diff --git a/.jules/bolt.md b/.jules/bolt.md
index 91ce70b..8495dac 100644
--- a/.jules/bolt.md
+++ b/.jules/bolt.md
@@ -1,5 +1,9 @@
 # Bolt's Journal
 
+## 2026-03-09 - GCP ID Token Credentials Caching & Proxy Overhead
+**Learning:** Invoking Google Cloud credential discovery (`google.FindDefaultCredentials`) inside `GetIDToken` for every forwarded request via the local reverse proxy introduces massive disk I/O and configuration parsing latency (~300ms per request). Standardizing thread-safe lazy credentials caching via `sync.Mutex` ensures subsequent ID token generation requests are nearly instantaneous because they reuse the cached credentials' thread-safe `TokenSource` which handles in-memory token caching and refreshes automatically.
+**Action:** Always identify and cache underlying `google.Credentials` and `TokenSource` configurations across performance-critical request paths instead of repeatedly invoking credential discovery.
+
 ## 2026-03-08 - GCP Logging Client Caching & Connection Longevity
 **Learning:** Establishing the GCP Stackdriver Logging client requires repeated Google credential discovery and connection establishment, causing high latency (~300ms) inside a reactive TUI interface. Caching `logadmin.Client` instances via a project-aware map with thread-safe `sync.Mutex` ensures subsequent streaming and log extraction operations are instantaneous. Crucially, calling `Close()` on individual stream terminations must be a no-op to prevent premature teardown of connection pools shared across other active streaming views.
 **Action:** Keep GCP Logging clients cached globally by project and handle connection termination via a no-op `Close` method, while adding test-isolation resets in unit tests.
diff --git a/internal/run/auth/auth.go b/internal/run/auth/auth.go
index 6f8a3db..8989a01 100644
--- a/internal/run/auth/auth.go
+++ b/internal/run/auth/auth.go
@@ -24,6 +24,7 @@ import (
 	"os/user"
 	"path/filepath"
 	"strings"
+	"sync"
 
 	api_region "github.com/JulienBreux/run-cli/internal/run/api/region"
 	"github.com/JulienBreux/run-cli/internal/run/model/common/info"
@@ -130,12 +131,30 @@ func parseConfig(path string) (info.Info, error) {
 	}, nil
 }
 
+var (
+	credsMu                sync.Mutex
+	cachedCreds            *google.Credentials
+	findDefaultCredentials = google.FindDefaultCredentials
+)
+
 // GetIDToken retrieves an identity token for the given audience using Google Cloud credentials.
+// It caches the credentials structure in a thread-safe manner to prevent repetitive credential discovery
+// disk read and metadata server lookup overhead (~300ms latency) on every subsequent call.
 var GetIDToken = func(ctx context.Context) (string, error) {
-	creds, err := google.FindDefaultCredentials(ctx, scopes...)
-	if err != nil {
-		return "", fmt.Errorf("failed to find default credentials: %w", err)
+	credsMu.Lock()
+	if cachedCreds == nil {
+		// Use a background context to ensure credential discovery is not canceled
+		// if a request-scoped context is canceled.
+		bgCtx := context.Background()
+		creds, err := findDefaultCredentials(bgCtx, scopes...)
+		if err != nil {
+			credsMu.Unlock()
+			return "", fmt.Errorf("failed to find default credentials: %w", err)
+		}
+		cachedCreds = creds
 	}
+	creds := cachedCreds
+	credsMu.Unlock()
 
 	token, err := creds.TokenSource.Token()
 	if err != nil {
diff --git a/internal/run/auth/auth_test.go b/internal/run/auth/auth_test.go
index 6dcffa4..4023cf2 100644
--- a/internal/run/auth/auth_test.go
+++ b/internal/run/auth/auth_test.go
@@ -17,11 +17,14 @@ limitations under the License.
 package auth
 
 import (
+	"context"
 	"os"
 	"path/filepath"
 	"testing"
 
 	api_region "github.com/JulienBreux/run-cli/internal/run/api/region"
+	"golang.org/x/oauth2"
+	"golang.org/x/oauth2/google"
 )
 
 func TestGetInfo(t *testing.T) {
@@ -133,3 +136,64 @@ func TestGetInfo_Defaults(t *testing.T) {
 		t.Errorf("Expected default Region 'all', got '%s'", info.Region)
 	}
 }
+
+type mockTokenSource struct {
+	token *oauth2.Token
+}
+
+func (m *mockTokenSource) Token() (*oauth2.Token, error) {
+	return m.token, nil
+}
+
+func TestGetIDToken_Caching(t *testing.T) {
+	// Reset/save global variables
+	origFindDefaultCredentials := findDefaultCredentials
+	origCachedCreds := cachedCreds
+	defer func() {
+		findDefaultCredentials = origFindDefaultCredentials
+		cachedCreds = origCachedCreds
+	}()
+
+	// Mock token source and extra fields
+	baseToken := &oauth2.Token{
+		AccessToken: "mock-access-token",
+	}
+	mockToken := baseToken.WithExtra(map[string]interface{}{
+		"id_token": "mock-id-token",
+	})
+	mockTS := &mockTokenSource{
+		token: mockToken,
+	}
+
+	callCount := 0
+	findDefaultCredentials = func(ctx context.Context, scopes ...string) (*google.Credentials, error) {
+		callCount++
+		return &google.Credentials{
+			TokenSource: mockTS,
+		}, nil
+	}
+
+	// First call (cache miss)
+	cachedCreds = nil // Ensure cache is empty
+	token, err := GetIDToken(context.Background())
+	if err != nil {
+		t.Fatalf("GetIDToken failed: %v", err)
+	}
+	if token != "mock-id-token" {
+		t.Errorf("Expected token 'mock-id-token', got '%s'", token)
+	}
+
+	// Second call (cache hit)
+	token, err = GetIDToken(context.Background())
+	if err != nil {
+		t.Fatalf("GetIDToken second call failed: %v", err)
+	}
+	if token != "mock-id-token" {
+		t.Errorf("Expected token 'mock-id-token', got '%s'", token)
+	}
+
+	// Verify discovery was only called once
+	if callCount != 1 {
+		t.Errorf("Expected findDefaultCredentials to be called exactly 1 time, called %d times", callCount)
+	}
+}