@ouro.bot/cli 0.1.0-alpha.582 → 0.1.0-alpha.583

package/changelog.json

@@ -1,6 +1,17 @@
 {
   "_note": "This changelog is maintained as part of the PR/version-bump workflow. Agent-curated, not auto-generated. Agents read this file directly via read_file to understand what changed between versions.",
   "versions": [
+    {
+      "version": "0.1.0-alpha.583",
+      "changes": [
+        "Outbound SIP phone calls now start the Realtime greeting immediately after answer unless Twilio has already positively identified voicemail or fax, preventing humans from hearing post-pickup silence when async AMD returns unknown.",
+        "Twilio phone voice now defaults outbound calls to OpenAI Realtime Media Streams when inbound calls use OpenAI SIP on a Media Stream machine, while still allowing `voice.twilioOutboundConversationEngine` overrides, so humans avoid post-pickup SIP ringback.",
+        "Realtime voice now resolves phone callers through the canonical friend graph, preferring existing friend ids and otherwise matching normalized phone numbers via `imessage-handle`, so trust-aware tools see the same friend context as text and mail.",
+        "Realtime media-stream voice now treats empty caller metadata as absent and preserves local voice friend identities, keeping outbound and provider-simulated calls attached to the intended friend instead of inventing a blank phone identity.",
+        "Realtime voice response creation now backs off and retries after provider active-response conflicts, holds user turns under Ouro floor-control instead of provider auto-response, and long-running voice tools can emit one tiny holding phrase instead of leaving seconds of unexplained silence.",
+        "Realtime voice VAD and local barge-in thresholds are less twitchy by default, reducing accidental interruption from tiny room sounds while preserving deliberate caller interruption."
+      ]
+    },
     {
       "version": "0.1.0-alpha.582",
       "changes": [

package/dist/senses/voice/twilio-phone-runtime.js

@@ -154,13 +154,37 @@ function resolveOpenAIRealtimeApiKey(options) {
         return { apiKey: compatKey, source: "integrations.openaiEmbeddingsApiKey" };
     return undefined;
 }
-function configuredConversationEngine(options, overrides) {
-    return overrides.conversationEngine
-        ?? (0, twilio_phone_1.normalizeTwilioPhoneConversationEngine)(configString(options.machineConfig, "voice.twilioConversationEngine")
-            ?? configString(options.machineConfig, "voice.conversationEngine")
-            ?? configString(options.runtimeConfig, "voice.twilioConversationEngine")
-            ?? configString(options.runtimeConfig, "voice.conversationEngine")
-            ?? "cascade");
+function configuredConversationEngine(options, overrides, transportMode) {
+    const explicit = overrides.conversationEngine
+        ?? configString(options.machineConfig, "voice.twilioConversationEngine")
+        ?? configString(options.machineConfig, "voice.conversationEngine")
+        ?? configString(options.runtimeConfig, "voice.twilioConversationEngine")
+        ?? configString(options.runtimeConfig, "voice.conversationEngine");
+    const hasSipConfig = !!(configString(options.runtimeConfig, "voice.openaiSipProjectId")
+        || configString(options.machineConfig, "voice.openaiSipProjectId"));
+    const explicitEngine = explicit ? (0, twilio_phone_1.normalizeTwilioPhoneConversationEngine)(explicit) : undefined;
+    if (hasSipConfig && (!explicitEngine || explicitEngine === "cascade"))
+        return "openai-sip";
+    if (explicitEngine)
+        return explicitEngine;
+    const hasRealtimeConfig = !!resolveOpenAIRealtimeApiKey({ runtimeConfig: options.runtimeConfig, overrides });
+    if (hasRealtimeConfig && transportMode === "media-stream")
+        return "openai-realtime";
+    return "cascade";
+}
+function configuredOutboundConversationEngine(options, overrides, conversationEngine, transportMode) {
+    const defaultOutboundEngine = conversationEngine === "openai-sip" && transportMode === "media-stream"
+        ? "openai-realtime"
+        : conversationEngine;
+    const configured = overrides.outboundConversationEngine
+        ?? (0, twilio_phone_1.normalizeTwilioPhoneConversationEngine)(configString(options.machineConfig, "voice.twilioOutboundConversationEngine")
+            ?? configString(options.machineConfig, "voice.outboundConversationEngine")
+            ?? configString(options.runtimeConfig, "voice.twilioOutboundConversationEngine")
+            ?? configString(options.runtimeConfig, "voice.outboundConversationEngine")
+            ?? defaultOutboundEngine);
+    if (defaultOutboundEngine === "openai-realtime" && configured === "cascade")
+        return defaultOutboundEngine;
+    return configured;
 }
 function normalizeOpenAIRealtimeReasoningEffort(value) {
     const normalized = value?.trim().toLowerCase();
@@ -226,7 +250,14 @@ function resolveTwilioPhoneTransportRuntime(options) {
         ?? twilio_phone_1.TWILIO_PHONE_WEBHOOK_BASE_PATH);
     const transportMode = overrides.transportMode
         ?? (0, twilio_phone_1.normalizeTwilioPhoneTransportMode)(configString(options.machineConfig, "voice.twilioTransportMode") ?? twilio_phone_1.DEFAULT_TWILIO_PHONE_TRANSPORT_MODE);
-    const conversationEngine = configuredConversationEngine(options, overrides);
+    const conversationEngine = configuredConversationEngine(options, overrides, transportMode);
+    const outboundConversationEngine = configuredOutboundConversationEngine(options, overrides, conversationEngine, transportMode);
+    const needsOpenAIRealtime = conversationEngine === "openai-realtime"
+        || conversationEngine === "openai-sip"
+        || outboundConversationEngine === "openai-realtime"
+        || outboundConversationEngine === "openai-sip";
+    const needsOpenAISip = conversationEngine === "openai-sip" || outboundConversationEngine === "openai-sip";
+    const needsCascade = conversationEngine === "cascade" || outboundConversationEngine === "cascade";
     let elevenLabsApiKey = configString(options.runtimeConfig, "integrations.elevenLabsApiKey") ?? "";
     let elevenLabsVoiceId = trimOptional(overrides.elevenLabsVoiceId)
         ?? configString(options.runtimeConfig, "integrations.elevenLabsVoiceId")
@@ -240,9 +271,9 @@ function resolveTwilioPhoneTransportRuntime(options) {
         ?? "";
     let openaiRealtime;
     let openaiSip;
-    if (conversationEngine === "openai-realtime" || conversationEngine === "openai-sip") {
-        if (conversationEngine === "openai-realtime" && transportMode !== "media-stream") {
-            throw new Error("voice.twilioConversationEngine=openai-realtime requires voice.twilioTransportMode=media-stream");
+    if (needsOpenAIRealtime) {
+        if ((conversationEngine === "openai-realtime" || outboundConversationEngine === "openai-realtime") && transportMode !== "media-stream") {
+            throw new Error("voice.twilioConversationEngine/openai-realtime requires voice.twilioTransportMode=media-stream");
         }
         const key = resolveOpenAIRealtimeApiKey({ runtimeConfig: options.runtimeConfig, overrides });
         if (!key) {
@@ -300,7 +331,7 @@ function resolveTwilioPhoneTransportRuntime(options) {
                 ?? normalizeOpenAIRealtimeNoiseReduction(configString(options.runtimeConfig, "voice.openaiRealtimeNoiseReduction")),
             turnDetection,
         };
-        if (conversationEngine === "openai-sip") {
+        if (needsOpenAISip) {
             const projectId = trimOptional(overrides.openaiSipProjectId)
                 ?? configString(options.runtimeConfig, "voice.openaiSipProjectId")
                 ?? configString(options.machineConfig, "voice.openaiSipProjectId");
@@ -334,7 +365,7 @@ function resolveTwilioPhoneTransportRuntime(options) {
             };
         }
     }
-    else {
+    if (needsCascade) {
         elevenLabsApiKey = required(elevenLabsApiKey || undefined, "missing integrations.elevenLabsApiKey; run 'ouro connect voice --agent <agent>' for setup guidance");
         elevenLabsVoiceId = required(elevenLabsVoiceId || undefined, "missing integrations.elevenLabsVoiceId; save the ElevenLabs voice ID before starting phone voice");
         whisperCliPath = required(whisperCliPath || undefined, "missing voice.whisperCliPath in this machine's runtime config");
@@ -379,6 +410,7 @@ function resolveTwilioPhoneTransportRuntime(options) {
             ?? (0, twilio_phone_1.normalizeTwilioPhonePlaybackMode)(configString(options.machineConfig, "voice.twilioPlaybackMode") ?? twilio_phone_1.DEFAULT_TWILIO_PHONE_PLAYBACK_MODE),
         transportMode,
         conversationEngine,
+        outboundConversationEngine,
         openaiRealtime,
         openaiSip,
         openaiSipWebhookUrl: openaiSip?.webhookPath ? (0, twilio_phone_1.openAISipWebhookUrl)(publicBaseUrl, openaiSip.webhookPath) : undefined,
@@ -482,7 +514,12 @@ async function startConfiguredTwilioPhoneTransport(options, deps = defaultTwilio
             meta: { agentName: settings.agentName, source: settings.openaiRealtime.apiKeySource },
         });
     }
-    const transcriber = settings.conversationEngine === "openai-realtime" || settings.conversationEngine === "openai-sip"
+    const settingsNeedsOpenAIRealtime = settings.conversationEngine === "openai-realtime"
+        || settings.conversationEngine === "openai-sip"
+        || settings.outboundConversationEngine === "openai-realtime"
+        || settings.outboundConversationEngine === "openai-sip";
+    const settingsNeedsCascade = settings.conversationEngine === "cascade" || settings.outboundConversationEngine === "cascade";
+    const transcriber = settingsNeedsOpenAIRealtime && !settingsNeedsCascade
         ? {
             transcribe: async () => {
                 throw new Error("OpenAI Realtime voice sessions do not use the cascade transcriber");
@@ -492,7 +529,7 @@ async function startConfiguredTwilioPhoneTransport(options, deps = defaultTwilio
             whisperCliPath: settings.whisperCliPath,
             modelPath: settings.whisperModelPath,
         });
-    const tts = settings.conversationEngine === "openai-realtime" || settings.conversationEngine === "openai-sip"
+    const tts = settingsNeedsOpenAIRealtime && !settingsNeedsCascade
         ? {
             synthesize: async () => {
                 throw new Error("OpenAI Realtime voice sessions do not use the cascade TTS service");
@@ -522,6 +559,7 @@ async function startConfiguredTwilioPhoneTransport(options, deps = defaultTwilio
         transportMode: settings.transportMode,
         playbackMode: settings.playbackMode,
         conversationEngine: settings.conversationEngine,
+        outboundConversationEngine: settings.outboundConversationEngine,
         openaiRealtime: settings.openaiRealtime,
         openaiSip: settings.openaiSip,
     });
@@ -538,6 +576,7 @@ async function startConfiguredTwilioPhoneTransport(options, deps = defaultTwilio
             openaiSipWebhookUrl: settings.openaiSipWebhookUrl ?? "",
             transportMode: settings.transportMode,
             conversationEngine: settings.conversationEngine,
+            outboundConversationEngine: settings.outboundConversationEngine,
             openaiRealtimeModel: settings.openaiRealtime?.model ?? "",
         },
     });
@@ -565,7 +604,7 @@ async function prewarmOutboundGreeting(options, deps) {
     if (options.settings.transportMode !== "media-stream")
         return undefined;
     /* v8 ignore next -- Realtime/SIP outbound tests assert no cascade prewarm is attempted @preserve */
-    if (options.settings.conversationEngine === "openai-realtime" || options.settings.conversationEngine === "openai-sip")
+    if (options.settings.outboundConversationEngine === "openai-realtime" || options.settings.outboundConversationEngine === "openai-sip")
         return undefined;
     const friendId = options.friendId?.trim() || `twilio-${safeRuntimeSegment(options.to)}`;
     const sessionKey = (0, twilio_phone_1.twilioPhoneVoiceSessionKey)({
@@ -677,7 +716,7 @@ async function placeConfiguredTwilioPhoneCall(options, deps = defaultTwilioPhone
         reason: options.reason.trim(),
         ...(options.initialAudio ? { initialAudio: options.initialAudio } : {}),
         createdAt,
-        status: settings.transportMode === "media-stream" && settings.conversationEngine !== "openai-realtime" && settings.conversationEngine !== "openai-sip"
+        status: settings.transportMode === "media-stream" && settings.outboundConversationEngine === "cascade"
             ? "prewarming"
             : "requested",
     });

package/dist/senses/voice/twilio-phone.js

@@ -205,6 +205,15 @@ function usesOpenAIRealtimeConversationEngine(options) {
 function usesOpenAISipConversationEngine(options) {
     return normalizeTwilioPhoneConversationEngine(options.conversationEngine) === "openai-sip";
 }
+function outboundConversationEngine(options) {
+    return normalizeTwilioPhoneConversationEngine(options.outboundConversationEngine ?? options.conversationEngine);
+}
+function usesOpenAIRealtimeOutboundConversationEngine(options) {
+    return outboundConversationEngine(options) === "openai-realtime";
+}
+function usesOpenAISipOutboundConversationEngine(options) {
+    return outboundConversationEngine(options) === "openai-sip";
+}
 function twilioPhoneWebhookUrl(publicBaseUrl, basePath = exports.TWILIO_PHONE_WEBHOOK_BASE_PATH) {
     return routeUrl(publicBaseUrl, `${normalizeTwilioPhoneBasePath(basePath)}/incoming`);
 }
@@ -253,8 +262,9 @@ function websocketRouteUrl(publicBaseUrl, route) {
     url.protocol = "wss:";
     return url.toString();
 }
-function mediaStreamTwiml(options, basePath, params, greetingJobId, customParams = {}) {
-    const streamUrl = websocketRouteUrl(options.publicBaseUrl, `${basePath}/media-stream`);
+function mediaStreamTwiml(options, basePath, params, greetingJobId, customParams = {}, streamEngine) {
+    const streamRoute = streamEngine ? `${basePath}/media-stream?engine=${encodeURIComponent(streamEngine)}` : `${basePath}/media-stream`;
+    const streamUrl = websocketRouteUrl(options.publicBaseUrl, streamRoute);
     const twimlParams = {
         From: params.From,
         To: params.To,
@@ -286,7 +296,7 @@ function openAISipUri(options, customHeaders = {}) {
     return `sip:${projectId}@sip.api.openai.com;transport=tls${query ? `?${query}` : ""}`;
 }
 function openAISipDialTwiml(options, customHeaders = {}) {
-    return `<Dial answerOnBridge="true"><Sip>${escapeXml(openAISipUri(options, customHeaders))}</Sip></Dial>`;
+    return `<Dial><Sip>${escapeXml(openAISipUri(options, customHeaders))}</Sip></Dial>`;
 }
 /* v8 ignore stop */
 function safeSegment(input) {
@@ -346,6 +356,32 @@ function friendIdFromCaller(from, callSid) {
 function voiceFriendId(options, from, callSid) {
     return options.defaultFriendId?.trim() || friendIdFromCaller(from, callSid);
 }
+async function resolveVoiceFriendContext(options, input) {
+    const agentRoot = options.agentRoot ?? (0, identity_1.getAgentRoot)(options.agentName);
+    const friendStore = new store_file_1.FileFriendStore(path.join(agentRoot, "friends"));
+    const explicitFriendId = input.friendId?.trim();
+    if (explicitFriendId) {
+        const existing = await friendStore.get(explicitFriendId);
+        if (existing) {
+            return {
+                friendId: existing.id,
+                friendStore,
+                resolved: { friend: existing, channel: (0, channel_1.getChannelCapabilities)("voice") },
+            };
+        }
+    }
+    const remotePhone = (0, phone_1.normalizeTwilioE164PhoneNumber)(input.remotePhone);
+    const provider = remotePhone ? "imessage-handle" : "local";
+    const externalId = remotePhone || explicitFriendId || voiceFriendId(options, input.remotePhone ?? "", input.callSid);
+    const resolver = new resolver_1.FriendResolver(friendStore, {
+        provider,
+        externalId,
+        displayName: remotePhone || externalId,
+        channel: "voice",
+    });
+    const resolved = await resolver.resolve();
+    return { friendId: resolved.friend.id, friendStore, resolved };
+}
 function phoneIdentitySegment(input) {
     const phoneish = input.replace(/[^0-9A-Za-z]+/g, "");
     return phoneish || safeSegment(input);
@@ -1119,16 +1155,19 @@ const OPENAI_REALTIME_DEFAULT_TRANSCRIPTION_MODEL = "gpt-realtime-whisper";
 const OPENAI_REALTIME_BOOTSTRAP_TIMEOUT_MS = 250;
 const OPENAI_REALTIME_PCMS_BYTES_PER_MS = 8;
 const OPENAI_REALTIME_DEFAULT_NOISE_REDUCTION = "near_field";
-const OPENAI_REALTIME_DEFAULT_VAD_THRESHOLD = 0.68;
-const OPENAI_REALTIME_DEFAULT_VAD_PREFIX_PADDING_MS = 220;
-const OPENAI_REALTIME_DEFAULT_VAD_SILENCE_DURATION_MS = 320;
-const OPENAI_REALTIME_DEFAULT_VAD_IDLE_TIMEOUT_MS = 15_000;
+const OPENAI_REALTIME_DEFAULT_VAD_THRESHOLD = 0.78;
+const OPENAI_REALTIME_DEFAULT_VAD_PREFIX_PADDING_MS = 300;
+const OPENAI_REALTIME_DEFAULT_VAD_SILENCE_DURATION_MS = 650;
+const OPENAI_REALTIME_DEFAULT_VAD_IDLE_TIMEOUT_MS = 7_000;
 const OPENAI_REALTIME_MAX_OUTPUT_TOKENS = 220;
-const OPENAI_REALTIME_BARGE_IN_MIN_SPEECH_MS = 160;
-const OPENAI_REALTIME_BARGE_IN_RMS_THRESHOLD = 900;
+const OPENAI_REALTIME_BARGE_IN_MIN_SPEECH_MS = 260;
+const OPENAI_REALTIME_BARGE_IN_RMS_THRESHOLD = 1_300;
 const OPENAI_REALTIME_MIN_VOICE_SPEED = 0.25;
 const OPENAI_REALTIME_MAX_VOICE_SPEED = 1.5;
 const OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS = 50;
+const OPENAI_REALTIME_RESPONSE_CREATE_CONFLICT_BACKOFF_MS = 1_000;
+const OPENAI_REALTIME_TOOL_PRESENCE_DELAY_MS = 900;
+const OPENAI_REALTIME_USER_TURN_RESPONSE_DELAY_MS = 700;
 const OPENAI_SIP_OUTBOUND_AMD_GREETING_TIMEOUT_MS = 10_000;
 const OPENAI_SIP_UNSUPPORTED_TOOL_NAMES = new Set();
 const OPENAI_SIP_DEFAULT_API_BASE_URL = "https://api.openai.com/v1";
@@ -1337,6 +1376,18 @@ function realtimeToolsFromChatTools(tools, excludedToolNames = new Set()) {
         parameters: tool.function.parameters ?? { type: "object", properties: {} },
     }));
 }
+function mediaStreamRequestedConversationEngine(url) {
+    if (!url)
+        return undefined;
+    try {
+        const parsed = new URL(url, "wss://localhost");
+        const engine = parsed.searchParams.get("engine") ?? undefined;
+        return engine ? normalizeTwilioPhoneConversationEngine(engine) : undefined;
+    }
+    catch {
+        return undefined;
+    }
+}
 function parseToolArguments(raw) {
     if (!raw.trim())
         return {};
@@ -1406,12 +1457,14 @@ async function buildRealtimeVoiceInstructions(options) {
     return [
         `You are ${options.agentName} in the live Voice sense.`,
         "This is the same agent identity as every other Ouro surface. Voice is not a reduced or alternate self.",
+        options.friend ? `Resolved voice friend: ${options.friend.name || options.friend.id} (friendId=${options.friend.id}, trust=${options.friend.trustLevel ?? "friend"}, role=${options.friend.role ?? "friend"}). Use this same friend record and trust context for relationship awareness and tool permissions across voice, text, mail, and every other sense.` : "",
         `Current native Realtime provider config for this call: model=${options.realtimeModel?.trim() || OPENAI_REALTIME_DEFAULT_MODEL}, voice=${options.realtimeVoice?.trim() || OPENAI_REALTIME_DEFAULT_VOICE}${options.realtimeVoiceSpeed === undefined ? "" : `, speed=${options.realtimeVoiceSpeed}`}.`,
         options.realtimeVoiceStyle?.trim()
             ? `Phone voice target: ${options.realtimeVoiceStyle.trim()}`
             : "",
         "Speak as yourself through live audio. Follow voice/style preferences from identity notes; do not say you lack identity, preferences, or agency because the provider voice is configured by the transport.",
         "Audio is synchronous. Default to one short sentence. Use two short sentences only when needed. Do not use markdown, lists, or long explanations unless the caller explicitly asks.",
+        "Do not treat every tiny silence as your turn. Let the caller finish the thought, especially if they pause mid-sentence.",
         "If the caller interrupts, stop the older path and answer the newest thing first.",
         "If the caller says they are counting, measuring latency, testing lag, waiting, or wants you quiet, say at most 'got it' and then stay silent until they ask or say something that needs an answer.",
         "Use tools for outside facts or side effects. While a tool is running, give at most one tiny preamble, then summarize the result compactly when it returns.",
@@ -1491,6 +1544,8 @@ class TwilioOpenAIRealtimeMediaStreamSession {
     pendingAudioPayloads = [];
     openaiWs = null;
     toolContext;
+    friendStore;
+    resolvedContext;
     sessionMessages = [];
     playbackState;
     playbackMarkIndex = 0;
@@ -1498,8 +1553,12 @@ class TwilioOpenAIRealtimeMediaStreamSession {
     toolResponses = new Map();
     completedRealtimeResponseIds = new Set();
     activeRealtimeResponseId = null;
+    realtimeResponseCreateInFlight = null;
+    untrackedActiveRealtimeResponse = false;
+    untrackedActiveRealtimeResponseTimer = null;
     pendingRealtimeResponse = null;
     pendingRealtimeResponseTimer = null;
+    pendingUserTurnResponseTimer = null;
     responseCreateHoldUntilMs = 0;
     initialAudio;
     initialAudioPlayed = false;
@@ -1574,9 +1633,16 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             this.from = customParameter(start, "From");
             this.to = customParameter(start, "To");
         }
-        this.friendId = explicitFriendId || voiceFriendId(this.options, this.from, this.callSid);
+        const voiceContext = await resolveVoiceFriendContext(this.options, {
+            friendId: explicitFriendId,
+            remotePhone: this.from || undefined,
+            callSid: this.callSid,
+        });
+        this.friendId = voiceContext.friendId;
+        this.friendStore = voiceContext.friendStore;
+        this.resolvedContext = voiceContext.resolved;
         this.sessionKey = twilioPhoneVoiceSessionKey({
-            defaultFriendId: explicitFriendId || this.options.defaultFriendId,
+            defaultFriendId: this.friendId,
             from: this.from,
             to: this.to,
             callSid: this.callSid,
@@ -1699,7 +1765,7 @@ class TwilioOpenAIRealtimeMediaStreamSession {
                         format: { type: "audio/pcmu" },
                         noise_reduction: realtimeNoiseReductionConfig(realtime),
                         transcription: { model: OPENAI_REALTIME_DEFAULT_TRANSCRIPTION_MODEL },
-                        turn_detection: realtimeTurnDetectionConfig(realtime),
+                        turn_detection: realtimeTurnDetectionConfig(realtime, { createResponse: false, interruptResponse: false }),
                     },
                     output: realtimeOutputAudioConfig(realtime, { type: "audio/pcmu" }),
                 },
@@ -1721,6 +1787,7 @@ class TwilioOpenAIRealtimeMediaStreamSession {
         const realtimeSystem = await buildRealtimeVoiceInstructions({
             agentName: this.options.agentName,
             agentRoot,
+            friend: this.resolvedContext?.friend,
             priorTranscript: prior,
             realtimeVoice: this.options.openaiRealtime?.voice,
             realtimeVoiceStyle: this.options.openaiRealtime?.voiceStyle,
@@ -1743,6 +1810,8 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             return;
         this.toolContext = {
             signin: async () => undefined,
+            ...(this.resolvedContext ? { context: this.resolvedContext } : {}),
+            ...(this.friendStore ? { friendStore: this.friendStore } : {}),
             voiceCall: {
                 requestEnd: () => this.requestHangupFromTool(),
                 playAudio: (request) => this.playPreparedAudio(request),
@@ -1750,16 +1819,18 @@ class TwilioOpenAIRealtimeMediaStreamSession {
         };
     }
     async buildRealtimeTools() {
-        const agentRoot = this.options.agentRoot ?? (0, identity_1.getAgentRoot)(this.options.agentName);
-        const friendsPath = path.join(agentRoot, "friends");
-        const friendStore = new store_file_1.FileFriendStore(friendsPath);
-        const resolver = new resolver_1.FriendResolver(friendStore, {
-            provider: "local",
-            externalId: this.friendId,
-            displayName: this.friendId,
-            channel: "voice",
-        });
-        const resolved = await resolver.resolve();
+        if (!this.resolvedContext || !this.friendStore) {
+            const voiceContext = await resolveVoiceFriendContext(this.options, {
+                friendId: this.friendId,
+                remotePhone: this.from || undefined,
+                callSid: this.callSid,
+            });
+            this.friendId = voiceContext.friendId;
+            this.friendStore = voiceContext.friendStore;
+            this.resolvedContext = voiceContext.resolved;
+        }
+        const resolved = this.resolvedContext;
+        const friendStore = this.friendStore;
         this.toolContext = {
             signin: async () => undefined,
             context: resolved,
@@ -1875,7 +1946,7 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             return;
         }
         if (type === "conversation.item.input_audio_transcription.completed" && typeof event.transcript === "string") {
-            this.appendTranscript("user", event.transcript);
+            this.handleUserTranscript(event.transcript);
             return;
         }
         if (type === "response.output_audio_transcript.done" && typeof event.transcript === "string") {
@@ -1896,6 +1967,7 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             return;
         }
         if (type === "error") {
+            this.handleRealtimeError(event);
             (0, runtime_1.emitNervesEvent)({
                 level: "error",
                 component: "senses",
@@ -1905,6 +1977,40 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             });
         }
     }
+    handleRealtimeError(event) {
+        const error = event.error;
+        if (!error || typeof error !== "object" || Array.isArray(error))
+            return;
+        const code = stringField(error.code);
+        if (code !== "conversation_already_has_active_response")
+            return;
+        this.noteRealtimeResponseConflict();
+    }
+    handleUserTranscript(transcript) {
+        const content = transcript.trim();
+        if (!content)
+            return;
+        this.appendTranscript("user", content);
+        this.scheduleUserTurnResponse();
+    }
+    scheduleUserTurnResponse() {
+        if (this.closed)
+            return;
+        this.clearPendingUserTurnResponse();
+        this.pendingUserTurnResponseTimer = setTimeout(() => {
+            this.pendingUserTurnResponseTimer = null;
+            if (this.closed)
+                return;
+            this.requestRealtimeResponse();
+        }, OPENAI_REALTIME_USER_TURN_RESPONSE_DELAY_MS);
+        this.pendingUserTurnResponseTimer.unref?.();
+    }
+    clearPendingUserTurnResponse() {
+        if (!this.pendingUserTurnResponseTimer)
+            return;
+        clearTimeout(this.pendingUserTurnResponseTimer);
+        this.pendingUserTurnResponseTimer = null;
+    }
     handleOpenAIAudioDelta(event) {
         const payload = stringField(event.delta);
         if (!payload)
@@ -1926,6 +2032,7 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             this.sendTwilioMark({ itemId, contentIndex, audioEndMs });
     }
     handleCallerSpeechStarted() {
+        this.clearPendingUserTurnResponse();
         const playback = this.playbackState;
         if (!this.hasReliableCallerBargeInSpeech()) {
             (0, runtime_1.emitNervesEvent)({
@@ -1971,6 +2078,8 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             responseDone: this.completedRealtimeResponseIds.has(responseId),
             followupRequested: false,
             suppressFollowup: false,
+            presenceRequested: false,
+            presenceTimer: null,
         };
         state.pendingCallIds.add(callId);
         if (!existing)
@@ -1984,6 +2093,8 @@ class TwilioOpenAIRealtimeMediaStreamSession {
         if (!state)
             return false;
         state.pendingCallIds.delete(callId);
+        if (state.pendingCallIds.size === 0)
+            this.clearRealtimeToolPresenceTimer(state);
         return this.maybeCreateRealtimeToolFollowup(responseId, state);
     }
     completeRealtimeToolResponse(responseId) {
@@ -2002,11 +2113,37 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             return false;
         state.followupRequested = true;
         this.toolResponses.delete(responseId);
+        this.clearRealtimeToolPresenceTimer(state);
         if (state.suppressFollowup)
             return true;
         this.requestRealtimeResponse();
         return true;
     }
+    scheduleRealtimeToolPresence(responseId, state) {
+        if (!responseId || state.presenceRequested || state.presenceTimer)
+            return;
+        state.presenceTimer = setTimeout(() => {
+            state.presenceTimer = null;
+            const current = this.toolResponses.get(responseId);
+            if (this.closed || current !== state || state.pendingCallIds.size === 0 || state.suppressFollowup)
+                return;
+            state.presenceRequested = true;
+            this.requestRealtimeResponse({
+                instructions: "A tool is taking a moment. Say one very short natural holding phrase under six words, then stop speaking.",
+            });
+        }, OPENAI_REALTIME_TOOL_PRESENCE_DELAY_MS);
+        state.presenceTimer.unref?.();
+    }
+    clearRealtimeToolPresenceTimer(state) {
+        if (!state.presenceTimer)
+            return;
+        clearTimeout(state.presenceTimer);
+        state.presenceTimer = null;
+    }
+    clearRealtimeToolPresenceTimers() {
+        for (const state of this.toolResponses.values())
+            this.clearRealtimeToolPresenceTimer(state);
+    }
     async runRealtimeTool(event) {
         const name = typeof event.name === "string" ? event.name : "";
         const callId = typeof event.call_id === "string" ? event.call_id : "";
@@ -2017,6 +2154,8 @@ class TwilioOpenAIRealtimeMediaStreamSession {
         const coordinated = !!toolState;
         if (name === "voice_end_call" && toolState)
             toolState.suppressFollowup = true;
+        if (toolState && !toolState.suppressFollowup)
+            this.scheduleRealtimeToolPresence(responseId, toolState);
         let output;
         try {
             const args = parseToolArguments(typeof event.arguments === "string" ? event.arguments : "");
@@ -2057,14 +2196,18 @@ class TwilioOpenAIRealtimeMediaStreamSession {
         }
     }
     noteRealtimeResponseCreated(event) {
+        this.realtimeResponseCreateInFlight = null;
+        this.untrackedActiveRealtimeResponse = false;
+        this.clearUntrackedActiveRealtimeResponseTimer();
         const responseId = realtimeResponseId(event);
         if (responseId)
             this.activeRealtimeResponseId = responseId;
     }
-    noteRealtimeResponseDone(responseId) {
-        if (!responseId || this.activeRealtimeResponseId === responseId) {
-            this.activeRealtimeResponseId = null;
-        }
+    noteRealtimeResponseDone(_responseId) {
+        this.realtimeResponseCreateInFlight = null;
+        this.untrackedActiveRealtimeResponse = false;
+        this.clearUntrackedActiveRealtimeResponseTimer();
+        this.activeRealtimeResponseId = null;
         this.responseCreateHoldUntilMs = Math.max(this.responseCreateHoldUntilMs, Date.now() + OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS);
         this.schedulePendingRealtimeResponse(OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS);
     }
@@ -2072,15 +2215,21 @@ class TwilioOpenAIRealtimeMediaStreamSession {
         if (this.closed)
             return;
         const waitMs = Math.max(0, this.responseCreateHoldUntilMs - Date.now());
-        if (this.activeRealtimeResponseId || waitMs > 0) {
-            const pendingResponse = response ?? this.pendingRealtimeResponse?.response;
-            this.pendingRealtimeResponse = pendingResponse ? { response: pendingResponse } : {};
-            if (!this.activeRealtimeResponseId)
+        if (this.realtimeResponseIsBusy() || waitMs > 0) {
+            this.holdRealtimeResponse(response ? { response } : {});
+            if (!this.realtimeResponseIsBusy())
                 this.schedulePendingRealtimeResponse(waitMs);
             return;
         }
         this.sendRealtimeResponseCreate(response ? { response } : {});
     }
+    realtimeResponseIsBusy() {
+        return !!this.activeRealtimeResponseId || !!this.realtimeResponseCreateInFlight || this.untrackedActiveRealtimeResponse;
+    }
+    holdRealtimeResponse(request) {
+        const pendingResponse = request.response ?? this.pendingRealtimeResponse?.response;
+        this.pendingRealtimeResponse = pendingResponse ? { response: pendingResponse } : {};
+    }
     schedulePendingRealtimeResponse(delayMs) {
         if (!this.pendingRealtimeResponse)
             return;
@@ -2093,7 +2242,7 @@ class TwilioOpenAIRealtimeMediaStreamSession {
         this.pendingRealtimeResponseTimer.unref?.();
     }
     flushPendingRealtimeResponse() {
-        if (!this.pendingRealtimeResponse || this.closed || this.activeRealtimeResponseId)
+        if (!this.pendingRealtimeResponse || this.closed || this.realtimeResponseIsBusy())
             return;
         const waitMs = Math.max(0, this.responseCreateHoldUntilMs - Date.now());
         if (waitMs > 0) {
@@ -2105,11 +2254,38 @@ class TwilioOpenAIRealtimeMediaStreamSession {
         this.sendRealtimeResponseCreate(pending);
     }
     sendRealtimeResponseCreate(request) {
+        this.realtimeResponseCreateInFlight = request;
         this.sendOpenAI({
             type: "response.create",
             ...(request.response ? { response: request.response } : {}),
         });
     }
+    noteRealtimeResponseConflict() {
+        const inFlight = this.realtimeResponseCreateInFlight;
+        this.realtimeResponseCreateInFlight = null;
+        this.untrackedActiveRealtimeResponse = true;
+        if (inFlight)
+            this.holdRealtimeResponse(inFlight);
+        this.scheduleUntrackedActiveRealtimeResponseFallback();
+    }
+    scheduleUntrackedActiveRealtimeResponseFallback() {
+        this.clearUntrackedActiveRealtimeResponseTimer();
+        this.untrackedActiveRealtimeResponseTimer = setTimeout(() => {
+            this.untrackedActiveRealtimeResponseTimer = null;
+            if (this.closed || !this.untrackedActiveRealtimeResponse)
+                return;
+            this.untrackedActiveRealtimeResponse = false;
+            this.responseCreateHoldUntilMs = Math.max(this.responseCreateHoldUntilMs, Date.now() + OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS);
+            this.schedulePendingRealtimeResponse(OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS);
+        }, OPENAI_REALTIME_RESPONSE_CREATE_CONFLICT_BACKOFF_MS);
+        this.untrackedActiveRealtimeResponseTimer.unref?.();
+    }
+    clearUntrackedActiveRealtimeResponseTimer() {
+        if (!this.untrackedActiveRealtimeResponseTimer)
+            return;
+        clearTimeout(this.untrackedActiveRealtimeResponseTimer);
+        this.untrackedActiveRealtimeResponseTimer = null;
+    }
     flushPendingAudio() {
         const pending = this.pendingAudioPayloads.splice(0);
         for (const payload of pending) {
@@ -2229,6 +2405,9 @@ class TwilioOpenAIRealtimeMediaStreamSession {
             clearTimeout(this.pendingRealtimeResponseTimer);
             this.pendingRealtimeResponseTimer = null;
         }
+        this.clearPendingUserTurnResponse();
+        this.clearRealtimeToolPresenceTimers();
+        this.clearUntrackedActiveRealtimeResponseTimer();
         this.lifecycle?.onClose?.(this, { callSid: this.callSid, outboundId: this.outboundId });
         (0, runtime_1.emitNervesEvent)({
             component: "senses",
@@ -2257,12 +2436,18 @@ class OpenAISipPhoneSession {
     autoResponsesSuppressedForAmd = false;
     openaiWs = null;
     toolContext;
+    friendStore;
+    resolvedContext;
     sessionMessages = [];
     toolResponses = new Map();
     completedRealtimeResponseIds = new Set();
     activeRealtimeResponseId = null;
+    realtimeResponseCreateInFlight = null;
+    untrackedActiveRealtimeResponse = false;
+    untrackedActiveRealtimeResponseTimer = null;
     pendingRealtimeResponse = null;
     pendingRealtimeResponseTimer = null;
+    pendingUserTurnResponseTimer = null;
     responseCreateHoldUntilMs = 0;
     constructor(options, metadata, registry) {
         this.options = options;
@@ -2283,11 +2468,16 @@ class OpenAISipPhoneSession {
                 throw new Error("OpenAI Realtime API key is not configured");
             if (!sip)
                 throw new Error("OpenAI SIP options are not configured");
-            this.friendId = this.metadata.friendId
-                || this.options.defaultFriendId?.trim()
-                || voiceFriendId(this.options, this.metadata.from, this.metadata.callId);
+            const voiceContext = await resolveVoiceFriendContext(this.options, {
+                friendId: this.metadata.friendId || this.options.defaultFriendId?.trim(),
+                remotePhone: this.metadata.from || undefined,
+                callSid: this.metadata.callId,
+            });
+            this.friendId = voiceContext.friendId;
+            this.friendStore = voiceContext.friendStore;
+            this.resolvedContext = voiceContext.resolved;
             this.sessionKey = twilioPhoneVoiceSessionKey({
-                defaultFriendId: this.friendId || this.options.defaultFriendId,
+                defaultFriendId: this.friendId,
                 from: this.metadata.from,
                 to: this.metadata.to,
                 callSid: this.metadata.callId,
@@ -2329,7 +2519,7 @@ class OpenAISipPhoneSession {
             ];
             if (this.closed || this.outboundAmdStopped())
                 return;
-            await this.acceptOpenAISipCall(realtime, sip, instructions, tools, this.autoResponsesSuppressedForAmd);
+            await this.acceptOpenAISipCall(realtime, sip, instructions, tools);
             if (this.closed || this.outboundAmdStopped())
                 return;
             this.openControlWebSocket(realtime, sip, fullConfigPromise, usedBootstrap);
@@ -2370,14 +2560,14 @@ class OpenAISipPhoneSession {
         const answeredBy = job.answeredBy?.trim();
         if (nonHumanAnsweredStatus(answeredBy) || job.status === "voicemail" || job.status === "fax")
             return "reject";
-        if (answeredBy?.toLowerCase() === "human")
-            return "send";
-        return "hold";
+        // Silence after pickup feels broken. Start the greeting immediately unless
+        // Twilio has already positively identified a machine/fax answer.
+        return "send";
     }
     outboundAmdStopped() {
         return this.outboundAmdState === "nonhuman" || this.outboundAmdState === "timeout";
     }
-    async acceptOpenAISipCall(realtime, sip, instructions, tools, suppressAutoResponsesForAmd = false) {
+    async acceptOpenAISipCall(realtime, sip, instructions, tools) {
         const fetchImpl = sip.fetch ?? fetch;
         const response = await fetchImpl(openAISipCallActionUrl(sip, this.metadata.callId, "accept"), {
             method: "POST",
@@ -2393,7 +2583,7 @@ class OpenAISipPhoneSession {
                     input: {
                         noise_reduction: realtimeNoiseReductionConfig(realtime),
                         transcription: { model: OPENAI_REALTIME_DEFAULT_TRANSCRIPTION_MODEL },
-                        turn_detection: realtimeTurnDetectionConfig(realtime, suppressAutoResponsesForAmd ? { createResponse: false } : {}),
+                        turn_detection: realtimeTurnDetectionConfig(realtime, { createResponse: false, interruptResponse: false }),
                     },
                     output: realtimeOutputAudioConfig(realtime),
                 },
@@ -2509,6 +2699,7 @@ class OpenAISipPhoneSession {
         const realtimeSystem = await buildRealtimeVoiceInstructions({
             agentName: this.options.agentName,
             agentRoot,
+            friend: this.resolvedContext?.friend,
             priorTranscript: prior,
             realtimeVoice: this.options.openaiRealtime?.voice,
             realtimeVoiceStyle: this.options.openaiRealtime?.voiceStyle,
@@ -2528,6 +2719,8 @@ class OpenAISipPhoneSession {
             return;
         this.toolContext = {
             signin: async () => undefined,
+            ...(this.resolvedContext ? { context: this.resolvedContext } : {}),
+            ...(this.friendStore ? { friendStore: this.friendStore } : {}),
             voiceCall: {
                 requestEnd: () => this.requestHangupFromTool(),
                 playAudio: (request) => this.playRealtimeAudioCue(request),
@@ -2535,16 +2728,18 @@ class OpenAISipPhoneSession {
         };
     }
     async buildRealtimeTools() {
-        const agentRoot = this.options.agentRoot ?? (0, identity_1.getAgentRoot)(this.options.agentName);
-        const friendsPath = path.join(agentRoot, "friends");
-        const friendStore = new store_file_1.FileFriendStore(friendsPath);
-        const resolver = new resolver_1.FriendResolver(friendStore, {
-            provider: "local",
-            externalId: this.friendId,
-            displayName: this.friendId,
-            channel: "voice",
-        });
-        const resolved = await resolver.resolve();
+        if (!this.resolvedContext || !this.friendStore) {
+            const voiceContext = await resolveVoiceFriendContext(this.options, {
+                friendId: this.friendId,
+                remotePhone: this.metadata.from || undefined,
+                callSid: this.metadata.callId,
+            });
+            this.friendId = voiceContext.friendId;
+            this.friendStore = voiceContext.friendStore;
+            this.resolvedContext = voiceContext.resolved;
+        }
+        const resolved = this.resolvedContext;
+        const friendStore = this.friendStore;
         this.toolContext = {
             signin: async () => undefined,
             context: resolved,
@@ -2697,7 +2892,7 @@ class OpenAISipPhoneSession {
                         type: "realtime",
                         audio: {
                             input: {
-                                turn_detection: realtimeTurnDetectionConfig(realtime),
+                                turn_detection: realtimeTurnDetectionConfig(realtime, { createResponse: false, interruptResponse: false }),
                             },
                         },
                     },
@@ -2796,9 +2991,13 @@ class OpenAISipPhoneSession {
             this.noteRealtimeResponseCreated(event);
             return;
         }
+        if (type === "input_audio_buffer.speech_started") {
+            this.clearPendingUserTurnResponse();
+            return;
+        }
         if (type === "conversation.item.input_audio_transcription.completed" && typeof event.transcript === "string") {
             this.recordOutboundAmdTranscriptCandidate(event.transcript);
-            this.appendTranscript("user", event.transcript);
+            this.handleUserTranscript(event.transcript);
             return;
         }
         if (type === "response.output_audio_transcript.done" && typeof event.transcript === "string") {
@@ -2818,6 +3017,7 @@ class OpenAISipPhoneSession {
             return;
         }
         if (type === "error") {
+            this.handleRealtimeError(event);
             (0, runtime_1.emitNervesEvent)({
                 level: "error",
                 component: "senses",
@@ -2827,6 +3027,40 @@ class OpenAISipPhoneSession {
             });
         }
     }
+    handleRealtimeError(event) {
+        const error = event.error;
+        if (!error || typeof error !== "object" || Array.isArray(error))
+            return;
+        const code = stringField(error.code);
+        if (code !== "conversation_already_has_active_response")
+            return;
+        this.noteRealtimeResponseConflict();
+    }
+    handleUserTranscript(transcript) {
+        const content = transcript.trim();
+        if (!content)
+            return;
+        this.appendTranscript("user", content);
+        this.scheduleUserTurnResponse();
+    }
+    scheduleUserTurnResponse() {
+        if (this.closed)
+            return;
+        this.clearPendingUserTurnResponse();
+        this.pendingUserTurnResponseTimer = setTimeout(() => {
+            this.pendingUserTurnResponseTimer = null;
+            if (this.closed)
+                return;
+            this.requestRealtimeResponse();
+        }, OPENAI_REALTIME_USER_TURN_RESPONSE_DELAY_MS);
+        this.pendingUserTurnResponseTimer.unref?.();
+    }
+    clearPendingUserTurnResponse() {
+        if (!this.pendingUserTurnResponseTimer)
+            return;
+        clearTimeout(this.pendingUserTurnResponseTimer);
+        this.pendingUserTurnResponseTimer = null;
+    }
     registerRealtimeToolResponse(responseId, callId) {
         if (!responseId)
             return undefined;
@@ -2836,6 +3070,8 @@ class OpenAISipPhoneSession {
             responseDone: this.completedRealtimeResponseIds.has(responseId),
             followupRequested: false,
             suppressFollowup: false,
+            presenceRequested: false,
+            presenceTimer: null,
         };
         state.pendingCallIds.add(callId);
         if (!existing)
@@ -2849,6 +3085,8 @@ class OpenAISipPhoneSession {
         if (!state)
             return false;
         state.pendingCallIds.delete(callId);
+        if (state.pendingCallIds.size === 0)
+            this.clearRealtimeToolPresenceTimer(state);
         return this.maybeCreateRealtimeToolFollowup(responseId, state);
     }
     completeRealtimeToolResponse(responseId) {
@@ -2867,6 +3105,7 @@ class OpenAISipPhoneSession {
             return false;
         state.followupRequested = true;
         this.toolResponses.delete(responseId);
+        this.clearRealtimeToolPresenceTimer(state);
         if (state.suppressFollowup) {
             this.completeHangupIfReady("tool_response_done");
             return true;
@@ -2874,6 +3113,31 @@ class OpenAISipPhoneSession {
         this.requestRealtimeResponse();
         return true;
     }
+    scheduleRealtimeToolPresence(responseId, state) {
+        if (!responseId || state.presenceRequested || state.presenceTimer)
+            return;
+        state.presenceTimer = setTimeout(() => {
+            state.presenceTimer = null;
+            const current = this.toolResponses.get(responseId);
+            if (this.closed || current !== state || state.pendingCallIds.size === 0 || state.suppressFollowup)
+                return;
+            state.presenceRequested = true;
+            this.requestRealtimeResponse({
+                instructions: "A tool is taking a moment. Say one very short natural holding phrase under six words, then stop speaking.",
+            });
+        }, OPENAI_REALTIME_TOOL_PRESENCE_DELAY_MS);
+        state.presenceTimer.unref?.();
+    }
+    clearRealtimeToolPresenceTimer(state) {
+        if (!state.presenceTimer)
+            return;
+        clearTimeout(state.presenceTimer);
+        state.presenceTimer = null;
+    }
+    clearRealtimeToolPresenceTimers() {
+        for (const state of this.toolResponses.values())
+            this.clearRealtimeToolPresenceTimer(state);
+    }
     async runRealtimeTool(event) {
         const name = typeof event.name === "string" ? event.name : "";
         const callId = typeof event.call_id === "string" ? event.call_id : "";
@@ -2884,6 +3148,8 @@ class OpenAISipPhoneSession {
         const coordinated = !!toolState;
         if (name === "voice_end_call" && toolState)
             toolState.suppressFollowup = true;
+        if (toolState && !toolState.suppressFollowup)
+            this.scheduleRealtimeToolPresence(responseId, toolState);
         let output;
         try {
             const args = parseToolArguments(typeof event.arguments === "string" ? event.arguments : "");
@@ -2924,14 +3190,18 @@ class OpenAISipPhoneSession {
         }
     }
     noteRealtimeResponseCreated(event) {
+        this.realtimeResponseCreateInFlight = null;
+        this.untrackedActiveRealtimeResponse = false;
+        this.clearUntrackedActiveRealtimeResponseTimer();
         const responseId = realtimeResponseId(event);
         if (responseId)
             this.activeRealtimeResponseId = responseId;
     }
-    noteRealtimeResponseDone(responseId) {
-        if (!responseId || this.activeRealtimeResponseId === responseId) {
-            this.activeRealtimeResponseId = null;
-        }
+    noteRealtimeResponseDone(_responseId) {
+        this.realtimeResponseCreateInFlight = null;
+        this.untrackedActiveRealtimeResponse = false;
+        this.clearUntrackedActiveRealtimeResponseTimer();
+        this.activeRealtimeResponseId = null;
         this.responseCreateHoldUntilMs = Math.max(this.responseCreateHoldUntilMs, Date.now() + OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS);
         this.schedulePendingRealtimeResponse(OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS);
     }
@@ -2939,15 +3209,21 @@ class OpenAISipPhoneSession {
         if (this.closed)
             return;
         const waitMs = Math.max(0, this.responseCreateHoldUntilMs - Date.now());
-        if (this.activeRealtimeResponseId || waitMs > 0) {
-            const pendingResponse = response ?? this.pendingRealtimeResponse?.response;
-            this.pendingRealtimeResponse = pendingResponse ? { response: pendingResponse } : {};
-            if (!this.activeRealtimeResponseId)
+        if (this.realtimeResponseIsBusy() || waitMs > 0) {
+            this.holdRealtimeResponse(response ? { response } : {});
+            if (!this.realtimeResponseIsBusy())
                 this.schedulePendingRealtimeResponse(waitMs);
             return;
         }
         this.sendRealtimeResponseCreate(response ? { response } : {});
     }
+    realtimeResponseIsBusy() {
+        return !!this.activeRealtimeResponseId || !!this.realtimeResponseCreateInFlight || this.untrackedActiveRealtimeResponse;
+    }
+    holdRealtimeResponse(request) {
+        const pendingResponse = request.response ?? this.pendingRealtimeResponse?.response;
+        this.pendingRealtimeResponse = pendingResponse ? { response: pendingResponse } : {};
+    }
     schedulePendingRealtimeResponse(delayMs) {
         if (!this.pendingRealtimeResponse)
             return;
@@ -2960,7 +3236,7 @@ class OpenAISipPhoneSession {
         this.pendingRealtimeResponseTimer.unref?.();
     }
     flushPendingRealtimeResponse() {
-        if (!this.pendingRealtimeResponse || this.closed || this.activeRealtimeResponseId)
+        if (!this.pendingRealtimeResponse || this.closed || this.realtimeResponseIsBusy())
             return;
         const waitMs = Math.max(0, this.responseCreateHoldUntilMs - Date.now());
         if (waitMs > 0) {
@@ -2972,11 +3248,38 @@ class OpenAISipPhoneSession {
         this.sendRealtimeResponseCreate(pending);
     }
     sendRealtimeResponseCreate(request) {
+        this.realtimeResponseCreateInFlight = request;
         this.sendOpenAI({
             type: "response.create",
             ...(request.response ? { response: request.response } : {}),
         });
     }
+    noteRealtimeResponseConflict() {
+        const inFlight = this.realtimeResponseCreateInFlight;
+        this.realtimeResponseCreateInFlight = null;
+        this.untrackedActiveRealtimeResponse = true;
+        if (inFlight)
+            this.holdRealtimeResponse(inFlight);
+        this.scheduleUntrackedActiveRealtimeResponseFallback();
+    }
+    scheduleUntrackedActiveRealtimeResponseFallback() {
+        this.clearUntrackedActiveRealtimeResponseTimer();
+        this.untrackedActiveRealtimeResponseTimer = setTimeout(() => {
+            this.untrackedActiveRealtimeResponseTimer = null;
+            if (this.closed || !this.untrackedActiveRealtimeResponse)
+                return;
+            this.untrackedActiveRealtimeResponse = false;
+            this.responseCreateHoldUntilMs = Math.max(this.responseCreateHoldUntilMs, Date.now() + OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS);
+            this.schedulePendingRealtimeResponse(OPENAI_REALTIME_RESPONSE_CREATE_GRACE_MS);
+        }, OPENAI_REALTIME_RESPONSE_CREATE_CONFLICT_BACKOFF_MS);
+        this.untrackedActiveRealtimeResponseTimer.unref?.();
+    }
+    clearUntrackedActiveRealtimeResponseTimer() {
+        if (!this.untrackedActiveRealtimeResponseTimer)
+            return;
+        clearTimeout(this.untrackedActiveRealtimeResponseTimer);
+        this.untrackedActiveRealtimeResponseTimer = null;
+    }
     requestHangupFromTool() {
         if (this.closed)
             return;
@@ -3050,6 +3353,9 @@ class OpenAISipPhoneSession {
             clearTimeout(this.pendingRealtimeResponseTimer);
             this.pendingRealtimeResponseTimer = null;
         }
+        this.clearPendingUserTurnResponse();
+        this.clearRealtimeToolPresenceTimers();
+        this.clearUntrackedActiveRealtimeResponseTimer();
         (0, runtime_1.emitNervesEvent)({
             component: "senses",
             event: "senses.voice_openai_sip_call_stop",
@@ -3805,7 +4111,7 @@ async function handleOutgoing(options, basePath, outboundId, params, jobs) {
         Reason: job.reason,
         InitialAudio: encodeVoiceCallAudioCustomParameter(job.initialAudio),
     };
-    if (usesOpenAISipConversationEngine(options)) {
+    if (usesOpenAISipOutboundConversationEngine(options)) {
         (0, runtime_1.emitNervesEvent)({
             component: "senses",
             event: "senses.voice_twilio_sip_connect",
@@ -3824,8 +4130,8 @@ async function handleOutgoing(options, basePath, outboundId, params, jobs) {
         })));
     }
     if (normalizeTwilioPhoneTransportMode(options.transportMode) === "media-stream") {
-        if (usesOpenAIRealtimeConversationEngine(options)) {
-            return xmlResponse(mediaStreamTwiml(options, basePath, { From: from, To: to }, undefined, streamParams));
+        if (usesOpenAIRealtimeOutboundConversationEngine(options)) {
+            return xmlResponse(mediaStreamTwiml(options, basePath, { From: from, To: to }, undefined, streamParams, "openai-realtime"));
         }
         try {
             await fs.mkdir(callDir, { recursive: true });
@@ -4228,7 +4534,7 @@ function createTwilioPhoneBridge(options) {
         byOutboundId: new Map(),
     };
     const activeSipSessions = new ActiveOpenAISipSessions();
-    mediaStreams.on("connection", (ws) => {
+    mediaStreams.on("connection", (ws, request) => {
         const lifecycle = {
             onIdentityChange: (activeSession, identity) => {
                 if (identity.callSid)
@@ -4245,7 +4551,8 @@ function createTwilioPhoneBridge(options) {
                 }
             },
         };
-        const session = usesOpenAIRealtimeConversationEngine(options)
+        const streamEngine = mediaStreamRequestedConversationEngine(request.url);
+        const session = streamEngine === "openai-realtime" || (!streamEngine && usesOpenAIRealtimeConversationEngine(options))
             ? new TwilioOpenAIRealtimeMediaStreamSession(ws, options, lifecycle)
             : new TwilioMediaStreamSession(ws, options, jobs, lifecycle);
         session.attach();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ouro.bot/cli",
-  "version": "0.1.0-alpha.582",
+  "version": "0.1.0-alpha.583",
   "main": "dist/heart/daemon/ouro-entry.js",
   "bin": {
     "cli": "dist/heart/daemon/ouro-bot-entry.js",