diff --git a/app/services/velorum/models_service.rb b/app/services/velorum/models_service.rb
index 6dad34fa..6d314d60 100644
--- a/app/services/velorum/models_service.rb
+++ b/app/services/velorum/models_service.rb
@@ -2,6 +2,8 @@
 
 module Velorum
   class ModelsService
+    include Code0::ZeroTrack::Loggable
+
     def initialize(client: nil, config: Sagittarius::Configuration.config[:velorum])
       @client = client
       @config = config
@@ -11,6 +13,13 @@ def execute
       return [] unless config[:enabled]
 
       client.models.models
+    rescue GRPC::BadStatus => e
+      logger.warn(
+        message: 'Failed to fetch Velorum models',
+        grpc_code: e.respond_to?(:code) ? e.code : nil,
+        grpc_details: e.respond_to?(:details) ? e.details : e.message
+      )
+      []
     end
 
     private
diff --git a/spec/requests/graphql/query/ai_query_spec.rb b/spec/requests/graphql/query/ai_query_spec.rb
index 4d91c483..ef05b662 100644
--- a/spec/requests/graphql/query/ai_query_spec.rb
+++ b/spec/requests/graphql/query/ai_query_spec.rb
@@ -85,4 +85,20 @@
       expect(Sagittarius::Velorum::Client).not_to have_received(:new)
     end
   end
+
+  context 'when Velorum is unreachable' do
+    before do
+      allow(client).to receive(:models).and_raise(
+        GRPC::BadStatus.new_status_exception(GRPC::Core::StatusCodes::UNAVAILABLE, 'Connection refused')
+      )
+    end
+
+    it 'returns enabled state and an empty model list without GraphQL errors' do
+      post_graphql(query, current_user: current_user)
+
+      expect(graphql_data_at(:ai, :enabled)).to be(true)
+      expect(graphql_data_at(:ai, :models)).to eq([])
+      expect(graphql_errors).to be_nil
+    end
+  end
 end
diff --git a/spec/services/velorum/models_service_spec.rb b/spec/services/velorum/models_service_spec.rb
new file mode 100644
index 00000000..978adf9b
--- /dev/null
+++ b/spec/services/velorum/models_service_spec.rb
@@ -0,0 +1,46 @@
+# frozen_string_literal: true
+
+require 'rails_helper'
+
+RSpec.describe Velorum::ModelsService do
+  subject(:models) { described_class.new(client: client, config: config).execute }
+
+  let(:client) { instance_double(Sagittarius::Velorum::Client) }
+  let(:config) { { enabled: true } }
+  let(:models_response) do
+    Tucana::Velorum::ModelsResponse.new(
+      models: [
+        Tucana::Velorum::Model.new(identifier: 'gpt-5', name: 'GPT-5')
+      ]
+    )
+  end
+
+  before do
+    allow(client).to receive(:models).and_return(models_response)
+  end
+
+  it 'returns models from Velorum' do
+    expect(models).to eq(models_response.models)
+  end
+
+  context 'when Velorum is disabled' do
+    let(:config) { { enabled: false } }
+
+    it 'returns an empty list without calling Velorum' do
+      expect(models).to eq([])
+      expect(client).not_to have_received(:models)
+    end
+  end
+
+  context 'when Velorum is unreachable' do
+    before do
+      allow(client).to receive(:models).and_raise(
+        GRPC::BadStatus.new_status_exception(GRPC::Core::StatusCodes::UNAVAILABLE, 'Connection refused')
+      )
+    end
+
+    it 'returns an empty list' do
+      expect(models).to eq([])
+    end
+  end
+end