feat(ui): AI Usage Sacred Valley card + /api/ai-usage proxy (Claude tokens + local model perf)

Summarises the Homelab Monitor (CT300 :8080) into a blackflame card: Claude Code
token usage today/week + top model, and OpenClaw/Ollama p50/p95 latency + error rate.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

lib/api/index.js

@@ -31,6 +31,7 @@ import { router as healthRouter } from './routes/health.js';
 import { router as securityRouter } from './routes/security.js';
 import { router as actionsRouter } from './routes/actions.js';
 import { router as littleblueRouter } from './routes/littleblue.js';
+import { router as aiUsageRouter } from './routes/ai_usage.js';
 
 export function mountApi(app) {
   const api = Router();
@@ -45,6 +46,7 @@ export function mountApi(app) {
   api.use('/security', securityRouter);
   api.use('/actions', actionsRouter);
   api.use('/little-blue', littleblueRouter);
+  api.use('/ai-usage', aiUsageRouter);
   api.use('/projects', projectsRouter);
   api.use('/projects/:project_id/tasks', tasksByProjectRouter);
   api.use('/tasks', tasksRouter);

lib/api/routes/ai_usage.js

@@ -0,0 +1,55 @@
+import { Router } from 'express';
+import { asyncWrap } from '../errors.js';
+
+// Surfaces the Homelab Monitor (phuryn/claude-usage fork) on CT 300 as a compact
+// summary for the Sacred Valley "AI Usage" card. The monitor owns the data
+// (scans ~/.claude/**/*.jsonl + OpenClaw trajectories → usage.db); we just fetch
+// + condense its API over the LAN.
+const MON = () => process.env.AI_USAGE_URL || 'http://192.168.1.212:8080';
+
+// Pure: condense the monitor's /api/data + /api/openclaw into the card summary.
+export function summarizeUsage(data, openclaw) {
+  const rows = (data && data.daily_by_model) || [];
+  const days = [...new Set(rows.map(r => r.day))].sort();
+  const today = days[days.length - 1] || null;
+  const weekDays = new Set(days.slice(-7));
+  const acc = (pred) => rows.filter(pred).reduce((a, r) => ({
+    input: a.input + (r.input || 0),
+    output: a.output + (r.output || 0),
+    cache: a.cache + (r.cache_read || 0) + (r.cache_creation || 0),
+    turns: a.turns + (r.turns || 0)
+  }), { input: 0, output: 0, cache: 0, turns: 0 });
+
+  const modelTokens = {};
+  for (const r of rows) if (weekDays.has(r.day)) modelTokens[r.model] = (modelTokens[r.model] || 0) + (r.input || 0) + (r.output || 0);
+  const topModel = Object.entries(modelTokens).sort((a, b) => b[1] - a[1])[0]?.[0] || null;
+
+  const byModel = (openclaw && openclaw.summary && openclaw.summary.by_model) || [];
+  const sorted = [...byModel].sort((a, b) => (b.count || 0) - (a.count || 0));
+  const clean = (m) => m.replace(/^ollama\//, '');
+  const local = {
+    runs: byModel.reduce((a, m) => a + (m.count || 0), 0),
+    models: sorted.map(m => ({ model: clean(m.model), count: m.count, p50_ms: m.p50_ms, p95_ms: m.p95_ms, error_rate: m.error_rate })),
+    top: sorted[0] ? { model: clean(sorted[0].model), p50_ms: sorted[0].p50_ms, p95_ms: sorted[0].p95_ms, error_rate: sorted[0].error_rate } : null
+  };
+
+  return {
+    ok: true,
+    generated_at: data?.generated_at || null,
+    today,
+    claude: { today: acc(r => r.day === today), week: acc(r => weekDays.has(r.day)), top_model: topModel },
+    local
+  };
+}
+
+export const router = Router();
+
+router.get('/', asyncWrap(async (_req, res) => {
+  const base = MON();
+  const [data, openclaw] = await Promise.all([
+    fetch(`${base}/api/data`, { signal: AbortSignal.timeout(5000) }).then(r => r.ok ? r.json() : null).catch(() => null),
+    fetch(`${base}/api/openclaw`, { signal: AbortSignal.timeout(5000) }).then(r => r.ok ? r.json() : null).catch(() => null)
+  ]);
+  if (!data) return res.json({ ok: false, error: 'monitor unreachable', url: base });
+  res.json(summarizeUsage(data, openclaw));
+}));