diff --git a/README.md b/README.md
index 8ed056a..497a8a2 100644
--- a/README.md
+++ b/README.md
@@ -50,3 +50,69 @@ services and APIs.
   compare with `sample-conf.yaml`.
 * Start aperture without any command line parameters (`./aperture`), all configuration
   is done in the `~/.aperture/aperture.yaml` file.
+
+## Rate Limiting
+
+Aperture supports optional per-endpoint rate limiting using a token bucket
+algorithm. Rate limits are configured per service and applied based on the
+client's L402 token ID for authenticated requests, or IP address for
+unauthenticated requests.
+
+### Features
+
+* **Token bucket algorithm**: Allows controlled bursting while maintaining a
+  steady-state request rate.
+* **Per-client isolation**: Each L402 token ID or IP address has independent
+  rate limit buckets.
+* **Path-based rules**: Different endpoints can have different rate limits using
+  regular expressions.
+* **Multiple rules**: All matching rules are evaluated; if any rule denies the
+  request, it is rejected. This allows layering global and endpoint-specific
+  limits.
+* **Protocol-aware responses**: Returns HTTP 429 with `Retry-After` header for
+  REST requests, and gRPC `ResourceExhausted` status for gRPC requests.
+
+### Configuration
+
+Rate limits are configured in the `ratelimits` section of each service:
+
+```yaml
+services:
+  - name: "myservice"
+    hostregexp: "api.example.com"
+    address: "127.0.0.1:8080"
+    protocol: https
+
+    ratelimits:
+      # Global rate limit for all endpoints
+      - requests: 100    # Requests allowed per time window
+        per: 1s          # Time window duration (1s, 1m, 1h, etc.)
+        burst: 100       # Max burst capacity (defaults to 'requests')
+
+      # Stricter limit for expensive endpoints
+      - pathregexp: '^/api/v1/expensive.*$'
+        requests: 5
+        per: 1m
+        burst: 5
+```
+
+This example configures two rate limit rules using a token bucket algorithm. Each
+client gets a "bucket" of tokens that refills at the `requests/per` rate, up to the
+`burst` capacity. A request consumes one token; if no tokens are available, the
+request is rejected. This allows clients to make quick bursts of requests (up to
+`burst`) while enforcing a steady-state rate limit over time.
+
+1. **Global limit**: All endpoints are limited to 100 requests per second per client,
+   with a burst capacity of 100.
+2. **Endpoint-specific limit**: Paths matching `/api/v1/expensive.*` have a stricter
+   limit of 5 requests per minute with a burst of 5. Since both rules are evaluated,
+   requests to expensive endpoints must satisfy both limits.
+
+### Configuration Options
+
+| Option | Description | Required |
+|--------|-------------|----------|
+| `pathregexp` | Regular expression to match request paths. If omitted, matches all paths. | No |
+| `requests` | Number of requests allowed per time window. | Yes |
+| `per` | Time window duration (e.g., `1s`, `1m`, `1h`). | Yes |
+| `burst` | Maximum burst size. Defaults to `requests` if not set. | No |
diff --git a/freebie/mem_store.go b/freebie/mem_store.go
index 4a97b98..848404a 100644
--- a/freebie/mem_store.go
+++ b/freebie/mem_store.go
@@ -3,10 +3,8 @@ package freebie
 import (
 	"net"
 	"net/http"
-)
 
-var (
-	defaultIPMask = net.IPv4Mask(0xff, 0xff, 0xff, 0x00)
+	"github.com/lightninglabs/aperture/netutil"
 )
 
 type Count uint16
@@ -17,7 +15,7 @@ type memStore struct {
 }
 
 func (m *memStore) getKey(ip net.IP) string {
-	return ip.Mask(defaultIPMask).String()
+	return netutil.MaskIP(ip).String()
 }
 
 func (m *memStore) currentCount(ip net.IP) Count {
@@ -38,10 +36,10 @@ func (m *memStore) TallyFreebie(r *http.Request, ip net.IP) (bool, error) {
 	return true, nil
 }
 
-// NewMemIPMaskStore creates a new in-memory freebie store that masks the last
-// byte of an IP address to keep track of free requests. The last byte of the
-// address is discarded for the mapping to reduce risk of abuse by users that
-// have a whole range of IPs at their disposal.
+// NewMemIPMaskStore creates a new in-memory freebie store that masks IP
+// addresses to keep track of free requests. IPv4 addresses are masked to /24
+// and IPv6 addresses to /48. This reduces risk of abuse by users that have a
+// whole range of IPs at their disposal.
 func NewMemIPMaskStore(numFreebies Count) DB {
 	return &memStore{
 		numFreebies:    numFreebies,
diff --git a/netutil/ip.go b/netutil/ip.go
new file mode 100644
index 0000000..d2ede7f
--- /dev/null
+++ b/netutil/ip.go
@@ -0,0 +1,29 @@
+package netutil
+
+import "net"
+
+var (
+	// ipv4Mask24 masks IPv4 addresses to /24 (last octet zeroed).
+	// This groups clients on the same subnet together.
+	ipv4Mask24 = net.CIDRMask(24, 32)
+
+	// ipv6Mask48 masks IPv6 addresses to /48.
+	// Residential connections typically receive /48 to /64 allocations,
+	// so /48 provides reasonable grouping for rate limiting purposes.
+	ipv6Mask48 = net.CIDRMask(48, 128)
+)
+
+// MaskIP returns a masked version of the IP address for grouping purposes.
+// IPv4 addresses are masked to /24 (zeroing the last octet).
+// IPv6 addresses are masked to /48.
+//
+// This is useful for rate limiting and freebie tracking where we want to
+// group requests from the same network segment rather than individual IPs,
+// reducing abuse potential from users with multiple addresses.
+func MaskIP(ip net.IP) net.IP {
+	if ip4 := ip.To4(); ip4 != nil {
+		return ip4.Mask(ipv4Mask24)
+	}
+
+	return ip.Mask(ipv6Mask48)
+}
diff --git a/netutil/ip_test.go b/netutil/ip_test.go
new file mode 100644
index 0000000..6c714d0
--- /dev/null
+++ b/netutil/ip_test.go
@@ -0,0 +1,117 @@
+package netutil
+
+import (
+	"net"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+// TestMaskIP verifies that MaskIP correctly applies /24 masks to IPv4 and /48
+// masks to IPv6 addresses.
+func TestMaskIP(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{
+			name:     "IPv4 masks last octet",
+			input:    "192.168.1.123",
+			expected: "192.168.1.0",
+		},
+		{
+			name:     "IPv4 already masked",
+			input:    "10.0.0.0",
+			expected: "10.0.0.0",
+		},
+		{
+			name:     "IPv4 different last octet same result",
+			input:    "192.168.1.255",
+			expected: "192.168.1.0",
+		},
+		{
+			name:     "IPv6 masks to /48",
+			input:    "2001:db8:1234:5678:9abc:def0:1234:5678",
+			expected: "2001:db8:1234::",
+		},
+		{
+			name:     "IPv6 already masked",
+			input:    "2001:db8:abcd::",
+			expected: "2001:db8:abcd::",
+		},
+		{
+			name:     "IPv6 loopback",
+			input:    "::1",
+			expected: "::",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			ip := net.ParseIP(tc.input)
+			require.NotNil(t, ip, "failed to parse input IP")
+
+			result := MaskIP(ip)
+			require.Equal(t, tc.expected, result.String())
+		})
+	}
+}
+
+// TestMaskIP_SameSubnetGroupsTogether verifies that IPv4 addresses in the same
+// /24 subnet produce identical masked results.
+func TestMaskIP_SameSubnetGroupsTogether(t *testing.T) {
+	// Verify that IPs in the same /24 subnet produce the same masked result.
+	ips := []string{
+		"192.168.1.1",
+		"192.168.1.100",
+		"192.168.1.255",
+	}
+
+	results := make([]string, 0, len(ips))
+	for _, ipStr := range ips {
+		ip := net.ParseIP(ipStr)
+		results = append(results, MaskIP(ip).String())
+	}
+
+	// All should be the same.
+	for i := 1; i < len(results); i++ {
+		require.Equal(t, results[0], results[i],
+			"IPs in same /24 should have same masked result")
+	}
+}
+
+// TestMaskIP_DifferentSubnetsDiffer verifies that IPv4 addresses in different
+// /24 subnets produce distinct masked results.
+func TestMaskIP_DifferentSubnetsDiffer(t *testing.T) {
+	ip1 := net.ParseIP("192.168.1.100")
+	ip2 := net.ParseIP("192.168.2.100")
+
+	result1 := MaskIP(ip1).String()
+	result2 := MaskIP(ip2).String()
+
+	require.NotEqual(t, result1, result2,
+		"IPs in different /24 subnets should have different masked results")
+}
+
+// TestMaskIP_IPv6SamePrefix48GroupsTogether verifies that IPv6 addresses
+// sharing the same /48 prefix produce identical masked results.
+func TestMaskIP_IPv6SamePrefix48GroupsTogether(t *testing.T) {
+	// IPs in the same /48 should produce the same masked result.
+	ips := []string{
+		"2001:db8:1234:0001::",
+		"2001:db8:1234:ffff::",
+		"2001:db8:1234:abcd:1234:5678:9abc:def0",
+	}
+
+	results := make([]string, 0, len(ips))
+	for _, ipStr := range ips {
+		ip := net.ParseIP(ipStr)
+		results = append(results, MaskIP(ip).String())
+	}
+
+	for i := 1; i < len(results); i++ {
+		require.Equal(t, results[0], results[i],
+			"IPs in same /48 should have same masked result")
+	}
+}
diff --git a/proxy/proxy.go b/proxy/proxy.go
index 7dfb5d4..ff47038 100644
--- a/proxy/proxy.go
+++ b/proxy/proxy.go
@@ -4,12 +4,14 @@ import (
 	"crypto/tls"
 	"crypto/x509"
 	"fmt"
+	"math"
 	"net"
 	"net/http"
 	"net/http/httputil"
 	"os"
 	"strconv"
 	"strings"
+	"time"
 
 	"github.com/lightninglabs/aperture/auth"
 	"github.com/lightninglabs/aperture/l402"
@@ -172,6 +174,26 @@ func (p *Proxy) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	// Determine auth level required to access service and dispatch request
 	// accordingly.
 	authLevel := target.AuthRequired(r)
+
+	// checkRateLimit is a helper that checks rate limits after determining
+	// the authentication status. This ensures we only use L402 token IDs
+	// for authenticated requests, preventing DoS via garbage tokens.
+	checkRateLimit := func(authenticated bool) bool {
+		if target.rateLimiter == nil {
+			return true
+		}
+		key := ExtractRateLimitKey(r, remoteIP, authenticated)
+		allowed, retryAfter := target.rateLimiter.Allow(r, key)
+		if !allowed {
+			prefixLog.Infof("Rate limit exceeded for key %s, "+
+				"retry after %v", key, retryAfter)
+			addCorsHeaders(w.Header())
+			sendRateLimitResponse(w, r, retryAfter)
+		}
+
+		return allowed
+	}
+
 	skipInvoiceCreation := target.SkipInvoiceCreation(r)
 	switch {
 	case authLevel.IsOn():
@@ -215,6 +237,11 @@ func (p *Proxy) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 
+		// User is authenticated, apply rate limit with L402 token ID.
+		if !checkRateLimit(true) {
+			return
+		}
+
 	case authLevel.IsFreebie():
 		// We only need to respect the freebie counter if the user
 		// is not authenticated at all.
@@ -267,6 +294,21 @@ func (p *Proxy) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 				)
 				return
 			}
+
+			// Unauthenticated freebie user, rate limit by IP.
+			if !checkRateLimit(false) {
+				return
+			}
+		} else if !checkRateLimit(true) {
+			// Authenticated user on freebie path, rate limit by
+			// L402 token.
+			return
+		}
+
+	default:
+		// Auth is off, rate limit by IP for unauthenticated access.
+		if !checkRateLimit(false) {
+			return
 		}
 	}
 
@@ -486,6 +528,36 @@ func sendDirectResponse(w http.ResponseWriter, r *http.Request,
 	}
 }
 
+// sendRateLimitResponse sends a rate limit exceeded response to the client.
+// For HTTP clients, it returns 429 Too Many Requests with Retry-After header.
+// For gRPC clients, it returns a ResourceExhausted status.
+func sendRateLimitResponse(w http.ResponseWriter, r *http.Request,
+	retryAfter time.Duration) {
+
+	// Round up to ensure clients don't retry before the limit resets.
+	retrySeconds := int(math.Ceil(retryAfter.Seconds()))
+	if retrySeconds < 1 {
+		retrySeconds = 1
+	}
+
+	// Set Retry-After header for both HTTP and gRPC.
+	w.Header().Set("Retry-After", strconv.Itoa(retrySeconds))
+
+	// Check if this is a gRPC request.
+	if strings.HasPrefix(r.Header.Get(hdrContentType), hdrTypeGrpc) {
+		w.Header().Set(
+			hdrGrpcStatus,
+			strconv.Itoa(int(codes.ResourceExhausted)),
+		)
+		w.Header().Set(hdrGrpcMessage, "rate limit exceeded")
+
+		// gRPC requires 200 OK even for errors.
+		w.WriteHeader(http.StatusOK)
+	} else {
+		http.Error(w, "rate limit exceeded", http.StatusTooManyRequests)
+	}
+}
+
 type trailerFixingTransport struct {
 	next http.RoundTripper
 }
diff --git a/proxy/ratelimit_config.go b/proxy/ratelimit_config.go
new file mode 100644
index 0000000..953b0db
--- /dev/null
+++ b/proxy/ratelimit_config.go
@@ -0,0 +1,56 @@
+package proxy
+
+import (
+	"regexp"
+	"time"
+)
+
+// RateLimitConfig defines a rate limiting rule for a specific path pattern.
+type RateLimitConfig struct {
+	// PathRegexp is a regular expression that matches request paths
+	// to which this rate limit applies. If empty, matches all paths.
+	PathRegexp string `long:"pathregexp" description:"Regular expression to match the path of the URL against for rate limiting"`
+
+	// Requests is the number of requests allowed per time window (Per).
+	Requests int `long:"requests" description:"Number of requests allowed per time window"`
+
+	// Per is the time window duration (e.g., 1s, 1m, 1h). Defaults to 1s.
+	Per time.Duration `long:"per" description:"Time window for rate limiting (e.g., 1s, 1m, 1h)"`
+
+	// Burst is the maximum number of requests that can be made in a burst,
+	// exceeding the steady-state rate. Defaults to Requests if not set.
+	Burst int `long:"burst" description:"Maximum burst size (defaults to Requests if not set)"`
+
+	// compiledPathRegexp is the compiled version of PathRegexp.
+	compiledPathRegexp *regexp.Regexp
+}
+
+// Rate returns the rate.Limit value (requests per second) for this
+// configuration.
+func (r *RateLimitConfig) Rate() float64 {
+	if r.Per == 0 {
+		return 0
+	}
+
+	return float64(r.Requests) / r.Per.Seconds()
+}
+
+// EffectiveBurst returns the burst value, defaulting to Requests if Burst
+// is 0.
+func (r *RateLimitConfig) EffectiveBurst() int {
+	if r.Burst == 0 {
+		return r.Requests
+	}
+
+	return r.Burst
+}
+
+// Matches returns true if the given path matches this rate limit's path
+// pattern.
+func (r *RateLimitConfig) Matches(path string) bool {
+	if r.compiledPathRegexp == nil {
+		return true // No pattern means match all
+	}
+
+	return r.compiledPathRegexp.MatchString(path)
+}
diff --git a/proxy/ratelimit_metrics.go b/proxy/ratelimit_metrics.go
new file mode 100644
index 0000000..2708eff
--- /dev/null
+++ b/proxy/ratelimit_metrics.go
@@ -0,0 +1,52 @@
+package proxy
+
+import (
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+)
+
+var (
+	// rateLimitAllowed counts requests that passed rate limiting.
+	rateLimitAllowed = promauto.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "aperture",
+			Subsystem: "ratelimit",
+			Name:      "allowed_total",
+			Help:      "Total number of requests allowed by rate limiter",
+		},
+		[]string{"service", "path_pattern"},
+	)
+
+	// rateLimitDenied counts requests denied by rate limiting.
+	rateLimitDenied = promauto.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "aperture",
+			Subsystem: "ratelimit",
+			Name:      "denied_total",
+			Help:      "Total number of requests denied by rate limiter",
+		},
+		[]string{"service", "path_pattern"},
+	)
+
+	// rateLimitCacheSize tracks the current size of the rate limiter cache.
+	rateLimitCacheSize = promauto.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Namespace: "aperture",
+			Subsystem: "ratelimit",
+			Name:      "cache_size",
+			Help:      "Current number of entries in the rate limiter cache",
+		},
+		[]string{"service"},
+	)
+
+	// rateLimitEvictions counts LRU cache evictions.
+	rateLimitEvictions = promauto.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "aperture",
+			Subsystem: "ratelimit",
+			Name:      "evictions_total",
+			Help:      "Total number of rate limiter cache evictions",
+		},
+		[]string{"service"},
+	)
+)
diff --git a/proxy/ratelimiter.go b/proxy/ratelimiter.go
new file mode 100644
index 0000000..ad837f9
--- /dev/null
+++ b/proxy/ratelimiter.go
@@ -0,0 +1,252 @@
+package proxy
+
+import (
+	"bytes"
+	"net"
+	"net/http"
+	"sync"
+	"time"
+
+	"github.com/lightninglabs/aperture/l402"
+	"github.com/lightninglabs/aperture/netutil"
+	"github.com/lightninglabs/neutrino/cache/lru"
+	"golang.org/x/time/rate"
+)
+
+const (
+	// DefaultMaxCacheSize is the default maximum number of rate limiter
+	// entries to keep in the LRU cache.
+	DefaultMaxCacheSize = 10_000
+)
+
+// limiterKey is a composite key for the rate limiter cache. Using a struct
+// instead of a concatenated string saves memory because the pathPattern field
+// can reference the same underlying string across multiple keys.
+type limiterKey struct {
+	// clientKey identifies the client (e.g., "ip:1.2.3.4" or "token:abc").
+	clientKey string
+	// pathPattern is the rate limit rule's PathRegexp (pointer to config's
+	// string, not a copy).
+	pathPattern string
+}
+
+// limiterEntry holds a rate.Limiter. Implements cache.Value interface.
+type limiterEntry struct {
+	limiter *rate.Limiter
+}
+
+// Size implements cache.Value. Returns 1 so the LRU cache counts entries
+// rather than bytes.
+func (e *limiterEntry) Size() (uint64, error) {
+	return 1, nil
+}
+
+// RateLimiter manages per-key rate limiters with LRU eviction.
+type RateLimiter struct {
+	// cacheMu protects the LRU cache which is not concurrency-safe.
+	cacheMu sync.Mutex
+
+	// configs is the list of rate limit configurations for this limiter.
+	configs []*RateLimitConfig
+
+	// cache is the LRU cache of rate limiter entries.
+	cache *lru.Cache[limiterKey, *limiterEntry]
+
+	// maxSize is the maximum number of entries in the cache.
+	maxSize int
+
+	// serviceName is used for metrics labels.
+	serviceName string
+}
+
+// RateLimiterOption is a functional option for configuring a RateLimiter.
+type RateLimiterOption func(*RateLimiter)
+
+// WithMaxCacheSize sets the maximum cache size.
+func WithMaxCacheSize(size int) RateLimiterOption {
+	return func(rl *RateLimiter) {
+		rl.maxSize = size
+	}
+}
+
+// NewRateLimiter creates a new RateLimiter with the given configurations.
+func NewRateLimiter(serviceName string, configs []*RateLimitConfig,
+	opts ...RateLimiterOption) *RateLimiter {
+
+	rl := &RateLimiter{
+		configs:     configs,
+		maxSize:     DefaultMaxCacheSize,
+		serviceName: serviceName,
+	}
+
+	for _, opt := range opts {
+		opt(rl)
+	}
+
+	// Initialize the LRU cache with the configured max size.
+	rl.cache = lru.NewCache[limiterKey, *limiterEntry](uint64(rl.maxSize))
+
+	return rl
+}
+
+// Allow checks if a request should be allowed based on all matching rate
+// limits. Returns (allowed, retryAfter) where retryAfter is the suggested
+// duration to wait if denied.
+func (rl *RateLimiter) Allow(r *http.Request, key string) (bool,
+	time.Duration) {
+
+	path := r.URL.Path
+
+	// Collect all matching configs and their reservations. We need to check
+	// all rules before consuming any tokens, so that if any rule denies we
+	// can cancel all reservations.
+	type ruleReservation struct {
+		cfg         *RateLimitConfig
+		reservation *rate.Reservation
+	}
+	reservations := make([]ruleReservation, 0, len(rl.configs))
+
+	for _, cfg := range rl.configs {
+		if !cfg.Matches(path) {
+			continue
+		}
+
+		// Create composite key: client key + path pattern for
+		// independent limiting per rule. Using a struct instead of
+		// string concatenation saves memory since pathPattern
+		// references the config's string.
+		cacheKey := limiterKey{
+			clientKey:   key,
+			pathPattern: cfg.PathRegexp,
+		}
+
+		limiter := rl.getOrCreateLimiter(cacheKey, cfg)
+		reservation := limiter.Reserve()
+
+		reservations = append(reservations, ruleReservation{
+			cfg:         cfg,
+			reservation: reservation,
+		})
+	}
+
+	// If no rules matched, allow the request.
+	if len(reservations) == 0 {
+		return true, 0
+	}
+
+	// Check if all reservations can proceed immediately. If any rule
+	// denies, we must cancel ALL reservations to avoid consuming tokens
+	// unfairly.
+	var maxWait time.Duration
+	allAllowed := true
+
+	for _, rr := range reservations {
+		if !rr.reservation.OK() {
+			// Rate is zero or infinity.
+			allAllowed = false
+			maxWait = time.Second
+
+			break
+		}
+
+		delay := rr.reservation.Delay()
+		if delay > 0 {
+			allAllowed = false
+			if delay > maxWait {
+				maxWait = delay
+			}
+		}
+	}
+
+	// If any rule denied, cancel all reservations and return denied.
+	if !allAllowed {
+		for _, rr := range reservations {
+			rr.reservation.Cancel()
+			rateLimitDenied.WithLabelValues(
+				rl.serviceName, rr.cfg.PathRegexp,
+			).Inc()
+		}
+
+		return false, maxWait
+	}
+
+	// All rules allowed - tokens are consumed, record metrics.
+	for _, rr := range reservations {
+		rateLimitAllowed.WithLabelValues(
+			rl.serviceName, rr.cfg.PathRegexp,
+		).Inc()
+	}
+
+	return true, 0
+}
+
+// getOrCreateLimiter retrieves an existing limiter or creates a new one.
+func (rl *RateLimiter) getOrCreateLimiter(key limiterKey,
+	cfg *RateLimitConfig) *rate.Limiter {
+
+	rl.cacheMu.Lock()
+	defer rl.cacheMu.Unlock()
+
+	// Try to get existing entry from cache (also updates LRU order).
+	if entry, err := rl.cache.Get(key); err == nil {
+		return entry.limiter
+	}
+
+	// Create a new limiter.
+	limiter := rate.NewLimiter(
+		rate.Limit(cfg.Rate()), cfg.EffectiveBurst(),
+	)
+
+	entry := &limiterEntry{
+		limiter: limiter,
+	}
+
+	// Put handles eviction automatically when cache is full.
+	evicted, _ := rl.cache.Put(key, entry)
+	if evicted {
+		rateLimitEvictions.WithLabelValues(rl.serviceName).Inc()
+	}
+
+	rateLimitCacheSize.WithLabelValues(rl.serviceName).Set(
+		float64(rl.cache.Len()),
+	)
+
+	return limiter
+}
+
+// Size returns the current number of entries in the cache.
+func (rl *RateLimiter) Size() int {
+	rl.cacheMu.Lock()
+	defer rl.cacheMu.Unlock()
+
+	return rl.cache.Len()
+}
+
+// ExtractRateLimitKey extracts the rate-limiting key from a request.
+// For authenticated requests, it uses the L402 token ID. For unauthenticated
+// requests, it falls back to the client IP address.
+//
+// IMPORTANT: The authenticated parameter should only be true if the L402 token
+// has been validated by the authenticator. Using unvalidated L402 tokens as
+// keys is a DoS vector since attackers can flood the cache with garbage tokens.
+func ExtractRateLimitKey(r *http.Request, remoteIP net.IP,
+	authenticated bool) string {
+
+	// Only use L402 token ID if the request has been authenticated.
+	// This prevents DoS attacks where garbage L402 tokens flood the cache.
+	if authenticated {
+		mac, _, err := l402.FromHeader(&r.Header)
+		if err == nil && mac != nil {
+			identifier, err := l402.DecodeIdentifier(
+				bytes.NewBuffer(mac.Id()),
+			)
+			if err == nil {
+				return "token:" + identifier.TokenID.String()
+			}
+		}
+	}
+
+	// Fall back to IP address for unauthenticated requests.
+	// Mask the IP to group clients from the same network segment.
+	return "ip:" + netutil.MaskIP(remoteIP).String()
+}
diff --git a/proxy/ratelimiter_test.go b/proxy/ratelimiter_test.go
new file mode 100644
index 0000000..a445fe7
--- /dev/null
+++ b/proxy/ratelimiter_test.go
@@ -0,0 +1,436 @@
+package proxy
+
+import (
+	"fmt"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"regexp"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+)
+
+// TestRateLimiterBasic tests basic rate limiting functionality.
+func TestRateLimiterBasic(t *testing.T) {
+	cfg := &RateLimitConfig{
+		PathRegexp: "^/api/.*$",
+		Requests:   10,
+		Per:        time.Second,
+		Burst:      10,
+	}
+	cfg.compiledPathRegexp = regexp.MustCompile(cfg.PathRegexp)
+
+	rl := NewRateLimiter("test-service", []*RateLimitConfig{cfg})
+
+	// First 10 requests should be allowed.
+	for i := 0; i < 10; i++ {
+		req := httptest.NewRequest("GET", "/api/test", nil)
+		allowed, _ := rl.Allow(req, "test-key")
+		require.True(t, allowed, "request %d should be allowed", i)
+	}
+
+	// 11th request should be denied.
+	req := httptest.NewRequest("GET", "/api/test", nil)
+	allowed, retryAfter := rl.Allow(req, "test-key")
+	require.False(t, allowed, "11th request should be denied")
+	require.Greater(t, retryAfter, time.Duration(0))
+}
+
+// TestRateLimiterNoMatchingRules tests that requests pass when no rules match.
+func TestRateLimiterNoMatchingRules(t *testing.T) {
+	cfg := &RateLimitConfig{
+		PathRegexp: "^/api/.*$",
+		Requests:   1,
+		Per:        time.Hour,
+		Burst:      1,
+	}
+	cfg.compiledPathRegexp = regexp.MustCompile(cfg.PathRegexp)
+
+	rl := NewRateLimiter("test-service", []*RateLimitConfig{cfg})
+
+	// Request to non-matching path should always be allowed.
+	for i := 0; i < 100; i++ {
+		req := httptest.NewRequest("GET", "/other/path", nil)
+		allowed, _ := rl.Allow(req, "test-key")
+		require.True(t, allowed, "non-matching request should be allowed")
+	}
+}
+
+// TestRateLimiterLRUEviction tests that the LRU cache evicts old entries.
+func TestRateLimiterLRUEviction(t *testing.T) {
+	cfg := &RateLimitConfig{
+		Requests: 100,
+		Per:      time.Second,
+		Burst:    100,
+	}
+
+	rl := NewRateLimiter(
+		"test-service", []*RateLimitConfig{cfg},
+		WithMaxCacheSize(5),
+	)
+
+	// Create 10 different keys.
+	for i := 0; i < 10; i++ {
+		req := httptest.NewRequest("GET", "/api/test", nil)
+		key := fmt.Sprintf("key-%d", i)
+		rl.Allow(req, key)
+	}
+
+	// Cache should be at max size.
+	require.Equal(t, 5, rl.Size())
+}
+
+// TestRateLimiterPathMatching tests that different path patterns have
+// independent limits.
+func TestRateLimiterPathMatching(t *testing.T) {
+	cfgApi := &RateLimitConfig{
+		PathRegexp: "^/api/.*$",
+		Requests:   5,
+		Per:        time.Second,
+		Burst:      5,
+	}
+	cfgApi.compiledPathRegexp = regexp.MustCompile(cfgApi.PathRegexp)
+
+	cfgAdmin := &RateLimitConfig{
+		PathRegexp: "^/admin/.*$",
+		Requests:   2,
+		Per:        time.Second,
+		Burst:      2,
+	}
+	cfgAdmin.compiledPathRegexp = regexp.MustCompile(cfgAdmin.PathRegexp)
+
+	rl := NewRateLimiter(
+		"test-service",
+		[]*RateLimitConfig{cfgApi, cfgAdmin},
+	)
+
+	// API path should allow 5 requests.
+	for i := 0; i < 5; i++ {
+		req := httptest.NewRequest("GET", "/api/users", nil)
+		allowed, _ := rl.Allow(req, "test-key")
+		require.True(t, allowed)
+	}
+
+	// Admin path should allow 2 requests.
+	for i := 0; i < 2; i++ {
+		req := httptest.NewRequest("GET", "/admin/settings", nil)
+		allowed, _ := rl.Allow(req, "test-key")
+		require.True(t, allowed)
+	}
+
+	// Next admin request should be denied.
+	req := httptest.NewRequest("GET", "/admin/settings", nil)
+	allowed, _ := rl.Allow(req, "test-key")
+	require.False(t, allowed)
+
+	// API should still have capacity (used 5, burst is 5, but we're testing
+	// a 6th).
+	req = httptest.NewRequest("GET", "/api/users", nil)
+	allowed, _ = rl.Allow(req, "test-key")
+	require.False(t, allowed, "6th API request should be denied")
+}
+
+// TestRateLimiterMultipleRulesAllMustPass tests that all matching rules must
+// pass for a request to be allowed.
+func TestRateLimiterMultipleRulesAllMustPass(t *testing.T) {
+	// Global rule: 100 req/sec.
+	cfgGlobal := &RateLimitConfig{
+		Requests: 100,
+		Per:      time.Second,
+		Burst:    100,
+	}
+
+	// Specific rule: 2 req/sec for /expensive.
+	cfgExpensive := &RateLimitConfig{
+		PathRegexp: "^/expensive$",
+		Requests:   2,
+		Per:        time.Second,
+		Burst:      2,
+	}
+	cfgExpensive.compiledPathRegexp = regexp.MustCompile(cfgExpensive.PathRegexp)
+
+	rl := NewRateLimiter(
+		"test-service",
+		[]*RateLimitConfig{cfgGlobal, cfgExpensive},
+	)
+
+	// Expensive should be limited by the stricter rule.
+	for i := 0; i < 2; i++ {
+		req := httptest.NewRequest("GET", "/expensive", nil)
+		allowed, _ := rl.Allow(req, "test-key")
+		require.True(t, allowed)
+	}
+
+	req := httptest.NewRequest("GET", "/expensive", nil)
+	allowed, _ := rl.Allow(req, "test-key")
+	require.False(t, allowed, "should be denied by /expensive rule")
+}
+
+// TestRateLimiterPerKeyIsolation tests that different keys have independent
+// rate limits.
+func TestRateLimiterPerKeyIsolation(t *testing.T) {
+	cfg := &RateLimitConfig{
+		Requests: 2,
+		Per:      time.Second,
+		Burst:    2,
+	}
+
+	rl := NewRateLimiter("test-service", []*RateLimitConfig{cfg})
+
+	// User 1 uses their quota.
+	for i := 0; i < 2; i++ {
+		req := httptest.NewRequest("GET", "/api/test", nil)
+		allowed, _ := rl.Allow(req, "user-1")
+		require.True(t, allowed)
+	}
+
+	// User 1 is now denied.
+	req := httptest.NewRequest("GET", "/api/test", nil)
+	allowed, _ := rl.Allow(req, "user-1")
+	require.False(t, allowed)
+
+	// User 2 should still have full quota.
+	for i := 0; i < 2; i++ {
+		req := httptest.NewRequest("GET", "/api/test", nil)
+		allowed, _ := rl.Allow(req, "user-2")
+		require.True(t, allowed)
+	}
+}
+
+// TestExtractRateLimitKeyIP tests IP-based key extraction for unauthenticated
+// requests.
+func TestExtractRateLimitKeyIP(t *testing.T) {
+	req := httptest.NewRequest("GET", "/api/test", nil)
+	ip := net.ParseIP("192.168.1.100")
+
+	// Unauthenticated request should use masked IP (/24 for IPv4).
+	key := ExtractRateLimitKey(req, ip, false)
+	require.Equal(t, "ip:192.168.1.0", key)
+}
+
+// TestExtractRateLimitKeyIPv6 tests IPv6 key extraction.
+func TestExtractRateLimitKeyIPv6(t *testing.T) {
+	req := httptest.NewRequest("GET", "/api/test", nil)
+	ip := net.ParseIP("2001:db8:1234:5678::1")
+
+	// IPv6 should be masked to /48.
+	key := ExtractRateLimitKey(req, ip, false)
+	require.Equal(t, "ip:2001:db8:1234::", key)
+}
+
+// TestExtractRateLimitKeyUnauthenticatedIgnoresL402 tests that unauthenticated
+// requests fall back to IP even if L402 header is present. This prevents DoS
+// attacks where garbage L402 tokens flood the cache.
+func TestExtractRateLimitKeyUnauthenticatedIgnoresL402(t *testing.T) {
+	req := httptest.NewRequest("GET", "/api/test", nil)
+	// Add a garbage L402 header that would be present before authentication.
+	req.Header.Set("Authorization", "L402 garbage:token")
+	ip := net.ParseIP("192.168.1.100")
+
+	// Even with L402 header present, unauthenticated=false should use
+	// masked IP.
+	key := ExtractRateLimitKey(req, ip, false)
+	require.Equal(t, "ip:192.168.1.0", key)
+}
+
+// TestRateLimitConfigRate tests the Rate() calculation.
+func TestRateLimitConfigRate(t *testing.T) {
+	tests := []struct {
+		name     string
+		requests int
+		per      time.Duration
+		wantRate float64
+	}{
+		{
+			name:     "10 per second",
+			requests: 10,
+			per:      time.Second,
+			wantRate: 10.0,
+		},
+		{
+			name:     "60 per minute",
+			requests: 60,
+			per:      time.Minute,
+			wantRate: 1.0,
+		},
+		{
+			name:     "1 per hour",
+			requests: 1,
+			per:      time.Hour,
+			wantRate: 1.0 / 3600.0,
+		},
+		{
+			name:     "zero per",
+			requests: 10,
+			per:      0,
+			wantRate: 0,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := &RateLimitConfig{
+				Requests: tt.requests,
+				Per:      tt.per,
+			}
+			require.InDelta(t, tt.wantRate, cfg.Rate(), 0.0001)
+		})
+	}
+}
+
+// TestRateLimitConfigEffectiveBurst tests the EffectiveBurst() calculation.
+func TestRateLimitConfigEffectiveBurst(t *testing.T) {
+	tests := []struct {
+		name      string
+		requests  int
+		burst     int
+		wantBurst int
+	}{
+		{
+			name:      "explicit burst",
+			requests:  10,
+			burst:     20,
+			wantBurst: 20,
+		},
+		{
+			name:      "default to requests",
+			requests:  10,
+			burst:     0,
+			wantBurst: 10,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := &RateLimitConfig{
+				Requests: tt.requests,
+				Burst:    tt.burst,
+			}
+			require.Equal(t, tt.wantBurst, cfg.EffectiveBurst())
+		})
+	}
+}
+
+// TestRateLimitConfigMatches tests the Matches() method.
+func TestRateLimitConfigMatches(t *testing.T) {
+	tests := []struct {
+		name       string
+		pathRegexp string
+		path       string
+		want       bool
+	}{
+		{
+			name:       "no pattern matches all",
+			pathRegexp: "",
+			path:       "/anything",
+			want:       true,
+		},
+		{
+			name:       "pattern matches",
+			pathRegexp: "^/api/.*$",
+			path:       "/api/users",
+			want:       true,
+		},
+		{
+			name:       "pattern does not match",
+			pathRegexp: "^/api/.*$",
+			path:       "/admin/users",
+			want:       false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := &RateLimitConfig{
+				PathRegexp: tt.pathRegexp,
+			}
+			if tt.pathRegexp != "" {
+				cfg.compiledPathRegexp = regexp.MustCompile(
+					tt.pathRegexp,
+				)
+			}
+			require.Equal(t, tt.want, cfg.Matches(tt.path))
+		})
+	}
+}
+
+// TestSendRateLimitResponseHTTP tests HTTP rate limit response.
+func TestSendRateLimitResponseHTTP(t *testing.T) {
+	w := httptest.NewRecorder()
+	req := httptest.NewRequest("GET", "/api/test", nil)
+
+	sendRateLimitResponse(w, req, 5*time.Second)
+
+	require.Equal(t, http.StatusTooManyRequests, w.Code)
+	require.Equal(t, "5", w.Header().Get("Retry-After"))
+	require.Contains(t, w.Body.String(), "rate limit exceeded")
+}
+
+// TestSendRateLimitResponseHTTPSubSecond tests that sub-second delays are
+// rounded up to 1 second.
+func TestSendRateLimitResponseHTTPSubSecond(t *testing.T) {
+	w := httptest.NewRecorder()
+	req := httptest.NewRequest("GET", "/api/test", nil)
+
+	sendRateLimitResponse(w, req, 500*time.Millisecond)
+
+	require.Equal(t, http.StatusTooManyRequests, w.Code)
+	require.Equal(t, "1", w.Header().Get("Retry-After"))
+}
+
+// TestSendRateLimitResponseHTTPRoundUp tests that fractional seconds are
+// rounded up, not down. This ensures clients don't retry before the limit
+// actually resets.
+func TestSendRateLimitResponseHTTPRoundUp(t *testing.T) {
+	w := httptest.NewRecorder()
+	req := httptest.NewRequest("GET", "/api/test", nil)
+
+	// 1.1 seconds should round up to 2 seconds, not down to 1.
+	sendRateLimitResponse(w, req, 1100*time.Millisecond)
+
+	require.Equal(t, http.StatusTooManyRequests, w.Code)
+	require.Equal(t, "2", w.Header().Get("Retry-After"))
+}
+
+// TestSendRateLimitResponseGRPC tests gRPC rate limit response.
+func TestSendRateLimitResponseGRPC(t *testing.T) {
+	w := httptest.NewRecorder()
+	req := httptest.NewRequest("POST", "/grpc.Service/Method", nil)
+	req.Header.Set("Content-Type", "application/grpc")
+
+	sendRateLimitResponse(w, req, 5*time.Second)
+
+	require.Equal(t, http.StatusOK, w.Code) // gRPC always returns 200.
+	require.Equal(t, "5", w.Header().Get("Retry-After"))
+	require.Equal(t, "8", w.Header().Get("Grpc-Status")) // ResourceExhausted.
+	require.Equal(t, "rate limit exceeded", w.Header().Get("Grpc-Message"))
+}
+
+// TestRateLimiterTokenRefill tests that tokens refill over time.
+func TestRateLimiterTokenRefill(t *testing.T) {
+	cfg := &RateLimitConfig{
+		Requests: 10,
+		Per:      100 * time.Millisecond, // Fast refill for testing.
+		Burst:    1,
+	}
+
+	rl := NewRateLimiter("test-service", []*RateLimitConfig{cfg})
+
+	// Use the one available token.
+	req := httptest.NewRequest("GET", "/api/test", nil)
+	allowed, _ := rl.Allow(req, "test-key")
+	require.True(t, allowed)
+
+	// Immediate second request should be denied.
+	allowed, _ = rl.Allow(req, "test-key")
+	require.False(t, allowed)
+
+	// Wait for refill.
+	time.Sleep(15 * time.Millisecond)
+
+	// Should have a token now.
+	allowed, _ = rl.Allow(req, "test-key")
+	require.True(t, allowed)
+}
diff --git a/proxy/service.go b/proxy/service.go
index 8f58d03..d2a952d 100644
--- a/proxy/service.go
+++ b/proxy/service.go
@@ -110,6 +110,12 @@ type Service struct {
 	// request, but still try to do the l402 authentication.
 	AuthSkipInvoiceCreationPaths []string `long:"authskipinvoicecreationpaths" description:"List of regular expressions for paths that will skip invoice creation'"`
 
+	// RateLimits is an optional list of rate-limiting rules for this
+	// service. Each rule specifies a path pattern and rate limit
+	// parameters. All matching rules are evaluated; if any rule denies
+	// the request, it is rejected.
+	RateLimits []*RateLimitConfig `long:"ratelimits" description:"List of rate limiting rules for this service"`
+
 	// compiledHostRegexp is the compiled host regex.
 	compiledHostRegexp *regexp.Regexp
 
@@ -123,8 +129,9 @@ type Service struct {
 	// invoice creation paths.
 	compiledAuthSkipInvoiceCreationPaths []*regexp.Regexp
 
-	freebieDB freebie.DB
-	pricer    pricer.Pricer
+	freebieDB   freebie.DB
+	pricer      pricer.Pricer
+	rateLimiter *RateLimiter
 }
 
 // ResourceName returns the string to be used to identify which resource a
@@ -275,6 +282,47 @@ func prepareServices(services []*Service) error {
 			)
 		}
 
+		// Validate and compile rate limit configurations.
+		if len(service.RateLimits) > 0 {
+			for i, rl := range service.RateLimits {
+				// Validate required fields.
+				if rl.Requests <= 0 {
+					return fmt.Errorf("service %s rate "+
+						"limit %d: requests must be "+
+						"positive", service.Name, i)
+				}
+				if rl.Per <= 0 {
+					return fmt.Errorf("service %s rate "+
+						"limit %d: per duration must "+
+						"be positive", service.Name, i)
+				}
+
+				// Compile path regex if provided.
+				if rl.PathRegexp != "" {
+					compiled, err := regexp.Compile(
+						rl.PathRegexp,
+					)
+					if err != nil {
+						return fmt.Errorf("service %s "+
+							"rate limit %d: error "+
+							"compiling path regex: "+
+							"%w", service.Name, i,
+							err)
+					}
+					rl.compiledPathRegexp = compiled
+				}
+			}
+
+			// Create the rate limiter for this service.
+			service.rateLimiter = NewRateLimiter(
+				service.Name, service.RateLimits,
+			)
+
+			log.Infof("Initialized rate limiter for service %s "+
+				"with %d rules", service.Name,
+				len(service.RateLimits))
+		}
+
 		// If dynamic prices are enabled then use the provided
 		// DynamicPrice options to initialise a gRPC backed
 		// pricer client.
diff --git a/sample-conf.yaml b/sample-conf.yaml
index aaa247d..f3f4ac2 100644
--- a/sample-conf.yaml
+++ b/sample-conf.yaml
@@ -195,6 +195,28 @@ services:
     authskipinvoicecreationpaths:
       - '^/streamingservice.*$'
 
+    # Optional per-endpoint rate limits using a token bucket algorithm.
+    # Rate limiting is applied per L402 token ID (or IP address for
+    # unauthenticated requests). All matching rules are evaluated; if any
+    # rule denies the request, it is rejected.
+    ratelimits:
+        # Rate limit for general API endpoints.
+      - pathregexp: '^/looprpc.SwapServer/LoopOutTerms.*$'
+        # Number of requests allowed per time window. Must be provided and
+        # positive.
+        requests: 5
+        # Time window duration (e.g., 1s, 1m, 1h).  Must be provided and
+        # positive.
+        per: 1s
+        # Maximum burst capacity. Must be positive if provided.
+        burst: 100
+
+        # Stricter rate limit for quote endpoints.
+      - pathregexp: '^/looprpc.SwapServer/LoopOutQuote.*$'
+        requests: 2
+        per: 1s
+        burst: 2
+
     # Options to use for connection to the price serving gRPC server.
     dynamicprice:
       # Whether or not a gRPC server is available to query price data from. If