diff --git a/apps/ios/Sources/Voice/TalkModeGatewayConfig.swift b/apps/ios/Sources/Voice/TalkModeGatewayConfig.swift
index 03d0e918c114..ab0eedb25b6d 100644
--- a/apps/ios/Sources/Voice/TalkModeGatewayConfig.swift
+++ b/apps/ios/Sources/Voice/TalkModeGatewayConfig.swift
@@ -401,13 +401,18 @@ enum TalkModeGatewayConfigParser {
         let realtimeVoiceId = Self.firstString(realtime, keys: ["voice"])
             ?? Self.firstString(realtimeProviderConfig, keys: ["voice"])
         let realtimeTransport = Self.firstString(realtime, keys: ["transport"])?.lowercased()
-        let requiresGatewayRealtimeTransport = realtimeTransport == "gateway-relay"
+        // Direct provider WebRTC can answer before consulting the agent, so this explicit
+        // policy must stay on the relay that enforces final-transcript consultations.
+        let requiresForcedAgentConsultRelay = Self.requiresForcedAgentConsultRelay(realtime)
+        let requiresGatewayRealtimeTransport = requiresForcedAgentConsultRelay
+            || realtimeTransport == "gateway-relay"
             || realtimeTransport == "provider-websocket"
             || Self.usesAzureOpenAI(provider: realtimeProvider, config: realtimeProviderConfig)
         let openAIProviderConfig = Self.realtimeProviderConfig(
             providers: realtimeProviders,
             provider: "openai")
-        let openAIRequiresGatewayRealtimeTransport = realtimeTransport == "gateway-relay"
+        let openAIRequiresGatewayRealtimeTransport = requiresForcedAgentConsultRelay
+            || realtimeTransport == "gateway-relay"
             || realtimeTransport == "provider-websocket"
             || Self.usesAzureOpenAI(provider: "openai", config: openAIProviderConfig)
         let executionMode = Self.resolvedExecutionMode(
@@ -450,6 +455,10 @@ enum TalkModeGatewayConfigParser {
         return nil
     }
 
+    private static func requiresForcedAgentConsultRelay(_ realtime: [String: AnyCodable]?) -> Bool {
+        self.firstString(realtime, keys: ["consultRouting"])?.lowercased() == "force-agent-consult"
+    }
+
     private static func resolvedExecutionMode(
         _ realtime: [String: AnyCodable]?,
         requiresGatewayRealtimeTransport: Bool) -> TalkModeExecutionMode
diff --git a/apps/ios/Tests/TalkModeConfigParsingTests.swift b/apps/ios/Tests/TalkModeConfigParsingTests.swift
index 5f463f6d4268..97252991c167 100644
--- a/apps/ios/Tests/TalkModeConfigParsingTests.swift
+++ b/apps/ios/Tests/TalkModeConfigParsingTests.swift
@@ -635,8 +635,78 @@ struct TalkModeManagerTests {
             defaultModelIdFallback: "eleven_v3",
             defaultRealtimeModelIdFallback: "gpt-realtime-2",
             defaultSilenceTimeoutMs: 900)
+        let routing = TalkModeRoutingResolver.resolve(
+            parsed: parsed,
+            providerSelection: .openAIRealtime,
+            defaultProvider: "elevenlabs",
+            defaultRealtimeModelId: "gpt-realtime-2")
 
         #expect(parsed.executionMode == .realtimeWebRTC)
+        #expect(routing.route == .realtimeWebRTC)
+    }
+
+    @Test func `routes forced agent consultation through gateway relay`() {
+        let config: [String: Any] = [
+            "talk": [
+                "realtime": [
+                    "provider": "openai",
+                    "mode": "realtime",
+                    "brain": "agent-consult",
+                    "consultRouting": "force-agent-consult",
+                ],
+            ],
+        ]
+
+        let parsed = TalkModeGatewayConfigParser.parse(
+            config: config,
+            defaultProvider: "elevenlabs",
+            defaultModelIdFallback: "eleven_v3",
+            defaultRealtimeModelIdFallback: "gpt-realtime-2",
+            defaultSilenceTimeoutMs: 900)
+        let gatewayDefaultRouting = TalkModeRoutingResolver.resolve(
+            parsed: parsed,
+            providerSelection: .gatewayDefault,
+            defaultProvider: "elevenlabs",
+            defaultRealtimeModelId: "gpt-realtime-2")
+        let openAIRouting = TalkModeRoutingResolver.resolve(
+            parsed: parsed,
+            providerSelection: .openAIRealtime,
+            defaultProvider: "elevenlabs",
+            defaultRealtimeModelId: "gpt-realtime-2")
+
+        #expect(parsed.requiresGatewayRealtimeTransport)
+        #expect(parsed.openAIRequiresGatewayRealtimeTransport)
+        #expect(parsed.executionMode == .realtimeRelay)
+        #expect(gatewayDefaultRouting.route == .realtimeRelay)
+        #expect(openAIRouting.route == .realtimeRelay)
+    }
+
+    @Test func `routes forced agent consultation through gateway relay without mode or brain`() {
+        let config: [String: Any] = [
+            "talk": [
+                "realtime": [
+                    "provider": "openai",
+                    "consultRouting": "force-agent-consult",
+                ],
+            ],
+        ]
+
+        let parsed = TalkModeGatewayConfigParser.parse(
+            config: config,
+            defaultProvider: "elevenlabs",
+            defaultModelIdFallback: "eleven_v3",
+            defaultRealtimeModelIdFallback: "gpt-realtime-2",
+            defaultSilenceTimeoutMs: 900)
+        let routing = TalkModeRoutingResolver.resolve(
+            parsed: parsed,
+            providerSelection: .openAIRealtime,
+            defaultProvider: "elevenlabs",
+            defaultRealtimeModelId: "gpt-realtime-2")
+
+        #expect(parsed.requiresGatewayRealtimeTransport)
+        #expect(parsed.openAIRequiresGatewayRealtimeTransport)
+        #expect(routing.executionMode == .realtimeRelay)
+        #expect(routing.route == .realtimeRelay)
     }
 
     @Test func `keeps Azure open AI realtime on gateway relay`() {