather than flattening
to neutral.
Payload is unchanged: create still sends toolset_name, description and
tools, an untouched description is still the empty string rather than
undefined, and the tool selection is still held outside the form. The antd
form carried no onFinish and its buttons sit outside the form element, so
the migrated form keeps submit on the footer button and neutralises its own
submit rather than introducing Enter to save.
Adds MCPToolsetsTab.test.tsx, proven green against the antd original before
any production change and re-run unedited afterwards.
* refactor(ui): migrate the MCP tool arguments form to react-hook-form and shadcn
Moves the schema-driven tool argument form off the antd Form store onto
react-hook-form. Validation moves to an explicit resolver that reproduces
antd's rules field by field, including the per-field required message and
the JSON object and array messages, and the same resolver is reused by
getSubmitValues so the imperative path and the rendered errors cannot
disagree.
getSubmitValues still rejects with a plain object carrying errorFields
rather than an Error. ChatUI branches on `err instanceof Error` to choose
its toast, so rejecting with an Error would have silently changed the
message the user sees. That is pinned by a test proven green against the
antd original with a Form.Item liveness gate, and proven red when the
rejection is switched to an Error.
Enum and boolean fields keep the antd Select, whose allowClear has no
shadcn equivalent; dropping it would remove the only way to unset an
optional enum. Everything else moves to the shadcn Input and Textarea and
onto semantic colour tokens.
Adds MCPToolArgumentsForm.test.tsx covering the string, integer, number,
boolean, object, array, nested-params and string-schema paths, written
against the antd original and re-run unedited afterwards.
* refactor(ui): drop the decorative antd Form.Item from the MCP connect guide
The connect guide rendered a single antd Form.Item with no field name and no
Form ancestor, so it registered nothing and carried no payload; it was only
supplying bottom margin. It becomes a div with the same margin class, which
removes the file's last antd Form dependency.
Also takes the guide onto semantic colour tokens so it renders in dark mode.
The blue and green callouts keep their hue and gain dark variants rather
than flattening to neutral, since the colour carries meaning there.
* chore(ui): ratchet the MCP tool arguments form lint suppressions
The react-hook-form migration removed four of the five nested ternaries
in MCPToolArgumentsForm, so lower the grandfathered count to match and
hoist the one inline object literal the budget rule flags.
* test(ui): classify the MCP modal batteries as integration tests
Both render a real component tree down to the form controls and stub only
the network boundary, which is the repo's definition of an integration
test rather than a unit test. The tool arguments battery renders a single
module in milliseconds, so it stays unsuffixed.
* refactor(ui): migrate agent forms to react-hook-form and shadcn
Move the agent create wizard and the agent detail editor off antd Form onto
react-hook-form with shadcn primitives. The two parents share three children
(agent_form_fields, dynamic_agent_form_fields, cost_config_fields), so the whole
form graph migrates in one commit.
The submit payload is unchanged. antd validates and submits only fields that are
currently mounted, and its Collapse panels mount lazily on first open and then
stay mounted, so a payload depends on which panels the user ever expanded.
react-hook-form keeps every registered value instead, so the panels track their
own mounted set and the detail editor filters the never-opened panels back out
before building the request. shouldUnregister stays off, since it drops values
for collapsed panels rather than merely excluding them from submit.
Tags, examples and forwarded header names move from antd tags-mode selects to a
combobox in the kit. Base UI clears the combobox input on blur before the blur
handler runs, so the pending text is committed from the input-clear reason,
which is what antd did when the field lost focus.
The agent type picker stays on antd Select: its popup content is not part of the
form graph, and the Base UI popup opens a macrotask later, which the existing
unit test cannot observe.
* refactor(ui): use the shared PasswordInput in the agent forms
* refactor(ui): drop the narration comments from the agent wizard
Moves the guardrail form graph off antd Form onto react-hook-form with the
shadcn field primitives. The graph migrates atomically: add_guardrail_form and
guardrail_info own the form instances, and guardrail_provider_fields,
guardrail_optional_params and LLMJudgeFields are field groups rendered inside
them, so an antd parent could not host a react-hook-form child either way.
The submit payload is unchanged. Two characterization suites, 25 cases, pin it:
each case was written against the antd original, proven green there, and passes
unedited against the migration.
Behaviour worth calling out. Nested provider fields are keyed with ":" rather
than "." so they stay flat keys the way antd stored them, since a dotted name
is a lodash path in react-hook-form and would have started shipping a nested
object. antd InputNumber clears to null and clamps on blur where a native
number input does neither, so the judge criteria weights reproduce that. The
guardrail_info submit handler is read through a ref at validation-resolution
time, matching how antd re-read onFinish, so a submit fired by the same click
that changed state still sees that state.
Two antd behaviours are preserved rather than fixed, both worth their own
follow-up: deselecting every mode blocks Next instead of falling back to the
seeded default, and a required provider-specific field is never enforced at
create time. One is fixed and disclosed: a failed validation now names the
problem instead of rendering "[object Object]", and the guardrail name label is
associated with its control, which it was not before.
MultiSelect takes an optional id so the label can point at the control.
SkipMessageSelect was duplicated verbatim in both parents and now lives in the
shared field module.
The edit project modal had seven payload tests and none of them cleared a
required field, so rewiring its submit button to send raw form values with
validation skipped left all seven green. The validation layer was untested
on that path while looking well covered.
Adds the missing case: clearing the project name blocks the save and shows
the error. It passes on current code, so the behaviour was always correct,
and it fails under the validation-bypass mutation, so it has teeth.
* refactor(ui): migrate the SCIM and Hashicorp Vault forms to react-hook-form and shadcn
Both forms move off antd Form onto react-hook-form plus the shadcn kit, with
neutral greys on semantic tokens and coloured callouts keeping their hue behind
a dark variant, so both are dark-mode ready.
Neither file had a test, so each one gained a characterization test written
against the antd original and proven green there before any source changed. The
same files pass unedited after the migration.
Two payload details the migration has to reproduce rather than tidy up. antd
onFinish emits a mounted but never-set field as a key holding undefined, and the
vault handler turns each of those into an empty string to clear it server-side,
so every rendered vault field is seeded to an empty string rather than left
absent. And the vault form still refuses to seed or send a blank sensitive
field, so a stored secret stays write-once.
SCIM keeps its Enter-to-submit path: its footer button was a Tremor Button
carrying an explicit type=submit, which Tremor forwards, so the form could
already be submitted from the keyboard.
* refactor(ui): migrate SSO, SCIM and vault forms to react-hook-form and shadcn
Moves the SSO settings form graph, the SCIM token form and the Hashicorp
Vault config form off antd Form onto react-hook-form plus the shadcn
FormField primitives, keeping today's submit payloads byte for byte.
The SSO graph migrates atomically because an antd Form.Item parent cannot
host a react-hook-form child. BaseSSOSettingsForm now owns the shared
schema, the field components and a mounted-field picker that reproduces
what antd's onFinish actually sent: rc-field-form validates only mounted
entities, so hidden provider and mapping fields never reached the wire.
submitMountedSSOValues keeps that behaviour explicit instead of leaving it
to which fields happen to be rendered.
EditSSOSettingsModal seeds through an explicit mapper rather than
spreading the server record, so a field the form does not declare cannot
leak into an update. The vault modal keeps its two distinct behaviours for
blank inputs, clearing non-sensitive fields with an empty string and
omitting blank secrets so a stored credential survives a save.
* fix(ui): render SSO select labels and guard seeding completeness
The migrated Select triggers rendered the raw stored value rather than the
option label, so an untouched Default Role showed "internal_user" and a
chosen provider showed "okta". Base UI resolves a label only through a
Value function child, so both selects now format through the same option
list that builds their items.
Adds three characterization cases the earlier suite did not reach: an
empty required provider credential blocks the submit and names the field,
reopening the modal against a different stored config replaces every
seeded value rather than merging, and every field the provider forms can
mount survives the seeding mapper. The last one fails by name when a key
is dropped from that mapper, which is the class of defect an explicit
allowlist invites.
* test(ui): cover the edit SSO modal against its real form tree
The existing modal test stubs BaseSSOSettingsForm out, so no field ever
registers and validation passes trivially. Rewiring the Save button to
call the submit handler with raw form values, skipping both validation and
the mounted-field filter, left all 112 tests green.
Adds an integration test that renders the real modal, the real form body
and the real antd shell, stubbing only the two data hooks. Clearing a
required credential now blocks the save and names the field, and a valid
save asserts the exact payload. The bypass mutation fails both cases, and
dropping only the mounted-field filter fails the payload one.
* refactor(ui): move the search tool, tag and vector store views off tremor
Swaps the tremor Button, TextInput, Text, Title, Card, Badge, Accordion and
TabGroup usages in the search tools, tag management and vector store views for
the shadcn primitives, following the tremor conversion cookbook. Both tab panels
in the vector store info view carry keepMounted so the tester's state survives
switching to Details and back, and a test pins that contract.
tremor's Button renders a bare button element with no type, so inside the antd
Forms here the Test Connection button in the create search tool modal and the
Cancel buttons in the tag editor and the vector store form were implicit submit
buttons. The shadcn Button defaults to type="button", so they can no longer
submit, and every button that is meant to submit now carries an explicit
type="submit". Clicking Test Connection and Cancel against a live proxy on the
merge base already only ran the connection test and only cancelled, so this
closes a latent trap rather than changing what the pages do.
Decrements the seven no-restricted-imports suppression counts these files no
longer need, leaving the antd half of each entry in place for the antd pass.
* refactor(ui): use the line tab strip in the vector store detail view
The detail view's tabs kept the default pill TabsList, so it no longer matched
the underline strip tremor rendered before the swap or the one the vector store
list view already uses.
* test(ui): pin the tag and vector store form save and cancel buttons
Cancel in the tag editor used to submit the form and save the tag because the
tremor button carried no type; nothing in the suite failed if it started doing
that again. Each form now has a pair of cases: Save Changes and Create still
submit, and Cancel leaves the record alone.
* fix(ui): keep the reveal toggle on the search tool API key
tremor's TextInput drew its own show/hide button whenever the type was
password, and the shadcn Input is a plain native input, so the straight
prop pass-through silently deleted that affordance from the create search
tool form's API key field.
Puts it on antd's Input.Password instead, which is what the sibling edit
form in the same directory (SearchTools.tsx) already uses for the very
same field, so the reveal survives and the two forms behave the same.
The file already imports antd, so this adds no import and no suppression.
* test(ui): pin the antd submit payloads for the key create and edit forms
Characterization only, no source change. Both suites are green against the
current antd components, so they can gate the react-hook-form migration that
follows without being edited.
key_edit_view had no exact-payload assertion, only objectContaining, so nothing
caught a form that started sending server-only key fields. The new case asserts
the whole object.
create_key_button's existing suite runs against a hand-rolled antd fake and
stubs out KeyLifecycleSettings and RateLimitTypeFormItem, so neither the real
store nor those two controls were covered. The new file drives the real antd
form and pins the network payload instead.
* refactor(ui): move the shared key form controls off antd onto shadcn
KeyLifecycleSettings and RateLimitTypeFormItem each owned an antd Form.Item and
took the parent's FormInstance as a prop, so neither could be hosted by anything
but an antd form. That is what made the key create and edit forms one
inseparable migration unit.
Both are now presentational: they take value and onChange and let the parent own
the binding, so an antd Form.Item and a react-hook-form FormField can host them
equally. The two parents keep their antd forms for now and pass the binding down
unchanged, which is why every existing payload assertion still holds.
Controls are shadcn Select, Input, Switch and Checkbox on semantic colour
tokens, so both are dark-mode ready. The rotation notice keeps its blue hue and
gains a dark variant rather than flattening to a neutral.
The form prop the two components took was already inert: antd dispatches its own
store update before calling the child's onChange, so setFieldValue was writing a
value the store had just been given.
KeyLifecycleSettings.test.tsx keeps every assertion; only the harness moves the
duration binding up into a Form.Item, and one case disables user-event's
pointer-events check because Base UI leaves a reopened select popup inert under
jsdom, reproduced on a bare shadcn Select with none of this code involved.
* test(ui): pin the role-gated key fields and tidy the new assertions
Adds the case that proves policies and prompts leave the payload entirely for a
role that cannot see them, which a react-hook-form port would otherwise start
sending from defaultValues. Green against antd like the rest.
Also hoists the two large expected payloads into named constants and drops two
unused exports, so the lane adds no new lint-budget pressure.
* fix(ui): give the key expiry input and Never Expire checkbox separate labels
The expiry label carried htmlFor for the duration input while also wrapping
the Never Expire checkbox and its own label, so the two controls shared one
ambiguous association. Splitting the row into a plain container with a label
per control makes each name resolve to the control it describes.
Also drops the prop and test comments added in this branch, which the
repository comment policy does not allow.
* test(ui): pin the create-form expiry binding to the generate payload
create_key_button coalesces a missing or blank duration to null before it
calls keyCreateCall, so the key is present in the payload whether or not the
control is bound to the form. Every existing case stayed green with the
Form.Item removed, which left the binding uncovered.
The new case opens Key Lifecycle, types an expiry, and asserts it arrives as
that value. Proven red with the Form.Item removed and green with it restored.
* refactor(ui): migrate the model settings and credential rotation modals to react-hook-form
Both modals owned a self-contained antd FormInstance with no shared form
children, so each migrates on its own without touching the add_model graph.
ModelSettingsModal keeps its antd Modal shell, footer buttons and Skeleton
placeholder. The single store_model_in_db field becomes a shadcn Switch inside
a FormField, the antd Form.Item tooltip stays a hover tooltip rather than
becoming always-visible description text, and the remount-on-new-config
behaviour that the antd `key` provided is now RHF's `values` option.
UpdateModelCredentialsModal keeps the antd Modal and warning Alert. The
Input.Password becomes an InputGroup with an Eye/EyeOff reveal toggle so the
reveal affordance survives, and the required rule ports to the same message.
Both submit paths stay exactly as they were: Enter still submits here because
the antd Form had onFinish and a real submit button, while the settings modal
keeps submitting only from its footer button.
* refactor(ui): reuse the shared PasswordInput in the credential rotation modal
* refactor(ui): move the internal user create, edit and detail views off tremor
The tremor SelectItem rows nested inside the antd role Select become
antd Select.Option so the antd control keeps driving the form; the antd
removal is left to the antd pass. The user detail tabs move from a
numeric index to overview/details slugs (the public initialTab number
prop is unchanged and mapped at the boundary), every panel is
keepMounted, and a test pins that contract. The per-team remove button
keeps tremor's light red look as a ghost icon button rather than a
solid destructive one. Prunes the tremor no-restricted-imports
suppressions these files no longer need.
* refactor(ui): keep the user detail tab strip on the line variant
The tremor TabList defaulted to the underline strip, so the shadcn
TabsList needs variant="line" to keep that look instead of the filled
segmented pill; the triggers pick up the same active classes the users
page strip right above already uses. Also drops the vestigial empty
placeholder on the embedded create user email field, which only existed
to suppress tremor's built-in "Type..." hint.
* feat(complexity_router): custom classifier plugins via classifier_type 'plugin'
Adds a third classification mode where an operator-supplied hook decides the
tier instead of the heuristic scorer or the LLM classifier. The hook implements
an async classify(context) returning a tier name (built-in value, tier_labels
label, or tier_definitions name) or None to decline; failures, timeouts, and
unknown tiers fall back exactly like a failed LLM classifier. The context
carries the request messages and metadata, including caller identity, so a
plugin can route by team, spend, or any business rule.
The plugin resolves from a dotted path at proxy startup with a load-time check
that classify is a coroutine function, and is closed off over HTTP like the
routing plugins list. Routing decisions record the new classifier_plugin cause.
tier_definitions now accepts classifier_type 'plugin' alongside 'llm'.
* fix(proxy): resolve plugin dotted paths in _delete_deployment before hashing ids
The db-sync reconcile re-reads the raw config and hashes litellm_params to
compute which ids the config wants served, but the router's ids were hashed
from the resolved params where plugin dotted paths are live instances. The
mismatched ids made the reconcile evict every plugin-bearing auto-router one
sync after startup, on any proxy with a database connected. This also affected
the existing routing plugins list, not just the new classifier plugin.
Resolving the plugins in _delete_deployment the same way load_config does makes
both sides hash the same canonical form. A plugin module broken on disk at
reconcile time skips cleanup instead of evicting valid deployments, matching
how a get_config failure is handled
* fix(complexity_router): treat non-string plugin verdicts as declines, centralize the empty-mapping sentinel
A hook returning a non-string raised inside resolve_classified_tier outside the
plugin exception boundary, failing the request instead of falling back. Also
moves the read-only empty mapping to constants.py per repo convention and moves
the classifier plugin product docs out of the package README for the docs repo
* refactor(complexity_router): rename the plugin classifier mode to classifier_type 'custom'
The mode value now names the operator's intent while classifier_plugin keeps
naming the mechanism; routing decisions keep the classifier_plugin cause
* refactor(proxy): pin plugin-bearing deployment ids from the raw params instead of resolving in the reconcile
Replaces the previous approach of re-running plugin resolution inside
_delete_deployment, which imported operator modules on every reconcile cycle
and skipped the whole cleanup pass when any one module was broken on disk.
load_config now stamps model_info.id from the raw litellm_params before
resolution swaps dotted paths for live instances, so the reconcile's raw-config
hash matches by construction and needs no resolution at all: a broken module
cannot stall cleanup for unrelated models, and any future param-transforming
resolution is covered by the same pin. _generate_model_id becomes a staticmethod
so the pin can run before the Router exists; its statically dead non-string key
branches are removed. Also documents candidate_models as an informational
snapshot for classifier plugins, unlike the narrowing surface RoutingPlugin
filters
* fix(router): restore _generate_model_id key handling, align classifier context with the routing-plugin pattern
The staticmethod conversion accidentally dropped the non-string-key branches
from _generate_model_id, a silent hash change for any params with non-string
keys; they are restored verbatim. The classifier plugin context now follows
the Router-level routing-plugin recipe exactly: structured messages come from
resolve_structured_messages over the raw messages, and the metadata key comes
from the shared get_metadata_variable_name_from_kwargs helper, which also
replaces the duplicated inline sniff in _pick_model_for_tier. This removes the
raw-or-resolved fallback where a plugin could silently receive resolved
messages when a call site forgot to pass the raw ones
* refactor(router): make generate_model_id public, guard classifier context construction
Two modules legitimately hash deployment ids with the same helper now (Router
and the proxy's config-load pin), so the private name was lying about its
audience and the cross-module call needed a pyright suppression; renaming it
public restores the static safety net. The classifier plugin's RoutingContext
construction moves inside the failure boundary, matching the LLM path where
litellm-side prompt building also falls back rather than failing the request,
and a prompt-only call with no message list is now covered by a test
* refactor(ui): move the MCP server forms and detail tabs off tremor
Swaps the tremor Button, TextInput, Title, Text and Tab primitives in the six
MCP server components for the shadcn layer, and leaves the antd Modals, Forms
and Selects alone for the antd pass. In the two files that mix both input
libraries, antd's Input is imported as AntdInput so the shadcn Input keeps its
canonical name.
Two behaviours needed care. Base UI's Button forces type="button", so the
create button in CreateMCPServer now carries an explicit type="submit"; Cancel
and the OAuth authorize button stay non submitting, which also drops the
accidental implicit submit they inherited from tremor. Every TabsContent gets
keepMounted, because Base UI unmounts inactive panels while tremor only hid
them, and a save started from the Cost Configuration tab reads fields that live
in the Server Configuration panel. mcp_server_edit.test.tsx gains a regression
test for that: drop keepMounted and the pending edit never reaches the update
payload.
One affordance is gone: password fields no longer draw tremor's built in reveal
toggle, since the shadcn Input is a plain native input.
Prunes the six no-restricted-imports suppressions these files no longer need.
* fix(ui): keep the reveal toggle on the MCP secret fields
tremor's TextInput drew its own show/hide button whenever the type was
password, and the shadcn Input is a plain native input, so the straight
prop pass-through silently deleted that affordance from five fields: the
create modal's authentication value and the OAuth client id and secret in
both the M2M and the interactive flow.
Puts them on antd's Input.Password instead, which is what every sibling
secret field in this directory already uses (TokenExchangeFormFields,
IdJagFormFields, AwsSigV4Fields and the edit form), so the reveal survives
and the five fields now match their neighbours instead of behaving
differently inside the same form. Both files already import antd, so this
adds no import and no suppression.
* test(ui): pin the connect tab mount contract
mcp_connect's per-card "limit tools to specific MCP servers" toggle lives
in panel local state that feeds the rendered header block, so the panels
have to stay mounted across a tab switch. Base UI unmounts an inactive
panel unless keepMounted is set, and unlike the edit form there was no
test holding that down.
Toggles the header on from the LiteLLM Proxy panel, switches to Cursor and
back, and asserts both the switch and the x-mcp-servers line in the curl
example survived. Dropping keepMounted from that panel fails it.
* fix(ui): keep the MCP tab strips underlined instead of segmented
A bare tremor TabList is variant="line", so the connect strip and the
server settings strip both drew an underlined tab on a full width
divider. Converting them bare turned each into a filled segmented
control, because the shadcn TabsList defaults to the pill.
Both strips now use variant="line" with the divider recipe, and the
connect strip gets back the grey rounded box tremor drew around its four
tabs.
* refactor(ui): migrate the vector store creation form to shadcn
CreateVectorStore and S3VectorsConfig were the last antd Form.Item users on
the vector stores page. Both are now built from the shared Field primitives
and shadcn controls, so the page picks up the design tokens and dark mode.
CreateVectorStore's antd Form was inert: no Form.Item carried a name, there
was no onFinish, and the submit button sat outside the form element, so the
form store never held anything. Form.useForm is dropped rather than replaced
with react-hook-form, and the existing imperative validation is unchanged.
S3VectorsConfig's four Form.Item wrappers had no name either, so its inputs
were already prop-controlled and decoupled from the parent store. The
embedding model picker keeps its typeahead by moving to Combobox.
The submit payload is unchanged. A new characterization suite pins it: it was
written against the antd originals, proved green there first, and passes
unedited against the migration.
* test(ui): cover the S3 embedding model combobox end to end
The migration moved this control from an antd Select with showSearch to a
Combobox, and nothing exercised it: the suite pinned the payload but never
loaded the option list, filtered it, or selected from it.
The case drives the whole interaction. It stubs three models, one of which is
a chat model, opens the list and asserts the chat model is absent, types to
filter, selects the remaining embedding model and asserts it arrives in the
providerParams argument.
Proved green against the antd originals of both files first, then unedited
against the migration. It queries the control by role rather than by label,
because the antd original rendered a label with no control associated to it,
which the migration fixes.
Moves six antd Form graphs onto react-hook-form + shadcn: the pass-through
create and edit forms with their two shared sections, the project create and
edit modals, and the access group edit modal.
Each form graph moved atomically. An antd Form.Item parent cannot host a
react-hook-form child, so a shared section that renders a Form.Item has to move
with every parent that mounts it or the field silently stops reaching the
payload. PassThroughSecuritySection holds Form.Item name="auth" and is imported
by both pass-through parents, so all four files move together.
Submit payloads are unchanged and pinned by characterization tests written
against the antd originals first. That includes the parts that look like bugs:
add_pass_through still sends timeout and cost_per_request as strings while
pass_through_info sends them as numbers, and both edit modals still omit fields
whose antd Form.Item never mounted.
One behaviour does change. Enter in the add pass-through form ran Cancel and
discarded the filled form, because HTML implicit submission activates the first
submit button in tree order and Tremor renders buttons with no type attribute,
making the footer Cancel that button. Cancel is now type="button" and Enter
submits.
* refactor(ui): migrate the caching, cost tracking, alerting and user detail forms to react-hook-form and shadcn
Moves the last of this lane's antd Form usage onto react-hook-form plus the
shadcn field kit, and takes the neutral greys in the files touched onto
semantic tokens so these screens are dark-mode ready.
Each unit is characterization-first: a payload-pinning test was written and
proven green against the antd original before any production code changed,
then re-run unedited against the migration. Every pre-existing test for these
files still passes without edits.
Two behaviour changes are deliberate and disclosed rather than buried.
Saving cache settings with the Advanced section collapsed used to drop
namespace, ttl, max_connections and the GCP fields from the payload, and to
send ssl as false even when the loaded value was true, because antd reports
only mounted fields at submit while its store keeps the rest. Keeping the
values in form state fixes that, and the payload no longer depends on whether
the section was expanded.
Adding a provider discount used to call the handler twice per click, because
the child submit button sat inside a Form carrying onFinish that called the
same function the button's onClick already called. Removing the leftover
wrapper collapses it to one request. Both calls read the same config and sent
the same body, so this changes request count rather than stored state.
* fix(ui): restore Enter submission on the provider discount form
Removing the antd Form wrapper from cost_tracking_settings also removed
native Enter submission from the Add Provider Discount modal, where the
original carried onFinish and AddProviderForm renders a type="submit"
button. Wrapping the fields in a native form whose onSubmit only calls
preventDefault restores it: implicit submission clicks the default
button, so Enter and a click each produce exactly one request rather
than the two the antd version fired on both paths.
The margin modal is deliberately left as a div. Its antd Form carried no
onFinish and AddMarginForm renders type="button", so Enter was already
inert there and a test now pins that alongside the working click path.
Also swaps five any[] mock forwarders for unknown[] in the user detail
test and asserts the role trigger renders its label rather than a blank
or raw value.
openai/responses/<model> deployments reached OpenAI as the literal model id
responses/<model> on /v1/responses and on /v1/messages (which rides the
Responses API for the openai provider) and 400ed with model_not_found, while
/v1/chat/completions already stripped the prefix. Strip a leading responses/
right after provider resolution so every Responses API entrypoint (HTTP,
websocket, compaction, the /v1/messages adapter) sends the real model id
* refactor(ui): retire the tremor date range picker in favour of the shared advanced picker
UsageDatePicker was the last tremor DateRangePicker surface. Its three call sites in the old usage page and the caching dashboard now render AdvancedDatePicker, which already had the same prop interface, preset list and idle-callback day-boundary adjustment. AdvancedDatePicker drops its own tremor Button and Text for the shadcn Button and a plain paragraph, and it now applies the className prop it already declared so the mb-4 the tag-based usage tab passes keeps landing on the picker root. usage_date_picker.tsx and its calendar-grid test are removed, and the two no-restricted-imports suppressions those files carried are pruned
* fix(ui): let the advanced date picker anchor its panel to the trigger's left edge
The picker's dropdown is 600px wide and right-anchored to a 300px trigger, which was fine while every caller sat at the right edge of its row. The two old usage tabs place it in the left column, so the preset column landed left of the main scroll container and was clipped. AdvancedDatePicker gains an align prop (default right, unchanged for existing callers) and the old usage call sites pass left. The caching dashboard grid gives the picker an auto track instead of a third equal share, so the fixed-width trigger no longer spills past the card at laptop widths
* fix(ui): give the advanced date picker a real focusable trigger
The picker's display was a click-only div, so tabbing through the usage,
old usage, caching, cost optimization and guardrails monitor pages skipped
the date range control entirely and its focus ring classes never fired. It
is now a type="button" element carrying aria-expanded, which restores the
keyboard and screen reader access the tremor picker had. The panel also
reports its anchoring as data-align so the test can assert intent instead
of a Tailwind class
* fix(ui): make date picker relative-range presets keyboard-operable
The presets were non-focusable divs with click handlers, so a keyboard-only admin tabbed past Today / Last 7 days / Last 30 days / MTD / YTD and had to type both dates by hand. They are now buttons carrying aria-pressed.
* feat(ui): plan-mode override tier in the auto-router create and edit forms
The backend plan_mode_min_tier field (#37230) was API-only. Both forms now
carry an Advanced: Plan-Mode Override panel in the shared complexity config
component: a toggle derived from field presence, so on writes the highest
tier that has models and off deletes the key, and a tier select limited to
tiers with models because the backend rejects a floor at an empty tier. The
edit modal manages the key on every save, so clearing it actually clears the
stored config instead of the preserved copy resurrecting it, while unmanaged
keys like plan_mode_patterns still round-trip untouched
* refactor(ui): hoist the eligible plan-mode tier list out of the panel JSX
* refactor(ui): move tierOptions into complexity_router_tiers, the shared tier-utility module
* refactor(ui): migrate login, onboarding and search tool forms to react-hook-form and shadcn
Moves four forms off antd Form and Tremor widgets onto react-hook-form plus
the shadcn kit, and onto semantic colour tokens so the screens are dark-mode
ready. antd Modal and Alert stay as the shells.
The submit payload is unchanged in all four. Each unit is pinned by a
characterization test that was proven green against the antd original before
any production code changed, and the pre-existing test files pass unedited.
Extracts the search tool payload builder, which was duplicated verbatim
between the create and edit forms, into searchToolPayload.ts with unit tests,
and adds a shared PasswordInput so the four reveal toggles antd and Tremor
gave for free are preserved on one component.
* refactor(ui): keep the search tool Test Connection button an implicit submit
Tremor's Button renders no type attribute, so inside a form it defaults to
submit. The Test Connection button therefore fires both its own onClick and
the form's onFinish today, which creates the search tool as a side effect of
testing the connection. shadcn's Button renders type="button", so the naive
swap silently dropped that second path.
Restores parity with an explicit type="submit" and pins it with a test, so
the double submit is recorded rather than quietly changed. Fixing it belongs
in its own change.
Also prunes the two now-stale eslint suppression counts for the migrated
search tool files, scoped to those keys only.
* refactor(ui): announce the login button spinner the way antd did
antd's Button renders its loading indicator as role="img" with aria-label
"loading", so a screen reader announces the request in flight. The shadcn
spinner is a bare svg, which drops that. Labels it on the login button to
match, as already done on the onboarding submit button.
* fix(ui): drop noValidate from the migrated login, onboarding and search-tool forms
antd's Form renders no novalidate attribute and its required rules emit
aria-required rather than the native required attribute, so nothing in
these four forms was ever gated by native constraint validation. The only
type="email" input is disabled and readOnly, which bars it from validation
in every browser. Measured in jsdom and again in Chrome against a live
proxy: form.checkValidity() is true with the fields empty, a native submit
reaches react-hook-form, and zod blocks it with the same messages.
Removing the attribute keeps the rendered form faithful to antd, and keeps
a constraint added later behaving the way antd would have behaved instead
of being silently suppressed.
* fix(ui): accept a null api_key when seeding the search tool edit form
The list endpoint declares api_key as str | None and search_tool_info as
dict | None, and the masking helper returns non-string values untouched, so
a tool stored without an API key comes back as "api_key": null. zod's
optional() accepts undefined and rejects null, so the edit form for any
such tool failed with "expected string, received null" and could never be
submitted. antd carried no schema and forwarded whatever the server sent.
nullish() restores that, and the payload still forwards the null rather
than coercing it to an empty string. The new case seeds both null vectors
and fails without this change.
* chore(ui): drop the narration comments from the search tool forms
These restate the state change or the JSX block directly below them, which
the repo's comment policy rules out, and both files are rewritten by this
change rather than merely touched.
Keeps an explicit empty messages list on the pre-existing path (default
model, provider validation error) instead of dropping the routing
decision and surfacing a misleading tags 401.
