@elizaos/ui 2.0.0-beta.2 → 2.0.11-beta.7

package/api/ios-local-agent-kernel.js

@@ -18,7 +18,10 @@ const ASSIGNMENTS_KEY = `${STORAGE_PREFIX}:assignments:v1`;
 const BROWSER_WORKSPACE_KEY = `${STORAGE_PREFIX}:browser-workspace:v1`;
 const WALLET_MARKET_OVERVIEW_KEY = `${STORAGE_PREFIX}:wallet-market-overview:v1`;
 const BUNDLE_INDEX_KEY = `${STORAGE_PREFIX}:eliza-1-bundles:v1`;
+const ACTIVE_SERVER_STORAGE_KEY = "elizaos:active-server";
 const AGENT_NAME = "Eliza";
+const IOS_LOCAL_AGENT_IPC_BASE = "eliza-local-agent://ipc";
+const DIRECT_CLOUD_API_BASE = "https://api.elizacloud.ai";
 const DEFAULT_SYSTEM_PROMPT = "You are Eliza, a private on-device assistant. Answer directly and concisely.";
 const DEFAULT_CLOUD_MARKET_PREVIEW_BASE_URL = "https://www.elizacloud.ai";
 const CLOUD_WALLET_MARKET_OVERVIEW_PATH = "/market/preview/wallet-overview";
@@ -51,6 +54,8 @@ const EMPTY_ROUTING_PREFERENCES = {
     preferredProvider: {},
     policy: {},
 };
+const IOS_LOCAL_BACKGROUND_UNAVAILABLE_REASON = "iOS local mode uses the WebView ITTP route kernel. Capacitor BackgroundRunner wakes in a separate JSContext and cannot call that WebView kernel while the app is suspended.";
+const IOS_LOCAL_TTS_EXECUTOR_AVAILABLE = false;
 const EMPTY_WALLET_ADDRESSES = {
     evmAddress: null,
     solanaAddress: null,
@@ -127,9 +132,102 @@ function writeJson(key, value) {
         // localStorage can be unavailable in embedded shells.
     }
 }
+function stringValue(value) {
+    return typeof value === "string" && value.trim() ? value.trim() : null;
+}
+function readIosCloudPairing() {
+    const fallback = {
+        paired: false,
+        agentId: null,
+        token: null,
+        apiBase: DIRECT_CLOUD_API_BASE,
+        label: null,
+    };
+    const activeServer = readJson(ACTIVE_SERVER_STORAGE_KEY, null);
+    if (activeServer?.kind !== "cloud") {
+        return fallback;
+    }
+    const id = stringValue(activeServer.id);
+    const agentId = id?.startsWith("cloud:") ? id.slice("cloud:".length) : null;
+    const storedToken = stringValue(activeServer.accessToken);
+    const globalToken = stringValue(globalThis.__ELIZA_CLOUD_AUTH_TOKEN__);
+    const token = storedToken ?? globalToken;
+    const label = stringValue(activeServer.label);
+    return {
+        paired: Boolean(agentId && token),
+        agentId,
+        token,
+        apiBase: DIRECT_CLOUD_API_BASE,
+        label,
+    };
+}
 function readBundleIndex() {
     return readJson(BUNDLE_INDEX_KEY, {});
 }
+function isIosVoiceAssetPath(rawPath) {
+    const path = rawPath.toLowerCase().replace(/\\/g, "/");
+    if (!/\.(bin|codec|gguf|json)$/i.test(path))
+        return false;
+    if (/(^|\/)(asr|tts|vad|voice|voices|wakeword)\//.test(path))
+        return true;
+    return /(^|\/)cache\/[^/]*voice[^/]*\.(bin|codec|gguf|json)$/i.test(path);
+}
+function inferVoiceReadiness(installed, bundles) {
+    const installedVoicePaths = new Set();
+    for (const model of installed) {
+        if (isIosVoiceAssetPath(model.path))
+            installedVoicePaths.add(model.path);
+    }
+    const bundleIdsWithVoice = new Set();
+    for (const record of Object.values(bundles)) {
+        const paths = [
+            ...Object.keys(record.files),
+            ...Object.values(record.files),
+        ].filter(isIosVoiceAssetPath);
+        if (paths.length > 0) {
+            bundleIdsWithVoice.add(record.modelId);
+            for (const path of paths)
+                installedVoicePaths.add(path);
+        }
+    }
+    const modelId = [...bundleIdsWithVoice][0] ??
+        installed.find((model) => model.id.startsWith("eliza-1"))?.id ??
+        null;
+    const installedFiles = installedVoicePaths.size;
+    if (installedFiles === 0) {
+        return {
+            status: "missing",
+            installedFiles: 0,
+            modelId,
+            message: "Eliza-1 voice assets are not visible to the iOS local agent.",
+        };
+    }
+    if (!IOS_LOCAL_TTS_EXECUTOR_AVAILABLE) {
+        return {
+            status: "unavailable",
+            installedFiles,
+            modelId,
+            message: "Eliza-1 voice assets are installed. This build is missing the iOS local voice playback engine.",
+        };
+    }
+    return {
+        status: "assets-ready",
+        installedFiles,
+        modelId,
+        message: "Eliza-1 voice assets are installed. Voice engine will warm on first playback.",
+    };
+}
+function sanitizeLocalSpeechText(input) {
+    let text = input.normalize("NFKC");
+    text = text.replace(/<think\b[^>]*>[\s\S]*?(?:<\/think>|$)/gi, " ");
+    text = text.replace(/<(analysis|reasoning|tool_calls?|tools?)\b[^>]*>[\s\S]*?(?:<\/\1>|$)/gi, " ");
+    text = text.replace(/```[\s\S]*?```/g, " ");
+    text = text.replace(/`([^`]+)`/g, "$1");
+    text = text.replace(/\[([^\]]+)\]\([^)]+\)/g, "$1");
+    text = text.replace(/<[^>\n]+>/g, " ");
+    text = text.replace(/\bhttps?:\/\/\S+/gi, " ");
+    return text.replace(/\s+/g, " ").trim();
+}
 function writeBundleRecord(record) {
     const current = readBundleIndex();
     current[record.modelId] = record;
@@ -323,16 +421,73 @@ async function capacitorLlamaProviderStatus() {
     };
 }
 function localConfig() {
+    const cloud = readIosCloudPairing();
     return {
-        meta: { onboardingComplete: true },
+        meta: { firstRunComplete: true },
         ui: {},
         cloud: {
-            enabled: false,
-            connectionStatus: "disconnected",
-            cloudProvisioned: false,
+            enabled: cloud.paired,
+            connectionStatus: cloud.paired ? "connected" : "disconnected",
+            cloudProvisioned: cloud.paired,
+            activeAgentId: cloud.agentId,
         },
     };
 }
+function localAgentCapabilities() {
+    return {
+        mode: "ios-local",
+        apiBase: IOS_LOCAL_AGENT_IPC_BASE,
+        transport: {
+            foreground: "ittp",
+            background: "unavailable",
+            tcpListener: false,
+            nativeRequestProxy: false,
+        },
+        routeKernel: {
+            shape: "fetch",
+            hostedIn: "webview",
+            honoRoutesDetected: false,
+        },
+        backendRuntime: {
+            state: "compatibility-kernel",
+            fullAgentRuntime: false,
+            node: false,
+            bun: false,
+            taskService: false,
+            pluginLoader: false,
+        },
+        storage: {
+            conversations: "native-synced-localStorage",
+            localInferenceState: "native-synced-localStorage",
+        },
+        localInference: {
+            state: "available-when-native-llama-plugin-is-present",
+            provider: "capacitor-llama",
+        },
+        scheduledTasks: {
+            state: "unavailable",
+            primitive: "ScheduledTask",
+            reason: IOS_LOCAL_BACKGROUND_UNAVAILABLE_REASON,
+        },
+        apps: {
+            state: "catalog-unavailable",
+            reason: "The runtime AppManager is not mounted in the iOS ITTP kernel.",
+        },
+        plugins: {
+            state: "loader-unavailable",
+            reason: "The runtime plugin loader is not mounted in the iOS ITTP kernel.",
+        },
+    };
+}
+function unavailableLocalBackendRoute(error, details = {}, status = 503) {
+    return json({
+        ok: false,
+        error,
+        mode: "ios-local",
+        capabilities: localAgentCapabilities(),
+        ...details,
+    }, status);
+}
 function localCharacter() {
     return {
         name: AGENT_NAME,
@@ -779,6 +934,44 @@ function commonBundleRoot(paths) {
     const [first, ...rest] = dirs;
     return rest.every((dir) => dir === first) ? first : undefined;
 }
+function normalizedSha256(value) {
+    return typeof value === "string" && /^[a-f0-9]{64}$/.test(value)
+        ? value
+        : null;
+}
+function normalizeNativeHashResult(result) {
+    if (typeof result === "string") {
+        const sha256 = normalizedSha256(result);
+        if (!sha256)
+            throw new Error("Native hashFile returned an invalid SHA256");
+        return { sha256 };
+    }
+    const sha256 = normalizedSha256(result.sha256 ?? result.hash);
+    if (!sha256)
+        throw new Error("Native hashFile returned an invalid SHA256");
+    const sizeBytes = typeof result.sizeBytes === "number"
+        ? result.sizeBytes
+        : typeof result.size === "number"
+            ? result.size
+            : undefined;
+    return {
+        sha256,
+        ...(typeof sizeBytes === "number" && sizeBytes >= 0 ? { sizeBytes } : {}),
+    };
+}
+async function hashNativeBundleFile(llama, filePath, label) {
+    if (!llama.hashFile) {
+        throw new Error(`Native Eliza-1 downloader cannot verify SHA256 for ${label}; refusing bundle install.`);
+    }
+    return normalizeNativeHashResult(await llama.hashFile(filePath));
+}
+async function verifyNativeBundleFile(llama, filePath, expectedSha256, label) {
+    const hashed = await hashNativeBundleFile(llama, filePath, label);
+    if (hashed.sha256 !== expectedSha256) {
+        throw new Error(`SHA256 mismatch for ${label}: expected ${expectedSha256}, got ${hashed.sha256}`);
+    }
+    return hashed;
+}
 function iosBundleFilename(model, filePath) {
     if (filePath === model.bundleManifestFile) {
         return `${model.id.replace(/[^a-zA-Z0-9._-]/g, "_")}.manifest.json`;
@@ -808,22 +1001,21 @@ function parseIosBundleManifest(input, model) {
         "voice",
         "asr",
         "vision",
-        "dflash",
         "cache",
+        "vad",
     ]) {
         if (!Array.isArray(raw.files[kind])) {
             throw new Error(`Invalid Eliza-1 manifest files.${kind}`);
         }
     }
+    for (const kind of ["text", "voice", "asr", "cache", "vad"]) {
+        if (raw.files[kind].length === 0) {
+            throw new Error(`Invalid Eliza-1 manifest files.${kind} must be non-empty`);
+        }
+    }
     if (!raw.files.text.some((entry) => entry.path === model.ggufFile)) {
         throw new Error(`Eliza-1 manifest missing text file ${model.ggufFile}`);
     }
-    const drafterId = model.runtime?.dflash?.drafterModelId;
-    const drafter = drafterId ? findCatalogModel(drafterId) : undefined;
-    if (drafter &&
-        !raw.files.dflash.some((entry) => entry.path === drafter.ggufFile)) {
-        throw new Error(`Eliza-1 manifest missing DFlash file ${drafter.ggufFile}`);
-    }
     return raw;
 }
 function collectIosBundleFiles(manifest) {
@@ -833,7 +1025,6 @@ function collectIosBundleFiles(manifest) {
         manifest.files.voice,
         manifest.files.asr,
         manifest.files.vision,
-        manifest.files.dflash,
         manifest.files.cache,
         manifest.files.embedding ?? [],
         manifest.files.vad ?? [],
@@ -890,7 +1081,7 @@ function gpuBackendForMobile(platform, backend) {
     return platform === "android" ? "vulkan" : "metal";
 }
 function mobileContextSize(hardware, catalog) {
-    const target = catalog?.runtime?.dflash?.contextSize ?? 4096;
+    const target = catalog?.contextLength ?? 4096;
     if (hardware.totalRamGb >= 12)
         return Math.min(target, 8192);
     if (hardware.totalRamGb >= 8)
@@ -902,7 +1093,7 @@ function mobileThreadCount(hardware) {
         return 0;
     return Math.max(2, Math.min(Math.floor(hardware.cpuCores) - 1, 6));
 }
-function companionInstalled(installed, modelId) {
+function _companionInstalled(installed, modelId) {
     return installed.find((entry) => entry.id === modelId);
 }
 async function listInstalledModels() {
@@ -917,11 +1108,7 @@ async function listInstalledModels() {
         const catalog = catalogForAvailableModel(model);
         const id = catalog?.id ??
             (model.name || model.path || randomId("model")).replace(/\.gguf$/i, "");
-        const bundle = catalog?.companionForModelId
-            ? bundles[catalog.companionForModelId]
-            : catalog
-                ? bundles[catalog.id]
-                : undefined;
+        const bundle = catalog ? bundles[catalog.id] : undefined;
         return {
             id,
             displayName: catalog?.displayName ?? model.name ?? id,
@@ -929,6 +1116,9 @@ async function listInstalledModels() {
             sizeBytes: typeof model.size === "number" ? model.size : 0,
             ...(bundle?.bundleRoot ? { bundleRoot: bundle.bundleRoot } : {}),
             ...(bundle?.manifestPath ? { manifestPath: bundle.manifestPath } : {}),
+            ...(bundle?.manifestSha256
+                ? { manifestSha256: bundle.manifestSha256 }
+                : {}),
             ...(bundle?.bundleVersion
                 ? { bundleVersion: bundle.bundleVersion }
                 : {}),
@@ -940,9 +1130,6 @@ async function listInstalledModels() {
             lastUsedAt: activeState.modelId === id ? activeState.loadedAt : null,
             source: catalog ? "eliza-download" : "external-scan",
             ...(catalog?.runtimeRole ? { runtimeRole: catalog.runtimeRole } : {}),
-            ...(catalog?.companionForModelId
-                ? { companionFor: catalog.companionForModelId }
-                : {}),
         };
     });
 }
@@ -990,40 +1177,32 @@ async function hardwareProbe() {
                 ? { thermalState: hardware.thermalState }
                 : {}),
             gpuSupported: hardware?.gpuSupported ?? Boolean(gpu),
-            dflashSupported: hardware?.dflashSupported ?? false,
-            dflashReason: hardware?.dflashReason ??
-                "native runtime has not reported DFlash drafter support",
+            mtpSupported: hardware?.mtpSupported ?? true,
+            mtpReason: hardware?.mtpReason ?? "native MTP is catalog-enabled",
             source: hardware?.source ?? "adapter-fallback",
         },
     };
 }
 function buildMobileLoadOptions(model, installed, hardware) {
+    void installed;
     const catalog = findCatalogModel(model.id);
+    const mtp = catalog?.runtime?.mtp;
     const options = {
         modelPath: model.path,
         contextSize: mobileContextSize(hardware, catalog),
         useGpu: hardware.mobile?.gpuSupported !== false,
         maxThreads: mobileThreadCount(hardware),
     };
-    const dflash = catalog?.runtime?.dflash;
-    if (!dflash)
-        return options;
-    if (hardware.mobile?.dflashSupported !== true)
-        return options;
-    const drafter = companionInstalled(installed, dflash.drafterModelId);
-    if (!drafter)
-        return options;
     return {
         ...options,
-        draftModelPath: drafter.path,
-        draftContextSize: dflash.draftContextSize,
-        draftMin: dflash.draftMin,
-        draftMax: dflash.draftMax,
-        speculativeSamples: Math.min(dflash.draftMax, 4),
+        draftContextSize: options.contextSize,
+        draftMin: mtp?.draftMin ?? 1,
+        draftMax: mtp?.draftMax ?? 4,
+        speculativeSamples: Math.min(mtp?.draftMax ?? 4, 4),
         mobileSpeculative: true,
-        cacheTypeK: catalog.runtime?.kvCache?.typeK,
-        cacheTypeV: catalog.runtime?.kvCache?.typeV,
-        disableThinking: dflash.disableThinking,
+        cacheTypeK: catalog?.runtime?.kvCache?.typeK,
+        cacheTypeV: catalog?.runtime?.kvCache?.typeV,
+        disableThinking: true,
     };
 }
 function runtimeSignature(options) {
@@ -1038,8 +1217,6 @@ function runtimeSignature(options) {
     ].join("|");
 }
 async function validateMobileModelFit(model) {
-    if (model.runtimeRole === "dflash-drafter")
-        return null;
     const hardware = await hardwareProbe();
     const fit = assessCatalogModelFit(hardware, model, MODEL_CATALOG);
     if (fit === "wontfit") {
@@ -1054,7 +1231,8 @@ async function validateMobileModelFit(model) {
     }
     return null;
 }
-async function ensureActiveModelLoaded() {
+let loadInFlightPromise = null;
+async function ensureActiveModelLoadedImpl() {
     if (activeState.status !== "ready" || !activeState.modelId) {
         throw new Error("No local model is active. Install and activate a GGUF model first.");
     }
@@ -1071,7 +1249,7 @@ async function ensureActiveModelLoaded() {
     }
     const llama = await loadCapacitorLlama();
     if (!llama?.load || !llama.generate) {
-        throw new Error("Capacitor llama runtime is not available on this build.");
+        throw new Error("Native Eliza-1 runtime is not available on this build.");
     }
     const hardware = await hardwareProbe();
     const loadOptions = buildMobileLoadOptions(model, installed, hardware);
@@ -1085,6 +1263,23 @@ async function ensureActiveModelLoaded() {
     await llama.load(loadOptions);
     loadedRuntimeSignature = signature;
 }
+/**
+ * Load the active local model into the native llama runtime.
+ *
+ * Mutex: concurrent callers (e.g. a chat send racing the background-runner
+ * pre-warm) share the same in-flight promise. Without this, two parallel
+ * `llama.load(...)` calls can land back-to-back during a cold boot and
+ * either tear down each other's mmap state or double-allocate the KV cache.
+ */
+async function ensureActiveModelLoaded() {
+    if (loadInFlightPromise) {
+        return loadInFlightPromise;
+    }
+    loadInFlightPromise = ensureActiveModelLoadedImpl().finally(() => {
+        loadInFlightPromise = null;
+    });
+    return loadInFlightPromise;
+}
 function emptyUsage(modelId) {
     return {
         promptTokens: 0,
@@ -1229,33 +1424,273 @@ function buildPrompt(messages, latestText) {
         .join("\n");
     return `${DEFAULT_SYSTEM_PROMPT}\n\n${history}${history ? "\n" : ""}User: ${latestText}\nAssistant:`;
 }
+function classifyIosLocalGenerateError(err) {
+    if (err instanceof Error) {
+        const name = err.name;
+        const msg = err.message.toLowerCase();
+        if (name === "AbortError") {
+            return { fallback: false, reason: "local-aborted-pre-completion" };
+        }
+        if (msg.includes("native eliza-1 runtime is not available") ||
+            msg.includes("not loaded") ||
+            msg.includes("not installed") ||
+            msg.includes("not staged")) {
+            return { fallback: true, reason: "local-unavailable" };
+        }
+        if (msg.includes("thermal") ||
+            msg.includes("low-power") ||
+            msg.includes("memory slot")) {
+            return { fallback: true, reason: "local-overloaded" };
+        }
+    }
+    return { fallback: false, reason: "local-error" };
+}
+const CLOUD_PAIRED_PROBE_TIMEOUT_MS = 2_000;
+const fetchWithOptionalPreconnect = fetch;
+const fetchIosKernelRoute = Object.assign(async function fetchIosKernelRoute(input, init) {
+    const request = input instanceof Request ? input : new Request(input, init);
+    return handleIosLocalAgentRequest(request);
+}, {
+    preconnect: (...args) => fetchWithOptionalPreconnect.preconnect?.(...args),
+});
+async function probeAgentCloudPaired(fetchImpl = fetchIosKernelRoute) {
+    const controller = new AbortController();
+    const timer = globalThis.setTimeout(() => controller.abort(), CLOUD_PAIRED_PROBE_TIMEOUT_MS);
+    let response;
+    try {
+        response = await fetchImpl(`${IOS_LOCAL_AGENT_IPC_BASE}/api/auth/status`, {
+            method: "GET",
+            headers: { accept: "application/json" },
+            signal: controller.signal,
+        });
+    }
+    catch (err) {
+        return {
+            kind: "unknown",
+            reason: err instanceof Error ? err.message : String(err),
+        };
+    }
+    finally {
+        globalThis.clearTimeout(timer);
+    }
+    if (!response.ok) {
+        return { kind: "unknown", reason: `auth-status http ${response.status}` };
+    }
+    const body = await response.json().catch(() => null);
+    if (!body || typeof body !== "object") {
+        return { kind: "unknown", reason: "auth-status non-object body" };
+    }
+    const cloudProvisioned = body
+        .cloudProvisioned;
+    if (cloudProvisioned === true) {
+        return { kind: "paired" };
+    }
+    return { kind: "not-paired", reason: "cloudProvisioned=false" };
+}
+function cloudBridgeResultText(result) {
+    const record = asRecord(result);
+    if (!record)
+        return null;
+    return (stringValue(record.text) ??
+        stringValue(record.reply) ??
+        stringValue(record.message));
+}
+async function sendPromptToIosCloud(prompt) {
+    const pairing = readIosCloudPairing();
+    if (!pairing.paired || !pairing.agentId || !pairing.token) {
+        throw new Error("Eliza Cloud is not paired.");
+    }
+    const response = await fetch(`${pairing.apiBase}/api/v1/eliza/agents/${encodeURIComponent(pairing.agentId)}/bridge`, {
+        method: "POST",
+        headers: {
+            "content-type": "application/json",
+            accept: "application/json",
+            authorization: `Bearer ${pairing.token}`,
+        },
+        body: JSON.stringify({
+            jsonrpc: "2.0",
+            id: randomId("cloud"),
+            method: "message.send",
+            params: { text: prompt },
+        }),
+    });
+    if (!response.ok) {
+        throw new Error(`Cloud bridge failed: HTTP ${response.status}`);
+    }
+    const body = asRecord(await response.json().catch(() => null));
+    if (!body) {
+        throw new Error("Cloud bridge returned a non-object response.");
+    }
+    const error = asRecord(body.error);
+    if (error) {
+        throw new Error(stringValue(error.message) ?? "Cloud bridge returned an error.");
+    }
+    const result = asRecord(body.result);
+    const text = cloudBridgeResultText(result);
+    if (!text) {
+        throw new Error("Cloud bridge response missing text.");
+    }
+    const modelId = stringValue(result?.model);
+    return {
+        text,
+        promptTokens: 0,
+        completionTokens: 0,
+        ...(modelId ? { modelId } : {}),
+    };
+}
+async function handleIosCloudChat(request) {
+    const body = await requestJson(request);
+    const prompt = stringValue(body.prompt) ??
+        stringValue(body.text) ??
+        stringValue(body.message);
+    if (!prompt)
+        return json({ error: "prompt is required" }, 400);
+    try {
+        return json(await sendPromptToIosCloud(prompt));
+    }
+    catch (error) {
+        const message = error instanceof Error ? error.message : String(error);
+        return json({ error: message }, message.includes("not paired") ? 409 : 502);
+    }
+}
+/**
+ * Forward a prompt to the paired Eliza Cloud agent. The local agent's
+ * `/api/cloud/chat` proxy already understands how to relay this to the
+ * cloud-side model; we send a typed POST and return the canonical
+ * `CloudForwardResult` shape. Errors propagate so the caller can decide
+ * whether to surface "honest cloud failed" or just rethrow.
+ */
+async function forwardToAgentCloudChat(prompt, fetchImpl = fetchIosKernelRoute) {
+    const response = await fetchImpl(`${IOS_LOCAL_AGENT_IPC_BASE}/api/cloud/chat`, {
+        method: "POST",
+        headers: {
+            "content-type": "application/json",
+            accept: "application/json",
+        },
+        body: JSON.stringify({ prompt, maxTokens: 256, temperature: 0.7 }),
+    });
+    if (!response.ok) {
+        throw new Error(`[ios-local-agent] Cloud fallback failed: HTTP ${response.status} ${response.statusText}`);
+    }
+    const body = await response.json();
+    if (!body || typeof body !== "object") {
+        throw new Error("[ios-local-agent] Cloud fallback: non-object response");
+    }
+    const record = body;
+    const textValue = record.text;
+    if (typeof textValue !== "string") {
+        throw new Error("[ios-local-agent] Cloud fallback: response missing string 'text'");
+    }
+    const promptTokens = Number.isFinite(record.promptTokens)
+        ? record.promptTokens
+        : 0;
+    const completionTokens = Number.isFinite(record.completionTokens)
+        ? record.completionTokens
+        : 0;
+    const modelId = typeof record.modelId === "string" ? record.modelId : undefined;
+    return {
+        text: textValue,
+        promptTokens,
+        completionTokens,
+        ...(modelId ? { modelId } : {}),
+    };
+}
 async function generateLocalReply(conversation, text) {
+    const cloudProbe = await probeAgentCloudPaired();
+    if (cloudProbe.kind === "paired") {
+        const prompt = buildPrompt(conversation.messages, text);
+        const cloud = await forwardToAgentCloudChat(prompt);
+        return {
+            text: cloud.text.trim() || "Cloud returned an empty response.",
+            usage: {
+                promptTokens: cloud.promptTokens,
+                completionTokens: cloud.completionTokens,
+                totalTokens: cloud.promptTokens + cloud.completionTokens,
+                ...(cloud.modelId ? { model: cloud.modelId } : {}),
+            },
+        };
+    }
     const statusReply = await localModelStatusReply(text);
     if (statusReply)
         return statusReply;
     await ensureActiveModelLoaded();
     const llama = await loadCapacitorLlama();
     if (!llama?.generate) {
-        throw new Error("Capacitor llama runtime is not available on this build.");
+        // Surface as a fallback-eligible failure so the caller / future cloud
+        // routing layer can detect it cleanly instead of grepping the message.
+        const err = new Error("Native Eliza-1 runtime is not available on this build.");
+        err.name = "LocalRuntimeUnavailableError";
+        throw err;
     }
     const prompt = buildPrompt(conversation.messages, text);
-    const result = await llama.generate({
-        prompt,
-        maxTokens: 256,
-        temperature: 0.7,
-        topP: 0.9,
-        stopSequences: ["\nUser:", "\nAssistant:"],
-    });
-    const cleaned = result.text.trim() || "I could not generate a local response.";
-    return {
-        text: cleaned,
-        usage: {
-            promptTokens: result.promptTokens,
-            completionTokens: result.outputTokens,
-            totalTokens: result.promptTokens + result.outputTokens,
-            ...(activeState.modelId ? { model: activeState.modelId } : {}),
-        },
-    };
+    try {
+        const result = await llama.generate({
+            prompt,
+            maxTokens: 256,
+            temperature: 0.7,
+            topP: 0.9,
+            stopSequences: ["\nUser:", "\nAssistant:"],
+        });
+        const cleaned = result.text.trim() || "I could not generate a local response.";
+        return {
+            text: cleaned,
+            usage: {
+                promptTokens: result.promptTokens,
+                completionTokens: result.outputTokens,
+                totalTokens: result.promptTokens + result.outputTokens,
+                ...(activeState.modelId ? { model: activeState.modelId } : {}),
+            },
+        };
+    }
+    catch (err) {
+        const cls = classifyIosLocalGenerateError(err);
+        if (!cls.fallback) {
+            throw err;
+        }
+        // Local failed in a fallback-eligible way. Probe the in-process agent
+        // for cloud-paired state via `/api/auth/status`; if paired, forward
+        // the prompt to the agent's cloud proxy and return its response. If
+        // NOT paired, surface an honest, actionable error — silently inventing
+        // a synthetic response is a bigger bug than telling the user "pair cloud
+        // or use a smaller model".
+        const probe = await probeAgentCloudPaired();
+        if (probe.kind === "paired") {
+            try {
+                const cloud = await forwardToAgentCloudChat(prompt);
+                return {
+                    text: cloud.text.trim() || "Cloud returned an empty response.",
+                    usage: {
+                        promptTokens: cloud.promptTokens,
+                        completionTokens: cloud.completionTokens,
+                        totalTokens: cloud.promptTokens + cloud.completionTokens,
+                        ...(cloud.modelId
+                            ? { model: cloud.modelId }
+                            : activeState.modelId
+                                ? { model: activeState.modelId }
+                                : {}),
+                    },
+                };
+            }
+            catch (cloudErr) {
+                // Cloud was paired but the forward failed. We can't pretend it
+                // succeeded, so we surface a precise error that names both the
+                // local cause AND the cloud attempt's failure.
+                const honest = new Error(`Local model failed (${cls.reason}) and the paired cloud fallback also failed: ${cloudErr instanceof Error ? cloudErr.message : String(cloudErr)}`);
+                honest.name = "LocalInferenceFallbackFailed";
+                honest.cause = cloudErr;
+                throw honest;
+            }
+        }
+        // Not paired (or probe unreachable). HONEST error per the wave-4 spec —
+        // do NOT swallow.
+        const honest = new Error("Local model failed and cloud is not paired. Pair Eliza Cloud or download a smaller model.");
+        honest.name = "LocalInferenceFallbackRequired";
+        const cause = err instanceof Error ? err : undefined;
+        if (cause) {
+            honest.cause = cause;
+        }
+        throw honest;
+    }
 }
 async function generateLocalGreeting() {
     const setupReply = await localModelStatusReply("download the default local model");
@@ -1296,6 +1731,7 @@ async function hubSnapshot() {
         listInstalledModels(),
         hardwareProbe(),
     ]);
+    const bundles = readBundleIndex();
     return {
         catalog: MODEL_CATALOG,
         installed,
@@ -1303,6 +1739,7 @@ async function hubSnapshot() {
         downloads: [...downloads.values()],
         hardware,
         assignments: readAssignments(),
+        voiceReadiness: inferVoiceReadiness(installed, bundles),
     };
 }
 function updateDownload(job, patch) {
@@ -1313,10 +1750,10 @@ function sleep(ms) {
     return new Promise((resolve) => setTimeout(resolve, ms));
 }
 function formatGb(bytes) {
-    return `${(bytes / 1024 ** 3).toFixed(1)}GB`;
+    return `${(bytes / 1024 ** 3).toFixed(1)} GB`;
 }
 function aggregateDownloadJobs(model) {
-    const ids = [model.id, ...(model.companionModelIds ?? [])];
+    const ids = [model.id];
     return ids.flatMap((id) => {
         const job = downloads.get(id);
         return job ? [job] : [];
@@ -1398,23 +1835,7 @@ function statusLine(model, jobs, detail) {
     return templates[index]();
 }
 async function queueCompanionDownloads(model) {
-    if (!model.companionModelIds?.length)
-        return;
-    const installed = await listInstalledModels().catch(() => []);
-    for (const companionId of model.companionModelIds) {
-        const companion = findCatalogModel(companionId);
-        if (!companion)
-            continue;
-        if (installed.some((entry) => entry.id === companionId))
-            continue;
-        const existing = downloads.get(companionId);
-        if (existing &&
-            existing.state !== "failed" &&
-            existing.state !== "cancelled") {
-            continue;
-        }
-        startDownload(companion);
-    }
+    void model;
 }
 async function downloadNativeModelFile(llama, url, filename) {
     if (!llama.downloadModel) {
@@ -1434,8 +1855,11 @@ async function downloadIosBundle(model, llama, job) {
     }
     const manifest = parseIosBundleManifest(await manifestResponse.json(), model);
     const files = {};
+    let bundleSizeBytes = 0;
     const manifestPath = await downloadNativeModelFile(llama, manifestUrl, iosBundleFilename(model, model.bundleManifestFile));
     files[model.bundleManifestFile] = manifestPath;
+    const manifestHash = await hashNativeBundleFile(llama, manifestPath, model.bundleManifestFile);
+    bundleSizeBytes += manifestHash.sizeBytes ?? 0;
     const entries = collectIosBundleFiles(manifest);
     const totalSteps = Math.max(entries.length + 1, 1);
     let completedSteps = 1;
@@ -1444,7 +1868,10 @@ async function downloadIosBundle(model, llama, job) {
     });
     for (const entry of entries) {
         const url = buildHuggingFaceResolveUrlForPath(model, entry.path);
-        files[entry.path] = await downloadNativeModelFile(llama, url, iosBundleFilename(model, entry.path));
+        const downloadedPath = await downloadNativeModelFile(llama, url, iosBundleFilename(model, entry.path));
+        files[entry.path] = downloadedPath;
+        const fileHash = await verifyNativeBundleFile(llama, downloadedPath, entry.sha256, entry.path);
+        bundleSizeBytes += fileHash.sizeBytes ?? 0;
         completedSteps += 1;
         updateDownload(job, {
             received: Math.min(job.total, Math.round((completedSteps / totalSteps) * job.total)),
@@ -1458,8 +1885,9 @@ async function downloadIosBundle(model, llama, job) {
         modelId: model.id,
         bundleVersion: manifest.version,
         manifestPath,
+        manifestSha256: manifestHash.sha256,
         bundleRoot: commonBundleRoot(Object.values(files)),
-        bundleSizeBytes: job.total,
+        bundleSizeBytes: bundleSizeBytes > 0 ? bundleSizeBytes : job.total,
         files,
         installedAt: nowIso(),
     });
@@ -1484,7 +1912,7 @@ function startDownload(model) {
     downloads.set(model.id, job);
     void (async () => {
         try {
-            if (model.runtimeRole !== "dflash-drafter" && !model.bundleManifestFile) {
+            if (!model.bundleManifestFile) {
                 void queueCompanionDownloads(model);
             }
             updateDownload(job, { state: "downloading" });
@@ -1492,7 +1920,7 @@ function startDownload(model) {
             if (!llama?.downloadModel) {
                 throw new Error("Native Eliza-1 downloader is unavailable.");
             }
-            if (model.bundleManifestFile && model.runtimeRole !== "dflash-drafter") {
+            if (model.bundleManifestFile) {
                 const textPath = await downloadIosBundle(model, llama, job);
                 updateDownload(job, {
                     state: "completed",
@@ -1557,13 +1985,6 @@ function startDownload(model) {
                 received: job.total,
                 etaMs: 0,
             });
-            if (model.runtimeRole === "dflash-drafter") {
-                if (model.companionForModelId &&
-                    activeState.modelId === model.companionForModelId) {
-                    await activateModel(model.companionForModelId).catch(() => undefined);
-                }
-                return;
-            }
             if (activeState.status === "idle" || !activeState.modelId) {
                 await activateModel(model.id, path).catch(() => undefined);
             }
@@ -1579,16 +2000,6 @@ function startDownload(model) {
 }
 async function activateModel(modelId, knownPath) {
     const catalog = findCatalogModel(modelId);
-    if (catalog?.runtimeRole === "dflash-drafter") {
-        const state = {
-            modelId,
-            loadedAt: null,
-            status: "error",
-            error: `${catalog.displayName} is a DFlash drafter companion, not a standalone chat model.`,
-        };
-        writeActiveModelState(state);
-        return state;
-    }
     const installed = await listInstalledModels();
     const model = installed.find((entry) => entry.id === modelId) ??
         (knownPath
@@ -1617,7 +2028,7 @@ async function activateModel(modelId, knownPath) {
     try {
         const llama = await loadCapacitorLlama();
         if (!llama?.load) {
-            throw new Error("Capacitor llama runtime is not available on this build.");
+            throw new Error("Native Eliza-1 runtime is not available on this build.");
         }
         const hardware = await hardwareProbe();
         if (catalog) {
@@ -1675,9 +2086,27 @@ export async function handleIosLocalAgentRequest(request, _context = {}) {
             connectors: {},
             uptime: Math.floor((Date.now() - startedAt) / 1000),
             agentState: running ? "running" : "not_started",
+            localAgent: {
+                mode: "ios-local",
+                transport: "ittp",
+                fullAgentRuntime: false,
+                taskService: false,
+            },
+        });
+    }
+    if (method === "GET" && pathname === "/api/local-agent/capabilities") {
+        return json(localAgentCapabilities());
+    }
+    if (method === "GET" && pathname === "/api/runtime/mode") {
+        return json({
+            mode: "local",
+            deploymentRuntime: "local",
+            isRemoteController: false,
+            remoteApiBaseConfigured: false,
         });
     }
     if (method === "GET" && pathname === "/api/status") {
+        const cloud = readIosCloudPairing();
         return json({
             state: running ? "running" : "not_started",
             agentName: AGENT_NAME,
@@ -1685,10 +2114,10 @@ export async function handleIosLocalAgentRequest(request, _context = {}) {
             startedAt,
             uptime: Date.now() - startedAt,
             cloud: {
-                connectionStatus: "disconnected",
-                activeAgentId: null,
-                cloudProvisioned: false,
-                hasApiKey: false,
+                connectionStatus: cloud.paired ? "connected" : "disconnected",
+                activeAgentId: cloud.agentId,
+                cloudProvisioned: cloud.paired,
+                hasApiKey: Boolean(cloud.token),
             },
             pendingRestart: false,
             pendingRestartReasons: [],
@@ -1701,6 +2130,7 @@ export async function handleIosLocalAgentRequest(request, _context = {}) {
     if (method === "POST" && pathname === "/api/agent/restart") {
         running = true;
         startedAt = Date.now();
+        const cloud = readIosCloudPairing();
         return json({
             status: {
                 state: "running",
@@ -1709,18 +2139,38 @@ export async function handleIosLocalAgentRequest(request, _context = {}) {
                 startedAt,
                 uptime: 0,
                 cloud: {
-                    connectionStatus: "disconnected",
-                    activeAgentId: null,
-                    cloudProvisioned: false,
-                    hasApiKey: false,
+                    connectionStatus: cloud.paired ? "connected" : "disconnected",
+                    activeAgentId: cloud.agentId,
+                    cloudProvisioned: cloud.paired,
+                    hasApiKey: Boolean(cloud.token),
                 },
                 pendingRestart: false,
                 pendingRestartReasons: [],
             },
         });
     }
+    if (method === "POST" && pathname === "/api/background/run-due-tasks") {
+        return unavailableLocalBackendRoute("task_service_unavailable", {
+            reason: IOS_LOCAL_BACKGROUND_UNAVAILABLE_REASON,
+            ranTasks: 0,
+        });
+    }
+    if (method === "POST" && pathname === "/api/internal/wake") {
+        return unavailableLocalBackendRoute("task_service_unavailable", {
+            reason: IOS_LOCAL_BACKGROUND_UNAVAILABLE_REASON,
+            ranTasks: 0,
+        });
+    }
     if (method === "GET" && pathname === "/api/auth/status") {
-        return json({ required: false, pairingEnabled: false, expiresAt: null });
+        const cloud = readIosCloudPairing();
+        return json({
+            required: false,
+            pairingEnabled: false,
+            expiresAt: null,
+            cloudProvisioned: cloud.paired,
+            cloudAgentId: cloud.agentId,
+            cloudConnectionStatus: cloud.paired ? "connected" : "disconnected",
+        });
     }
     if (method === "GET" && pathname === "/api/auth/me") {
         return json({
@@ -1737,11 +2187,12 @@ export async function handleIosLocalAgentRequest(request, _context = {}) {
             },
         });
     }
-    if (method === "GET" && pathname === "/api/onboarding/status") {
+    if (method === "GET" && pathname === "/api/first-run/status") {
+        const cloud = readIosCloudPairing();
         return json({
             complete: true,
-            cloudProvisioned: false,
-            deploymentTarget: "local",
+            cloudProvisioned: cloud.paired,
+            deploymentTarget: cloud.paired ? "cloud" : "local",
         });
     }
     if (method === "GET" && pathname === "/api/config") {
@@ -1750,6 +2201,9 @@ export async function handleIosLocalAgentRequest(request, _context = {}) {
     if (method === "PUT" && pathname === "/api/config") {
         return json(localConfig());
     }
+    if (method === "POST" && pathname === "/api/cloud/chat") {
+        return handleIosCloudChat(request);
+    }
     if (method === "GET" && pathname === "/api/config/schema") {
         return json({ schema: {}, defaults: localConfig() });
     }
@@ -1818,21 +2272,272 @@ export async function handleIosLocalAgentRequest(request, _context = {}) {
             triggers: [],
             todos: [],
             autonomy: { enabled: false, thinking: false, lastEventAt: null },
+            summary: {
+                totalTasks: 0,
+                completedTasks: 0,
+                totalTriggers: 0,
+                activeTriggers: 0,
+                totalTodos: 0,
+                completedTodos: 0,
+            },
+            tasksAvailable: false,
+            triggersAvailable: false,
+            todosAvailable: false,
         });
     }
+    if (method === "GET" &&
+        (pathname === "/api/workbench/tasks" || pathname === "/api/workbench/todos")) {
+        return json(pathname.endsWith("/todos") ? { todos: [] } : { tasks: [] });
+    }
+    if (method !== "GET" &&
+        (pathname === "/api/workbench/tasks" ||
+            pathname.startsWith("/api/workbench/tasks/") ||
+            pathname === "/api/workbench/todos" ||
+            pathname.startsWith("/api/workbench/todos/"))) {
+        return unavailableLocalBackendRoute("task_service_unavailable");
+    }
+    if (method === "GET" && pathname === "/api/triggers") {
+        return json({ triggers: [] });
+    }
+    if (method === "GET" && pathname === "/api/triggers/health") {
+        return json({
+            ok: true,
+            triggersEnabled: false,
+            workflowAvailable: false,
+            reason: "The AgentRuntime trigger service is not mounted in iOS local mode.",
+        });
+    }
+    if (pathname.startsWith("/api/triggers/") ||
+        (method !== "GET" && pathname === "/api/triggers")) {
+        return unavailableLocalBackendRoute("task_service_unavailable");
+    }
+    if (method === "GET" && pathname === "/api/documents/stats") {
+        return json({
+            totalDocuments: 0,
+            totalFragments: 0,
+            totalBytes: 0,
+            bySource: {},
+        });
+    }
+    if (method === "GET" && pathname === "/api/documents") {
+        return json({
+            documents: [],
+            total: 0,
+            limit: integerFromUnknown(url.searchParams.get("limit")) ?? 100,
+            offset: integerFromUnknown(url.searchParams.get("offset")) ?? 0,
+        });
+    }
+    if (method === "GET" && pathname === "/api/documents/search") {
+        return json({ documents: [], results: [], total: 0 });
+    }
+    if (method === "GET" && pathname.startsWith("/api/documents/")) {
+        if (pathname.endsWith("/fragments")) {
+            const documentId = decodeURIComponent(pathname.slice("/api/documents/".length, -"/fragments".length));
+            return json({ documentId, fragments: [], count: 0 });
+        }
+        return json({ error: "Document not found" }, 404);
+    }
+    if (pathname.startsWith("/api/documents")) {
+        return unavailableLocalBackendRoute("document_store_unavailable");
+    }
+    if (method === "GET" && pathname === "/api/memories/stats") {
+        return json({ total: 0, byType: {}, recent: [] });
+    }
+    if (method === "GET" && pathname === "/api/mcp/config") {
+        return json({ servers: {} });
+    }
+    if (method === "GET" && pathname === "/api/mcp/status") {
+        return json({ servers: [] });
+    }
+    if (method === "GET" && pathname === "/api/mcp/marketplace/search") {
+        return json({ results: [] });
+    }
+    if (pathname.startsWith("/api/mcp/")) {
+        return unavailableLocalBackendRoute("mcp_unavailable");
+    }
+    if (method === "GET" && pathname === "/api/secrets/manager/backends") {
+        return json({
+            backends: [
+                {
+                    id: "in-house",
+                    label: "Local (encrypted)",
+                    available: false,
+                    signedIn: false,
+                    detail: "Secrets manager backend is not mounted in iOS local mode.",
+                    authMode: null,
+                },
+            ],
+        });
+    }
+    if (pathname === "/api/secrets/manager/preferences") {
+        return json({ preferences: { enabled: ["in-house"], routing: {} } });
+    }
+    if (method === "GET" && pathname === "/api/secrets/manager/install/methods") {
+        return json({ methods: [] });
+    }
+    if (method === "GET" && pathname === "/api/secrets/inventory") {
+        return json({ entries: [] });
+    }
+    if (method === "GET" && pathname === "/api/secrets/routing") {
+        return json({ rules: [] });
+    }
+    if (method === "GET" && pathname === "/api/secrets/logins") {
+        return json({ logins: [] });
+    }
+    if (pathname.startsWith("/api/secrets/")) {
+        return unavailableLocalBackendRoute("secrets_manager_unavailable");
+    }
+    if (method === "GET" && pathname === "/api/training/auto/config") {
+        return json({ enabled: false });
+    }
+    if (method === "GET" && pathname === "/api/training/auto/status") {
+        return json({ enabled: false, running: false, jobs: [] });
+    }
+    if (method === "GET" &&
+        (pathname === "/api/training/status" ||
+            pathname === "/api/training/datasets" ||
+            pathname === "/api/training/jobs" ||
+            pathname === "/api/training/models" ||
+            pathname === "/api/training/inference/endpoints")) {
+        if (pathname.endsWith("/status"))
+            return json({ available: false });
+        if (pathname.endsWith("/datasets"))
+            return json({ datasets: [] });
+        if (pathname.endsWith("/jobs"))
+            return json({ jobs: [] });
+        if (pathname.endsWith("/models"))
+            return json({ models: [] });
+        return json({ endpoints: [] });
+    }
+    if (pathname.startsWith("/api/training/")) {
+        return unavailableLocalBackendRoute("training_service_unavailable");
+    }
     if (method === "GET" &&
         (pathname === "/api/apps" || pathname === "/api/catalog/apps")) {
         return json([]);
     }
+    if (method === "GET" &&
+        (pathname === "/api/apps/search" ||
+            pathname === "/api/apps/installed" ||
+            pathname === "/api/apps/runs" ||
+            pathname === "/api/apps/plugins" ||
+            pathname === "/api/apps/plugins/search" ||
+            pathname === "/api/apps/permissions")) {
+        return json([]);
+    }
+    if (pathname === "/api/apps/favorites") {
+        if (method === "GET")
+            return json({ favoriteApps: [] });
+        if (method === "PUT")
+            return json({ favoriteApps: [] });
+    }
+    if (method === "POST" && pathname === "/api/apps/favorites/replace") {
+        return json({ favoriteApps: [] });
+    }
+    if (method === "POST" && pathname === "/api/apps/overlay-presence") {
+        return json({ ok: true });
+    }
+    if (method === "POST" &&
+        (pathname === "/api/apps/launch" ||
+            pathname === "/api/apps/create" ||
+            pathname === "/api/apps/relaunch" ||
+            pathname === "/api/apps/load-from-directory")) {
+        return unavailableLocalBackendRoute("app_manager_unavailable");
+    }
     if (method === "GET" && pathname === "/api/plugins") {
         return json({ plugins: [] });
     }
+    if (method === "GET" && pathname === "/api/plugins/installed") {
+        return json({ count: 0, plugins: [] });
+    }
+    if (method === "GET" && pathname === "/api/plugins/core") {
+        return json({ core: [], optional: [] });
+    }
+    if (method === "POST" && pathname === "/api/plugins/core/toggle") {
+        return unavailableLocalBackendRoute("plugin_loader_unavailable");
+    }
+    if (method === "POST" &&
+        (pathname === "/api/plugins/install" ||
+            pathname === "/api/plugins/update" ||
+            pathname === "/api/plugins/uninstall")) {
+        return unavailableLocalBackendRoute("plugin_loader_unavailable");
+    }
     if (method === "GET" && pathname === "/api/skills") {
         return json({ skills: [] });
     }
+    if (method === "POST" && pathname === "/api/skills/refresh") {
+        return json({ ok: true, skills: [] });
+    }
+    if (method === "GET" && pathname === "/api/skills/curated") {
+        return json({ skills: [] });
+    }
+    if ((method === "POST" || method === "DELETE") &&
+        pathname.startsWith("/api/skills/curated/")) {
+        return json({ ok: true });
+    }
+    if (method === "GET" && pathname === "/api/skills/catalog") {
+        return json({
+            total: 0,
+            page: 1,
+            perPage: 50,
+            totalPages: 0,
+            installedCount: 0,
+            skills: [],
+        });
+    }
+    if (method === "GET" && pathname === "/api/skills/catalog/search") {
+        return json({
+            query: url.searchParams.get("q") ?? "",
+            count: 0,
+            results: [],
+        });
+    }
+    if (method === "POST" && pathname === "/api/skills/catalog/refresh") {
+        return json({ ok: true, count: 0 });
+    }
+    if (method === "POST" &&
+        (pathname === "/api/skills/catalog/install" ||
+            pathname === "/api/skills/catalog/uninstall" ||
+            pathname === "/api/skills/marketplace/install" ||
+            pathname === "/api/skills/marketplace/uninstall")) {
+        return unavailableLocalBackendRoute("skill_installer_unavailable");
+    }
+    if (method === "GET" && pathname === "/api/skills/marketplace/search") {
+        return json({ ok: true, results: [] });
+    }
+    if (pathname === "/api/skills/marketplace/config") {
+        if (method === "GET" || method === "PUT")
+            return json({ keySet: false });
+    }
+    if (method === "GET" && pathname === "/api/registry/plugins") {
+        return json({ plugins: [] });
+    }
+    if (method === "GET" && pathname.startsWith("/api/registry/plugins/")) {
+        return json({ plugin: null }, 404);
+    }
+    if (method === "GET" && pathname === "/api/models") {
+        return json({
+            provider: url.searchParams.get("provider") ?? null,
+            models: [],
+        });
+    }
     if (method === "GET" && pathname === "/api/local-inference/hub") {
         return json(await hubSnapshot());
     }
+    if (method === "POST" && pathname === "/api/tts/local-inference") {
+        const body = await requestJson(request);
+        const text = typeof body.text === "string" ? sanitizeLocalSpeechText(body.text) : "";
+        if (!text)
+            return json({ error: "Missing text" }, 400);
+        const voiceReadiness = inferVoiceReadiness(await listInstalledModels(), readBundleIndex());
+        return json({
+            error: voiceReadiness.message,
+            code: voiceReadiness.status === "unavailable"
+                ? "ios_local_tts_executor_missing"
+                : "ios_local_voice_assets_missing",
+            voiceReadiness,
+        }, 503);
+    }
     if (method === "GET" && pathname === "/api/local-inference/hardware") {
         return json(await hardwareProbe());
     }