diff --git a/backend/src/index.ts b/backend/src/index.ts
index 951a817..b6f6204 100644
--- a/backend/src/index.ts
+++ b/backend/src/index.ts
@@ -264,6 +264,16 @@ app.get('/api/admin/metrics/failed-items', createAdminLimiter(), adminAuth, asyn
   }
 });
 
+app.get('/api/admin/metrics/api-latency', createAdminLimiter(), adminAuth, async (req, res) => {
+  try {
+    const metrics = await monitoringService.getApiLatencyMetrics();
+    res.json(metrics);
+  } catch (error) {
+    logger.error('Error fetching API latency metrics:', error);
+    res.status(500).json({ error: 'Failed to fetch API latency metrics' });
+  }
+});
+
 app.get('/api/admin/metrics/ops-summary', createAdminLimiter(), adminAuth, async (req, res) => {
   try {
     const w = req.query.window as string;
@@ -271,7 +281,7 @@ app.get('/api/admin/metrics/ops-summary', createAdminLimiter(), adminAuth, async
     if (isNaN(windowHours) || windowHours < 1 || windowHours > 720) {
       return res.status(400).json({ error: 'window must be between 1 and 720 hours' });
     }
-    const [subscriptions, renewals, activity, trials, throughput, latency, retries] =
+    const [subscriptions, renewals, activity, trials, throughput, latency, retries, apiLatency] =
       await Promise.all([
         monitoringService.getSubscriptionMetrics(),
         monitoringService.getRenewalMetrics(),
@@ -280,6 +290,7 @@ app.get('/api/admin/metrics/ops-summary', createAdminLimiter(), adminAuth, async
         monitoringService.getThroughputMetrics(windowHours),
         monitoringService.getLatencyMetrics(windowHours),
         monitoringService.getRetryMetrics(windowHours),
+        monitoringService.getApiLatencyMetrics(),
       ]);
     res.json({
       generated_at: new Date().toISOString(),
@@ -291,6 +302,7 @@ app.get('/api/admin/metrics/ops-summary', createAdminLimiter(), adminAuth, async
       throughput,
       latency,
       retries,
+      api_latency: apiLatency,
       db_pool: monitoringService.getPoolMetrics(),
     });
   } catch (error) {
diff --git a/backend/src/middleware/requestLogger.ts b/backend/src/middleware/requestLogger.ts
index 5587e4b..b525113 100644
--- a/backend/src/middleware/requestLogger.ts
+++ b/backend/src/middleware/requestLogger.ts
@@ -1,5 +1,6 @@
 import { Request, Response, NextFunction } from 'express';
 import logger from '../config/logger';
+import { apiLatencyService } from '../services/api-latency-service';
 
 /**
  * Logs the start and end of every HTTP request including:
@@ -23,7 +24,7 @@ export function requestLoggerMiddleware(
   });
 
   // Hook into the response 'finish' event to log completion
-  res.on('finish', () => {
+  res.on('finish', async () => {
     const durationMs = Date.now() - startMs;
     const level = res.statusCode >= 500 ? 'error'
                 : res.statusCode >= 400 ? 'warn'
@@ -35,6 +36,14 @@ export function requestLoggerMiddleware(
       statusCode: res.statusCode,
       durationMs,
     });
+
+    // Record latency for endpoint family
+    try {
+      const family = apiLatencyService.getEndpointFamily(req.method, req.path);
+      await apiLatencyService.recordLatency(family, durationMs);
+    } catch (error) {
+      logger.error('Failed to record API latency metric:', error);
+    }
   });
 
   next();
diff --git a/backend/src/services/api-latency-service.ts b/backend/src/services/api-latency-service.ts
new file mode 100644
index 0000000..c352a3a
--- /dev/null
+++ b/backend/src/services/api-latency-service.ts
@@ -0,0 +1,154 @@
+
+import { RateLimitRedisStore } from '../lib/redis-store';
+import logger from '../config/logger';
+
+export interface EndpointLatencyMetrics {
+  family: string;
+  p50_ms: number;
+  p95_ms: number;
+  p99_ms: number;
+  avg_ms: number;
+  sample_count: number;
+}
+
+interface LatencyStoreEntry {
+  latencies: number[];
+  timestamp: number;
+}
+
+export class ApiLatencyService {
+  private static instance: ApiLatencyService | null = null;
+  private memoryStore: Map<string, LatencyStoreEntry> = new Map();
+  private readonly WINDOW_MS = 24 * 60 * 60 * 1000; // 24 hour window
+  private readonly MAX_SAMPLES_PER_FAMILY = 10000;
+  private readonly REDIS_KEY_PREFIX = 'api_latency:';
+
+  private constructor() {}
+
+  static getInstance(): ApiLatencyService {
+    if (!ApiLatencyService.instance) {
+      ApiLatencyService.instance = new ApiLatencyService();
+    }
+    return ApiLatencyService.instance;
+  }
+
+  /**
+   * Normalize a request path to an endpoint family.
+   * Replaces dynamic path segments like IDs with placeholders.
+   */
+  getEndpointFamily(method: string, path: string): string {
+    let normalizedPath = path;
+
+    // Remove leading /api/ prefix if present
+    if (normalizedPath.startsWith('/api/')) {
+      normalizedPath = normalizedPath.slice(4);
+    }
+
+    // Split into segments and replace UUIDs/numbers with placeholders
+    const segments = normalizedPath.split('/').map(segment => {
+      if (/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(segment)) {
+        return ':id';
+      }
+      if (/^\d+$/.test(segment)) {
+        return ':id';
+      }
+      return segment;
+    });
+
+    return `${method.toUpperCase()} /${segments.join('/')}`;
+  }
+
+  /**
+   * Record a latency measurement for an endpoint family.
+   */
+  async recordLatency(family: string, latencyMs: number): Promise<void> {
+    try {
+      const redisStore = RateLimitRedisStore.getInstance();
+      if (redisStore.isAvailable()) {
+        // TODO: Implement Redis-based storage once we have access to raw Redis client
+        this.recordToMemory(family, latencyMs);
+      } else {
+        this.recordToMemory(family, latencyMs);
+      }
+    } catch (error) {
+      logger.error('Failed to record API latency:', error);
+    }
+  }
+
+  private recordToMemory(family: string, latencyMs: number): void {
+    const now = Date.now();
+    const entry = this.memoryStore.get(family) || { latencies: [], timestamp: now };
+    
+    // Clean up old entries
+    if (now - entry.timestamp > this.WINDOW_MS) {
+      entry.latencies = [];
+      entry.timestamp = now;
+    }
+
+    // Add new latency
+    entry.latencies.push(latencyMs);
+
+    // Trim to max samples
+    if (entry.latencies.length > this.MAX_SAMPLES_PER_FAMILY) {
+      entry.latencies = entry.latencies.slice(-this.MAX_SAMPLES_PER_FAMILY);
+    }
+
+    this.memoryStore.set(family, entry);
+  }
+
+  /**
+   * Compute percentile metrics for all endpoint families.
+   */
+  async getLatencyMetrics(): Promise<EndpointLatencyMetrics[]> {
+    const metrics: EndpointLatencyMetrics[] = [];
+    const now = Date.now();
+
+    for (const [family, entry] of this.memoryStore.entries()) {
+      // Skip if entry is too old
+      if (now - entry.timestamp > this.WINDOW_MS) {
+        continue;
+      }
+
+      const sorted = [...entry.latencies].sort((a, b) => a - b);
+      const percentiles = this.computePercentiles(sorted);
+
+      metrics.push({
+        family,
+        ...percentiles,
+      });
+    }
+
+    // Sort by sample count descending
+    return metrics.sort((a, b) => b.sample_count - a.sample_count);
+  }
+
+  private computePercentiles(sorted: number[]): Omit<EndpointLatencyMetrics, 'family'> {
+    if (sorted.length === 0) {
+      return {
+        p50_ms: 0,
+        p95_ms: 0,
+        p99_ms: 0,
+        avg_ms: 0,
+        sample_count: 0,
+      };
+    }
+
+    const at = (pct: number) => {
+      const idx = Math.min(Math.ceil((pct / 100) * sorted.length) - 1, sorted.length - 1);
+      return sorted[Math.max(0, idx)];
+    };
+
+    const avg = sorted.reduce((sum, val) => sum + val, 0) / sorted.length;
+
+    return {
+      p50_ms: at(50),
+      p95_ms: at(95),
+      p99_ms: at(99),
+      avg_ms: Math.round(avg),
+      sample_count: sorted.length,
+    };
+  }
+}
+
+export const apiLatencyService = ApiLatencyService.getInstance();
+
diff --git a/backend/src/services/monitoring-service.ts b/backend/src/services/monitoring-service.ts
index 3ea60a1..46d0179 100644
--- a/backend/src/services/monitoring-service.ts
+++ b/backend/src/services/monitoring-service.ts
@@ -1,6 +1,7 @@
 import { supabase, monitorPool, PoolMetrics } from '../config/database';
 import logger from '../config/logger';
 import { ExternalServiceClient, ServiceMetrics } from '../utils/external-service-client';
+import { apiLatencyService, EndpointLatencyMetrics } from './api-latency-service';
 
 // ─── Existing interfaces ────────────────────────────────────────────────────
 
@@ -652,6 +653,13 @@ export class MonitoringService {
             }
         })(), contextId);
     }
+
+    /**
+     * Get API latency percentiles per endpoint family.
+     */
+    async getApiLatencyMetrics(): Promise<EndpointLatencyMetrics[]> {
+        return apiLatencyService.getLatencyMetrics();
+    }
 }
 
 export const monitoringService = new MonitoringService();
diff --git a/backend/tests/monitoring-api.test.ts b/backend/tests/monitoring-api.test.ts
index 1f7117d..10d37f1 100644
--- a/backend/tests/monitoring-api.test.ts
+++ b/backend/tests/monitoring-api.test.ts
@@ -15,6 +15,7 @@ jest.mock('../src/services/monitoring-service', () => ({
     getRetryMetrics: jest.fn(),
     getFailedItems: jest.fn(),
     getPoolMetrics: jest.fn(),
+    getApiLatencyMetrics: jest.fn(),
   },
 }));
 
@@ -129,13 +130,22 @@ app.get('/api/admin/metrics/failed-items', adminAuth, async (req, res) => {
   }
 });
 
+app.get('/api/admin/metrics/api-latency', adminAuth, async (_req, res) => {
+  try {
+    const metrics = await monitoringService.getApiLatencyMetrics();
+    res.json(metrics);
+  } catch {
+    res.status(500).json({ error: 'Failed to fetch API latency metrics' });
+  }
+});
+
 app.get('/api/admin/metrics/ops-summary', adminAuth, async (req, res) => {
   try {
     const windowHours = parseInt(req.query.window as string) || 24;
     if (windowHours < 1 || windowHours > 720) {
       return res.status(400).json({ error: 'window must be between 1 and 720 hours' });
     }
-    const [subscriptions, renewals, activity, trials, throughput, latency, retries] =
+    const [subscriptions, renewals, activity, trials, throughput, latency, retries, apiLatency] =
       await Promise.all([
         monitoringService.getSubscriptionMetrics(),
         monitoringService.getRenewalMetrics(),
@@ -144,6 +154,7 @@ app.get('/api/admin/metrics/ops-summary', adminAuth, async (req, res) => {
         monitoringService.getThroughputMetrics(windowHours),
         monitoringService.getLatencyMetrics(windowHours),
         monitoringService.getRetryMetrics(windowHours),
+        monitoringService.getApiLatencyMetrics(),
       ]);
     res.json({
       generated_at: new Date().toISOString(),
@@ -155,6 +166,7 @@ app.get('/api/admin/metrics/ops-summary', adminAuth, async (req, res) => {
       throughput,
       latency,
       retries,
+      api_latency: apiLatency,
       db_pool: monitoringService.getPoolMetrics(),
     });
   } catch {
@@ -167,6 +179,60 @@ app.get('/api/admin/metrics/ops-summary', adminAuth, async (req, res) => {
 const AUTH = { 'x-admin-api-key': 'test-admin-key' };
 const WRONG_AUTH = { 'x-admin-api-key': 'wrong-key' };
 
+// ── GET /api/admin/metrics/api-latency ───────────────────────────────────────
+
+describe('GET /api/admin/metrics/api-latency', () => {
+  const API_LATENCY_FIXTURE = [
+    {
+      family: 'GET /subscriptions',
+      p50_ms: 50,
+      p95_ms: 150,
+      p99_ms: 300,
+      avg_ms: 75,
+      sample_count: 100,
+    },
+    {
+      family: 'POST /webhooks',
+      p50_ms: 100,
+      p95_ms: 200,
+      p99_ms: 400,
+      avg_ms: 120,
+      sample_count: 50,
+    },
+  ];
+
+  it('returns 401 without auth key', async () => {
+    const res = await request(app).get('/api/admin/metrics/api-latency');
+    expect(res.status).toBe(401);
+  });
+
+  it('returns 401 with wrong auth key', async () => {
+    const res = await request(app)
+      .get('/api/admin/metrics/api-latency')
+      .set(WRONG_AUTH);
+    expect(res.status).toBe(401);
+  });
+
+  it('returns 200 with correct shape', async () => {
+    (monitoringService.getApiLatencyMetrics as jest.Mock).mockResolvedValue(API_LATENCY_FIXTURE);
+
+    const res = await request(app).get('/api/admin/metrics/api-latency').set(AUTH);
+
+    expect(res.status).toBe(200);
+    expect(res.body).toHaveLength(2);
+    expect(res.body[0].family).toBe('GET /subscriptions');
+    expect(res.body[0].p50_ms).toBe(50);
+    expect(res.body[0].sample_count).toBe(100);
+  });
+
+  it('returns 500 when service throws', async () => {
+    (monitoringService.getApiLatencyMetrics as jest.Mock).mockRejectedValue(new Error('fail'));
+
+    const res = await request(app).get('/api/admin/metrics/api-latency').set(AUTH);
+    expect(res.status).toBe(500);
+  });
+});
+
 // ── Original access-control suite ────────────────────────────────────────────
 
 describe('Monitoring API Access Control', () => {
@@ -498,6 +564,9 @@ describe('GET /api/admin/metrics/ops-summary', () => {
     (monitoringService.getLatencyMetrics as jest.Mock).mockResolvedValue({ notification_delivery_latency: { p50_ms: 100 } });
     (monitoringService.getRetryMetrics as jest.Mock).mockResolvedValue({ total_retried: 1 });
     (monitoringService.getPoolMetrics as jest.Mock).mockReturnValue({ activeConnections: 0 });
+    (monitoringService.getApiLatencyMetrics as jest.Mock).mockResolvedValue([
+      { family: 'GET /subscriptions', p50_ms: 50, sample_count: 10 },
+    ]);
   };
 
   it('returns 401 without auth key', async () => {
@@ -524,8 +593,10 @@ describe('GET /api/admin/metrics/ops-summary', () => {
     expect(res.body).toHaveProperty('throughput');
     expect(res.body).toHaveProperty('latency');
     expect(res.body).toHaveProperty('retries');
+    expect(res.body).toHaveProperty('api_latency');
     expect(res.body).toHaveProperty('db_pool');
     expect(res.body.subscriptions.total_subscriptions).toBe(5);
+    expect(res.body.api_latency).toHaveLength(1);
   });
 
   it('accepts custom window query param and passes it to all sub-metrics', async () => {
diff --git a/contracts/Cargo.lock b/contracts/Cargo.lock
index 7d28b28..1645608 100644
--- a/contracts/Cargo.lock
+++ b/contracts/Cargo.lock
@@ -2,6 +2,13 @@
 # It is not intended for manual editing.
 version = 4
 
+[[package]]
+name = "agent-registry"
+version = "0.1.0"
+dependencies = [
+ "soroban-sdk",
+]
+
 [[package]]
 name = "ahash"
 version = "0.8.12"