feat: LLM smart router with embedding-based prompt classification

apps/dashboard/src/api/client.ts

@@ -636,6 +636,7 @@ export type Schedule = {
   max_concurrency: number;
   timeout_ms?: number;
   digest_mode: DigestMode;
+  routing_profile?: string;
   fetch_config: FetchConfig;
   source_states: Record<string, SourceState>;
   max_catchup_runs: number;
@@ -671,6 +672,7 @@ export type CreateScheduleRequest = {
   max_concurrency?: number;
   timeout_ms?: number;
   digest_mode?: DigestMode;
+  routing_profile?: string;
   fetch_config?: Partial<FetchConfig>;
   max_catchup_runs?: number;
 };
@@ -688,6 +690,7 @@ export type UpdateScheduleRequest = {
   max_concurrency?: number;
   timeout_ms?: number | null;
   digest_mode?: DigestMode;
+  routing_profile?: string;
   fetch_config?: Partial<FetchConfig>;
   max_catchup_runs?: number;
 };
@@ -756,6 +759,38 @@ export type QuotaListResponse = {
   quotas: QuotaStatus[];
 };
 
+// ── Router types ────────────────────────────────────────────────────
+
+export type RouterStatus = {
+  enabled: boolean;
+  default_profile: string;
+  classifier: {
+    provider: string;
+    model: string;
+    connected: boolean;
+    avg_latency_ms?: number;
+  };
+  tiers: Record<string, string[]>;
+  thresholds: Record<string, number>;
+};
+
+export type ClassifyResult = {
+  tier: string;
+  scores: Record<string, number>;
+  resolved_model: string;
+  latency_ms: number;
+};
+
+export type RouterDecision = {
+  timestamp: string;
+  prompt_snippet: string;
+  profile: string;
+  tier: string;
+  model: string;
+  latency_ms: number;
+  bypassed: boolean;
+};
+
 // ── API functions ──────────────────────────────────────────────────
 
 export const api = {
@@ -871,4 +906,11 @@ export const api = {
   // Provider listing
   providers: () => get<{ providers: string[]; count: number }>("/v1/models"),
   roles: () => get<{ roles: Record<string, string> }>("/v1/models/roles"),
+
+  // Router
+  routerStatus: () => get<RouterStatus>("/v1/router/status"),
+  classifyPrompt: (prompt: string) =>
+    post<ClassifyResult>("/v1/router/classify", { prompt }),
+  routerDecisions: (limit = 100) =>
+    get<{ decisions: RouterDecision[]; count: number }>(`/v1/router/decisions?limit=${limit}`),
 };

apps/dashboard/src/pages/ScheduleDetail.vue

@@ -33,6 +33,7 @@ const editMissedPolicy = ref<MissedPolicy>("run_once");
 const editDigestMode = ref<DigestMode>("full");
 const editMaxConcurrency = ref(1);
 const editMaxCatchupRuns = ref(5);
+const editRoutingProfile = ref("");
 const editTimeoutMs = ref<number | null>(null);
 const editSubmitting = ref(false);
 const editError = ref("");
@@ -49,6 +50,7 @@ function startEdit() {
   editDigestMode.value = s.digest_mode;
   editMaxConcurrency.value = s.max_concurrency;
   editMaxCatchupRuns.value = s.max_catchup_runs;
+  editRoutingProfile.value = s.routing_profile ?? "";
   editTimeoutMs.value = s.timeout_ms ?? null;
   editError.value = "";
   editing.value = true;
@@ -83,6 +85,7 @@ async function submitEdit() {
     digest_mode: editDigestMode.value,
     max_concurrency: editMaxConcurrency.value,
     max_catchup_runs: editMaxCatchupRuns.value,
+    routing_profile: editRoutingProfile.value === "" ? null : editRoutingProfile.value,
     timeout_ms: editTimeoutMs.value,
   };
 
@@ -368,6 +371,17 @@ function goToRun(runId?: string) {
             <label>Max Catch-up Runs</label>
             <input type="number" v-model.number="editMaxCatchupRuns" min="1" max="100" />
           </div>
+          <div class="edit-field">
+            <label>Routing Profile</label>
+            <select v-model="editRoutingProfile">
+              <option value="">Default (inherit)</option>
+              <option value="auto">Auto</option>
+              <option value="eco">Eco</option>
+              <option value="premium">Premium</option>
+              <option value="free">Free</option>
+              <option value="reasoning">Reasoning</option>
+            </select>
+          </div>
         </div>
 
         <div class="edit-field">

apps/dashboard/src/pages/Schedules.vue

@@ -34,6 +34,7 @@ const formMissedPolicy = ref<"skip" | "run_once" | "catch_up">("run_once");
 const formDigestMode = ref<"full" | "changes_only">("full");
 const formMaxConcurrency = ref(1);
 const formMaxCatchupRuns = ref(5);
+const formRoutingProfile = ref("");
 const formSubmitting = ref(false);
 const formError = ref("");
 
@@ -90,6 +91,7 @@ function openForm() {
   formDigestMode.value = "full";
   formMaxConcurrency.value = 1;
   formMaxCatchupRuns.value = 5;
+  formRoutingProfile.value = "";
   formError.value = "";
 }
 
@@ -121,6 +123,7 @@ async function submitForm() {
     digest_mode: formDigestMode.value,
     max_concurrency: formMaxConcurrency.value,
     max_catchup_runs: formMaxCatchupRuns.value,
+    routing_profile: formRoutingProfile.value === "" ? undefined : formRoutingProfile.value,
   };
 
   formSubmitting.value = true;
@@ -299,6 +302,17 @@ function goToSchedule(id: string) {
           <label>Max Catch-up Runs</label>
           <input type="number" v-model.number="formMaxCatchupRuns" min="1" max="100" />
         </div>
+        <div class="field">
+          <label>Routing Profile</label>
+          <select v-model="formRoutingProfile">
+            <option value="">Default (inherit)</option>
+            <option value="auto">Auto</option>
+            <option value="eco">Eco</option>
+            <option value="premium">Premium</option>
+            <option value="free">Free</option>
+            <option value="reasoning">Reasoning</option>
+          </select>
+        </div>
       </div>
 
       <div class="field toggle-field">

apps/dashboard/src/pages/Settings.vue

@@ -2,6 +2,7 @@
 import { ref, computed, onMounted } from "vue";
 import { api, ApiError, setApiToken, getApiToken } from "@/api/client";
 import type { SystemInfo, ReadinessResponse } from "@/api/client";
+import type { RouterStatus, RouterDecision } from "@/api/client";
 import Card from "@/components/Card.vue";
 import LoadingPanel from "@/components/LoadingPanel.vue";
 import ConfigEditor from "@/components/ConfigEditor.vue";
@@ -20,6 +21,27 @@ const tokenSaved = ref(false);
 // Restart
 const restarting = ref(false);
 
+// Router
+const routerStatus = ref<RouterStatus | null>(null);
+const routerDecisions = ref<RouterDecision[]>([]);
+const routerLoading = ref(false);
+const routerError = ref("");
+const decisionsExpanded = ref(false);
+
+async function loadRouter() {
+  routerLoading.value = true;
+  routerError.value = "";
+  try {
+    routerStatus.value = await api.routerStatus();
+    const res = await api.routerDecisions(20);
+    routerDecisions.value = res.decisions;
+  } catch (e: unknown) {
+    routerError.value = e instanceof ApiError ? e.friendly : String(e);
+  } finally {
+    routerLoading.value = false;
+  }
+}
+
 const generatedToml = computed(() => {
   if (!sysInfo.value || !readiness.value) return "";
   return configToToml(sysInfo.value, readiness.value);
@@ -64,7 +86,10 @@ async function load() {
   }
 }
 
-onMounted(load);
+onMounted(() => {
+  load();
+  loadRouter();
+});
 </script>
 
 <template>
@@ -175,6 +200,63 @@ onMounted(load);
           </div>
         </Card>
       </template>
+
+      <!-- LLM Router -->
+      <Card v-if="routerLoading && !routerStatus" title="LLM Router">
+        <LoadingPanel message="Loading router status..." />
+      </Card>
+      <Card v-if="routerStatus || routerError" title="LLM Router">
+        <template v-if="routerStatus">
+          <div class="readiness-header">
+            <span :class="routerStatus.enabled ? 'status-ok' : 'status-warn'">
+              {{ routerStatus.enabled ? "Enabled" : "Disabled" }}
+            </span>
+            <span class="profile-badge">{{ routerStatus.default_profile }}</span>
+          </div>
+
+          <!-- Classifier Status -->
+          <div class="sub-heading">Classifier</div>
+          <div class="settings-grid">
+            <div><span class="label">Provider</span> <span class="mono val">{{ routerStatus.classifier.provider }}</span></div>
+            <div><span class="label">Model</span> <span class="mono val">{{ routerStatus.classifier.model }}</span></div>
+            <div><span class="label">Status</span>
+              <span :class="routerStatus.classifier.connected ? 'status-ok' : 'status-warn'">
+                {{ routerStatus.classifier.connected ? "Connected" : "Disconnected" }}
+              </span>
+            </div>
+            <div v-if="routerStatus.classifier.avg_latency_ms != null">
+              <span class="label">Avg Latency</span>
+              <span class="mono val">{{ routerStatus.classifier.avg_latency_ms }}ms</span>
+            </div>
+          </div>
+
+          <!-- Tier Assignments -->
+          <div class="sub-heading" style="margin-top: 0.8rem">Tier Assignments</div>
+          <div v-for="(models, tier) in routerStatus.tiers" :key="tier" class="tier-row">
+            <span class="tier-label">{{ tier }}</span>
+            <span class="mono val">{{ models.join(", ") || "\u2014" }}</span>
+          </div>
+
+          <!-- Recent Decisions (collapsible) -->
+          <div class="sub-heading clickable" style="margin-top: 0.8rem" @click="decisionsExpanded = !decisionsExpanded">
+            Recent Decisions {{ decisionsExpanded ? "\u25BE" : "\u25B8" }}
+          </div>
+          <div v-if="decisionsExpanded && routerDecisions.length > 0" class="decisions-log">
+            <div v-for="(d, idx) in routerDecisions" :key="idx" class="decision-row">
+              <span class="dim">{{ new Date(d.timestamp).toLocaleTimeString() }}</span>
+              <span class="tier-badge" :class="'tier-' + d.tier">{{ d.tier }}</span>
+              <span class="mono">{{ d.model }}</span>
+              <span class="dim">{{ d.latency_ms }}ms</span>
+              <span class="dim decision-snippet">{{ d.prompt_snippet }}</span>
+            </div>
+          </div>
+          <div v-if="decisionsExpanded && routerDecisions.length === 0" class="dim">
+            No routing decisions recorded yet.
+          </div>
+        </template>
+
+        <p v-if="routerError" class="error">{{ routerError }}</p>
+      </Card>
     </template>
 
     <!-- Edit mode: TOML config editor -->
@@ -297,4 +379,58 @@ button.secondary:hover { color: var(--text); border-color: var(--text-dim); }
 button.secondary.danger { border-color: var(--red); color: var(--red); }
 button.secondary.danger:hover { background: var(--red); color: #fff; }
 button.secondary.danger:disabled { opacity: 0.5; cursor: not-allowed; }
+
+/* Router card */
+.profile-badge {
+  background: var(--accent);
+  color: #fff;
+  padding: 0.15rem 0.5rem;
+  border-radius: 3px;
+  font-size: 0.75rem;
+  font-weight: 600;
+  text-transform: uppercase;
+}
+.tier-row {
+  display: flex;
+  align-items: center;
+  gap: 0.8rem;
+  padding: 0.2rem 0;
+  font-size: 0.82rem;
+}
+.tier-label {
+  min-width: 5rem;
+  color: var(--text-dim);
+  font-weight: 500;
+  text-transform: capitalize;
+}
+.clickable { cursor: pointer; user-select: none; }
+.decisions-log {
+  max-height: 300px;
+  overflow-y: auto;
+  font-size: 0.78rem;
+}
+.decision-row {
+  display: flex;
+  align-items: center;
+  gap: 0.6rem;
+  padding: 0.2rem 0;
+  border-bottom: 1px solid var(--border);
+}
+.decision-snippet {
+  flex: 1;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+.tier-badge {
+  padding: 0.1rem 0.4rem;
+  border-radius: 3px;
+  font-size: 0.7rem;
+  font-weight: 600;
+  text-transform: uppercase;
+}
+.tier-simple { background: var(--green); color: #000; }
+.tier-complex { background: var(--accent); color: #fff; }
+.tier-reasoning { background: var(--red); color: #fff; }
+.tier-free { background: var(--text-dim); color: #fff; }
 </style>