Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
Go agentagenticaichatbotdsh-pluginembeddingsevaluationgenerative-aigolangknowledge-basellmmulti-tenantollamaopenaiquestion-answeringragrerankingsemantic-searchvector-searchwiki
<!--
CredentialResource — per-field "configured / unconfigured / editing" card.
Why this component exists:
The previous UX (PR #990) showed a password input pre-filled with a redacted
placeholder plus a red "Remove this credential" checkbox below it. That
conflated three distinct user intents (preserve / replace / clear) into a
single form field, and bundled credential changes with unrelated config
edits in the same submit. Users could accidentally wipe a working key by
toggling the wrong checkbox.
This component splits credentials out as an independent resource:
- Read-only "configured" badge by default, derived from the `meta` prop.
- "Replace" expands an input + Save/Cancel; commit is an explicit
PUT to the credential subresource (no main-form submit needed).
- "Remove" pops a danger-themed confirmation and on confirm calls DELETE
immediately. No "save form to apply" intermediate state.
`meta` is the source of truth and comes from the parent resource's main
GET response (`<resource>.credentials` on every DTO) — there is no
dedicated GET /credentials endpoint. After a successful save/remove the
component derives the new local state from the save's return value (or,
for remove, by setting the field to unconfigured) and emits 'changed' so
the parent can re-fetch the main resource if it cares about anything
else that depends on credential state.
-->
<template>
<div class="credential-resource">
<div v-for="field in fields" :key="field.key" class="credential-row">
<!--
Per-field label, rendered only when there is more than one credential
(e.g. WeKnora Cloud's api_key + app_secret) so a single-field card
doesn't double up with the parent's outer .form-label. Single-field
consumers (ModelEditorDialog API key, WebSearch provider api_key,
McpService api_key) keep showing the parent label only.
-->
<div v-if="fields.length > 1" class="credential-row-label">{{ field.label }}</div>
<!-- Configured: faux input row (visually identical to t-input) -->
<template v-if="stateOf(field.key) === 'configured'">
<!--
Two possible looks:
a) default — ✓ 已配置 + [更换 | 移除]
b) confirm-remove pending — ⚠ 确认移除?此操作不可撤销 + [取消 | 确认移除]
We use a sub-state instead of a global modal because the modal
forces the user to context-switch to the screen center, then back
to this row to see the result. Inline confirm keeps focus on the
row that's actually changing.
-->
<div
class="credential-faux-input"
:class="{ 'is-confirm-remove': pendingRemove[field.key] }"
:title="pendingRemove[field.key] ? '' : t('credential.configured')"
>
<template v-if="pendingRemove[field.key]">
<t-icon name="error-circle-filled" class="status-icon warn" />
<span class="credential-faux-text danger">{{ t('credential.confirmRemovePrompt') }}</span>
<div class="credential-actions">
<t-button size="small" variant="text" @click="cancelPendingRemove(field.key)">
{{ t('common.cancel') }}
</t-button>
<span class="action-divider"></span>
<t-button size="small" variant="text" theme="danger" :loading="busy[field.key] === 'remove'"
@click="confirmRemove(field)">
{{ t('credential.confirmRemove') }}
</t-button>
</div>
</template>
<template v-else>
<t-icon name="check-circle-filled" class="status-icon success" />
<span class="credential-faux-text">{{ t('credential.configured') }}</span>
<div class="credential-actions">
<t-button size="small" variant="text" @click="enterEdit(field.key)">
{{ t('credential.update') }}
</t-button>
<span class="action-divider"></span>
<t-button size="small" variant="text" theme="danger" @click="requestRemove(field.key)">
{{ t('credential.remove') }}
</t-button>
</div>
</template>
</div>
</template>
<!-- Unconfigured: faux input row with a single "Configure" affordance -->
<template v-else-if="stateOf(field.key) === 'unconfigured'">
<div
class="credential-faux-input is-empty"
:class="{ 'is-just-removed': inlineToast[field.key]?.kind === 'removed' }"
@click="enterEdit(field.key)"
>
<!--
Right after a successful remove we hold the row in place but swap
the icon + placeholder text for a brief success state. After
~2.4s the row fades back to its plain "未配置" prompt. This
keeps feedback anchored to where the user just clicked, instead
of asking them to glance at a global toast somewhere else.
-->
<template v-if="inlineToast[field.key]?.kind === 'removed'">
<t-icon name="check-circle-filled" class="status-icon success" />
<span class="credential-faux-text">{{ t('credential.removedToast') }}</span>
</template>
<template v-else>
<t-icon name="lock-on" class="status-icon muted" />
<span class="credential-faux-text muted">{{ t('credential.unconfigured') }}</span>
<div class="credential-actions">
<t-button size="small" variant="text" theme="primary" @click.stop="enterEdit(field.key)">
{{ t('credential.configure') }}
</t-button>
</div>
</template>
</div>
</template>
<!-- Editing: real input + tiny action row beneath -->
<template v-else>
<div class="credential-edit">
<t-input v-model="drafts[field.key]" type="password"
:placeholder="field.placeholder ?? t('credential.inputPlaceholder')" :autocomplete="'new-password'"
class="credential-edit-input" @enter="onSave(field)">
<template #prefix-icon><t-icon name="lock-on" /></template>
</t-input>
<div class="credential-edit-actions">
<t-button size="small" variant="text" @click="cancelEdit(field.key)">
{{ t('common.cancel') }}
</t-button>
<t-button size="small" theme="primary" :loading="busy[field.key] === 'save'" :disabled="!drafts[field.key]"
@click="onSave(field)">
{{ t('common.save') }}
</t-button>
</div>
</div>
</template>
</div>
</div>
</template>
<script setup lang="ts" generic="K extends string">
import { onBeforeUnmount, reactive, ref, watch } from 'vue'
import { useI18n } from 'vue-i18n'
import { MessagePlugin } from 'tdesign-vue-next'
export interface CredentialFieldDef<K extends string = string> {
key: K
label: string
// Optional connector-specific placeholder shown only when the input is
// visible (e.g. "ntn_xxxx" for Notion). Defaults to a generic "Enter value".
placeholder?: string
}
export interface CredentialResourceApi<K extends string = string> {
// PUT /credentials — body keyed by field name, value is the new secret.
// Returns the updated per-field configured map.
save: (patch: Partial<Record<K, string>>) => Promise<Record<K, { configured: boolean }>>
// DELETE /credentials/:field
remove: (field: K) => Promise<void>
}
interface Props {
fields: CredentialFieldDef<K>[]
api: CredentialResourceApi<K>
// Initial per-field "configured?" map, sourced from the parent resource's
// main GET response. The component reads it on first render and after
// every reset; subsequent state transitions are tracked locally.
meta: Record<K, { configured: boolean }>
}
interface Emits {
// Fires after every successful save or remove so the parent can refresh
// any derived view (e.g. badges that depend on credential state) or just
// reload the main resource to keep `meta` in sync.
(e: 'changed'): void
}
const props = defineProps<Props>()
const emit = defineEmits<Emits>()
const { t } = useI18n()
type State = 'configured' | 'unconfigured' | 'editing'
// Local view state per field. Source of truth is props.meta, but we track
// the editing state and any locally-applied save/remove transitions here so
// the UI doesn't snap back when the parent re-renders before re-fetching.
const states = reactive<Record<string, State>>({})
const drafts = reactive<Record<string, string>>({})
const busy = reactive<Record<string, 'save' | 'remove' | null>>({})
// Per-field "did the user just press Remove" flag. While true, the
// configured row swaps to an inline confirm prompt instead of immediately
// deleting. Cleared on cancel, on the actual DELETE request firing, and
// any time the field leaves the configured state for any other reason.
const pendingRemove = reactive<Record<string, boolean>>({})
// Inline per-field flash message used as an anchored "toast" instead of the
// global MessagePlugin one for actions whose effect happens at this exact
// row (currently: remove). Cleared after a short delay so the row reverts
// to its normal placeholder. Keyed by field.key.
type InlineToastKind = 'removed'
const inlineToast = reactive<Record<string, { kind: InlineToastKind } | null>>({})
const inlineToastTimers: Record<string, ReturnType<typeof setTimeout> | null> = {}
function flashInlineToast(key: string, kind: InlineToastKind, ms = 2400) {
inlineToast[key] = { kind }
if (inlineToastTimers[key]) {
clearTimeout(inlineToastTimers[key]!)
}
inlineToastTimers[key] = setTimeout(() => {
inlineToast[key] = null
inlineToastTimers[key] = null
}, ms)
}
function deriveStatesFromMeta(meta: Record<string, { configured: boolean }>) {
for (const f of props.fields) {
// Preserve in-progress edits across parent re-renders — `meta` describes
// server state, the editing flag is user intent.
if (states[f.key] === 'editing') continue
states[f.key] = meta[f.key]?.configured ? 'configured' : 'unconfigured'
}
}
// Initialize from the first meta snapshot, and re-derive whenever the parent
// passes a new one (after a main-resource refresh). watch with immediate:true
// covers both cases in one place.
watch(
() => props.meta,
(m) => deriveStatesFromMeta(m ?? ({} as Record<K, { configured: boolean }>)),
{ immediate: true, deep: true },
)
// If the parent swaps the api (e.g. user opens a different resource), drop
// transient state. props.meta will follow and re-init via the watch above.
watch(() => props.api, () => {
for (const k of Object.keys(states)) delete states[k]
for (const k of Object.keys(drafts)) delete drafts[k]
for (const k of Object.keys(pendingRemove)) delete pendingRemove[k]
})
function stateOf(key: string): State {
return states[key] ?? 'unconfigured'
}
function enterEdit(key: string) {
drafts[key] = ''
states[key] = 'editing'
// If the user was mid-confirm and changed their mind ("update" instead
// of "remove"), drop the pending flag so we don't bounce back into the
// confirm UI when they cancel the edit.
pendingRemove[key] = false
}
// Cancel returns directly to whatever the parent told us via props.meta —
// no async re-fetch needed, and no risk of staying stuck in 'editing'
// because the previous implementation's refresh was a no-op when state
// was already 'editing'.
function cancelEdit(key: string) {
drafts[key] = ''
states[key] = props.meta?.[key as K]?.configured ? 'configured' : 'unconfigured'
}
async function onSave(field: CredentialFieldDef) {
const value = drafts[field.key]
if (!value) return
busy[field.key] = 'save'
try {
// Apply the save's returned metadata locally so the card flips to
// 'configured' immediately. Skip the editing-preserve guard since this
// particular field just finished editing.
const updated = await props.api.save({ [field.key]: value } as Partial<Record<K, string>>)
for (const f of props.fields) {
if (f.key === field.key) continue
if (states[f.key] === 'editing') continue
states[f.key] = updated[f.key as K]?.configured ? 'configured' : 'unconfigured'
}
states[field.key] = updated[field.key as K]?.configured ? 'configured' : 'unconfigured'
drafts[field.key] = ''
MessagePlugin.success(t('credential.savedToast'))
emit('changed')
} catch (err: any) {
MessagePlugin.error(err?.message || t('credential.saveFailed'))
} finally {
busy[field.key] = null
}
}
// Two-step remove. First click flips the row to a confirm state; second
// click actually fires DELETE. We deliberately do NOT use a global modal
// here — the row is also the place where the result will appear, and a
// modal forces an unnecessary screen-center detour. The danger-themed
// "确认移除" button on a tinted-warning row gives the same protection
// against fat-fingered destructive clicks without that detour.
//
// Errors still surface via global MessagePlugin so the user can't miss them.
function requestRemove(key: string) {
pendingRemove[key] = true
}
function cancelPendingRemove(key: string) {
pendingRemove[key] = false
}
async function confirmRemove(field: CredentialFieldDef) {
busy[field.key] = 'remove'
try {
await props.api.remove(field.key as K)
states[field.key] = 'unconfigured'
pendingRemove[field.key] = false
flashInlineToast(field.key, 'removed')
emit('changed')
} catch (err: any) {
MessagePlugin.error(err?.message || t('credential.removeFailed'))
} finally {
busy[field.key] = null
}
}
// Cancel pending inline-toast timers when the component unmounts so we
// don't write to a torn-down reactive object after navigation.
onBeforeUnmount(() => {
for (const k of Object.keys(inlineToastTimers)) {
if (inlineToastTimers[k]) {
clearTimeout(inlineToastTimers[k]!)
inlineToastTimers[k] = null
}
}
})
</script>
<style scoped lang="less">
.credential-resource {
display: flex;
flex-direction: column;
gap: 8px;
}
.credential-row {
/* Empty wrapper now — each state owns its own outer styling so the
"configured" and "unconfigured" rows can mimic a t-input exactly,
and the editing state can defer the chrome to t-input itself. */
}
.credential-row-label {
font-size: var(--app-text-md);
font-weight: 500;
color: var(--td-text-color-primary);
margin-bottom: 4px;
line-height: 1.4;
}
/*
Faux-input row: visually identical to a TDesign default t-input
- 32px tall
- same border (--td-component-border) + radius (6px) + bg
- 0 12px horizontal padding
This makes the credential card stop looking like "a card inside a card"
when it sits between Base URL and 自定义请求头 — it reads as a normal
field, just one that doesn't accept typed input.
*/
.credential-faux-input {
display: flex;
align-items: center;
gap: 8px;
height: 32px;
padding: 0 4px 0 12px;
background: var(--td-bg-color-container);
border: 1px solid var(--td-component-border);
border-radius: var(--app-radius-sm);
font-size: var(--app-text-md);
transition: border-color var(--app-motion-fast) ease, background-color var(--app-motion-fast) ease;
&:hover {
border-color: var(--td-brand-color-hover);
}
&.is-empty {
cursor: pointer;
background: var(--td-bg-color-container);
&:hover {
background: var(--td-bg-color-container-hover);
}
}
/*
Just-removed flash state: brief tinted background + brand-color border
so the row gives the user clear, anchored confirmation that the remove
actually happened — no need for them to find a corner toast. Auto-fades
after ~2.4s back to plain "未配置" via the inlineToast timer.
*/
&.is-just-removed {
background: var(--td-success-color-light);
border-color: var(--td-success-color-focus);
animation: credential-toast-flash 0.25s ease both;
cursor: default;
&:hover {
background: var(--td-success-color-light);
border-color: var(--td-success-color-focus);
}
}
/*
Confirm-remove state: warning-tinted background + danger-color border,
making it obvious that the row is in a destructive-action standoff. The
button group changes to [Cancel | Confirm-danger]; user has to make a
deliberate second click to actually delete.
*/
&.is-confirm-remove {
background: var(--td-error-color-light);
border-color: var(--td-error-color-focus);
animation: credential-confirm-flash 0.2s ease both;
&:hover {
border-color: var(--td-error-color-focus);
}
}
}
@keyframes credential-toast-flash {
from {
background: var(--td-bg-color-container);
border-color: var(--td-component-border);
}
to {
background: var(--td-success-color-light);
border-color: var(--td-success-color-focus);
}
}
@keyframes credential-confirm-flash {
from {
background: var(--td-bg-color-container);
border-color: var(--td-component-border);
}
to {
background: var(--td-error-color-light);
border-color: var(--td-error-color-focus);
}
}
.credential-faux-text {
flex: 1;
min-width: 0;
color: var(--td-text-color-primary);
overflow: hidden;
text-overflow: ellipsis;
white-space: nowrap;
&.muted {
color: var(--td-text-color-placeholder);
}
&.danger {
color: var(--td-error-color);
font-weight: 500;
}
}
.status-icon {
flex-shrink: 0;
font-size: var(--app-text-xl);
&.success {
color: var(--td-success-color);
}
&.muted {
color: var(--td-text-color-placeholder);
}
&.warn {
color: var(--td-error-color);
}
}
/*
Inline action buttons inside the faux input. We use `variant="text"` so
they read as inline text affordances — same visual weight as the eye
toggle on the create-mode API key input. A 1px divider between Update
and Remove gives them just enough separation without adding boxes.
*/
.credential-actions {
display: flex;
align-items: center;
gap: 2px;
flex-shrink: 0;
:deep(.t-button--variant-text) {
height: 24px;
padding: 0 8px;
font-size: var(--app-text-sm);
border-radius: var(--app-radius-xs);
}
}
.action-divider {
width: 1px;
height: 14px;
background: var(--td-component-stroke);
margin: 0 2px;
}
/*
Editing state — let t-input own its frame; we just stack a tiny
end-aligned action row underneath. No border on this wrapper.
*/
.credential-edit {
display: flex;
flex-direction: column;
gap: 6px;
}
.credential-edit-actions {
display: flex;
justify-content: flex-end;
align-items: center;
gap: 4px;
:deep(.t-button) {
height: 28px;
padding: 0 12px;
font-size: var(--app-text-sm);
}
}
</style>