diff --git a/Sources/Config/DeviceCapability.swift b/Sources/Config/DeviceCapability.swift
new file mode 100644
index 0000000..e1ac6a9
--- /dev/null
+++ b/Sources/Config/DeviceCapability.swift
@@ -0,0 +1,220 @@
+import Foundation
+import IOKit
+
+enum DeviceCapability {
+    struct Info: Sendable {
+        let chipName: String
+        let totalRAMGB: Double
+        let gpuCoreCount: Int
+        let neuralEngineCoreCount: Int
+        let availableDiskGB: Double
+        let totalDiskGB: Double
+    }
+
+    static let current: Info = {
+        let ram = Double(ProcessInfo.processInfo.physicalMemory) / 1_073_741_824
+        let (chipName, gpuCores, neCores) = readAppleSiliconInfo()
+        let (availDisk, totalDisk) = diskSpace()
+        return Info(
+            chipName: chipName,
+            totalRAMGB: ram,
+            gpuCoreCount: gpuCores,
+            neuralEngineCoreCount: neCores,
+            availableDiskGB: availDisk,
+            totalDiskGB: totalDisk
+        )
+    }()
+
+    // MARK: - Model Compatibility
+
+    enum Compatibility: Sendable {
+        case compatible
+        case marginal(String)
+        case incompatible(String)
+    }
+
+    struct ModelRequirements: Sendable {
+        let minRAMGB: Double
+        let recommendedRAMGB: Double
+        let diskGB: Double
+    }
+
+    static func requirements(for modelID: String, downloadSizeBytes: Int64?) -> ModelRequirements {
+        if let known = knownRequirements[modelID] { return known }
+        let diskGB = downloadSizeBytes.map { Double($0) / 1_073_741_824 } ?? 0
+        let ramNeeded = max(diskGB * 1.2, 1)
+        return ModelRequirements(
+            minRAMGB: ramNeeded,
+            recommendedRAMGB: ramNeeded * 1.5,
+            diskGB: diskGB
+        )
+    }
+
+    static func check(
+        modelID: String,
+        downloadSizeBytes: Int64?
+    ) -> Compatibility {
+        let reqs = requirements(for: modelID, downloadSizeBytes: downloadSizeBytes)
+        let info = current
+
+        if reqs.diskGB > 0, info.availableDiskGB < reqs.diskGB * 1.1 {
+            return .incompatible(L("device.insufficient_disk"))
+        }
+        if info.totalRAMGB < reqs.minRAMGB {
+            return .incompatible(L("device.insufficient_ram"))
+        }
+        if info.totalRAMGB < reqs.recommendedRAMGB {
+            return .marginal(L("device.marginal_ram"))
+        }
+        return .compatible
+    }
+
+    static func recommendedModelID(from candidates: [String]) -> String? {
+        let ram = current.totalRAMGB
+        for id in candidates.reversed() {
+            if let reqs = knownRequirements[id], ram >= reqs.recommendedRAMGB {
+                return id
+            }
+        }
+        return candidates.first
+    }
+
+    // MARK: - Hardware Detection
+
+    private static func readAppleSiliconInfo() -> (chipName: String, gpuCores: Int, neCores: Int) {
+        var chipName = sysctlString("machdep.cpu.brand_string") ?? "Apple Silicon"
+        if chipName.hasPrefix("Apple ") {
+            chipName = String(chipName.dropFirst(6))
+        }
+
+        let gpuCores = ioRegistryInt(className: "AGXAccelerator", key: "gpu-core-count")
+            ?? ioRegistryInt(className: "AGXAcceleratorG13G", key: "gpu-core-count")
+            ?? gpuCoresFromChipName(chipName)
+        let neCores = neuralEngineCoresFromChipName(chipName)
+        return (chipName, gpuCores, neCores)
+    }
+
+    private static func diskSpace() -> (available: Double, total: Double) {
+        guard let attrs = try? FileManager.default.attributesOfFileSystem(
+            forPath: NSHomeDirectory()
+        ) else { return (0, 0) }
+        let avail = (attrs[.systemFreeSize] as? NSNumber)?.doubleValue ?? 0
+        let total = (attrs[.systemSize] as? NSNumber)?.doubleValue ?? 0
+        return (avail / 1_073_741_824, total / 1_073_741_824)
+    }
+
+    private static func sysctlString(_ name: String) -> String? {
+        var size = 0
+        guard sysctlbyname(name, nil, &size, nil, 0) == 0, size > 0 else { return nil }
+        var buf = [CChar](repeating: 0, count: size)
+        guard sysctlbyname(name, &buf, &size, nil, 0) == 0 else { return nil }
+        return String(cString: buf)
+    }
+
+    private static func ioRegistryInt(className: String, key: String) -> Int? {
+        var iterator: io_iterator_t = 0
+        guard IOServiceGetMatchingServices(
+            kIOMainPortDefault,
+            IOServiceMatching(className),
+            &iterator
+        ) == KERN_SUCCESS else { return nil }
+        defer { IOObjectRelease(iterator) }
+
+        var entry = IOIteratorNext(iterator)
+        while entry != 0 {
+            defer { IOObjectRelease(entry); entry = IOIteratorNext(iterator) }
+            if let cfValue = IORegistryEntryCreateCFProperty(
+                entry, key as CFString, kCFAllocatorDefault, 0
+            )?.takeRetainedValue() as? NSNumber {
+                return cfValue.intValue
+            }
+        }
+        return nil
+    }
+
+    private static func gpuCoresFromChipName(_ name: String) -> Int {
+        let n = name.lowercased()
+        if n.contains("ultra") { return 76 }
+        if n.contains("max") { return 40 }
+        if n.contains("pro") { return 18 }
+        return 10
+    }
+
+    private static func neuralEngineCoresFromChipName(_ name: String) -> Int {
+        let n = name.lowercased()
+        if n.contains("m4") || n.contains("m3") { return 16 }
+        return 16
+    }
+
+    // MARK: - Known Model Requirements (RAM needed at runtime, disk for download)
+
+    private static let knownRequirements: [String: ModelRequirements] = [
+        "mlx-community/Qwen3.5-0.8B-MLX-4bit":
+            ModelRequirements(minRAMGB: 2, recommendedRAMGB: 4, diskGB: 0.7),
+        "mlx-community/Qwen3.5-2B-4bit":
+            ModelRequirements(minRAMGB: 4, recommendedRAMGB: 6, diskGB: 1.8),
+        "mlx-community/Qwen3.5-9B-5bit":
+            ModelRequirements(minRAMGB: 10, recommendedRAMGB: 12, diskGB: 7.1),
+        "mlx-community/Qwen3-30B-A3B-4bit":
+            ModelRequirements(minRAMGB: 12, recommendedRAMGB: 18, diskGB: 17.2),
+        "mlx-community/Qwen3.5-35B-A3B-4bit":
+            ModelRequirements(minRAMGB: 16, recommendedRAMGB: 24, diskGB: 20.5),
+        "mlx-community/Qwen2.5-0.5B-Instruct-4bit":
+            ModelRequirements(minRAMGB: 2, recommendedRAMGB: 4, diskGB: 0.3),
+        "mlx-community/Qwen2.5-1.5B-Instruct-4bit":
+            ModelRequirements(minRAMGB: 2, recommendedRAMGB: 4, diskGB: 0.9),
+        "mlx-community/Qwen2.5-3B-Instruct-4bit":
+            ModelRequirements(minRAMGB: 4, recommendedRAMGB: 6, diskGB: 1.8),
+        "mlx-community/Qwen3-0.6B-4bit":
+            ModelRequirements(minRAMGB: 2, recommendedRAMGB: 4, diskGB: 0.4),
+        "mlx-community/Qwen3-1.7B-4bit":
+            ModelRequirements(minRAMGB: 2, recommendedRAMGB: 4, diskGB: 1.0),
+        "mlx-community/Qwen3-4B-4bit":
+            ModelRequirements(minRAMGB: 4, recommendedRAMGB: 6, diskGB: 2.3),
+        "mlx-community/gemma-4-e2b-it-4bit":
+            ModelRequirements(minRAMGB: 6, recommendedRAMGB: 8, diskGB: 3.6),
+        "mlx-community/gemma-4-e4b-it-4bit":
+            ModelRequirements(minRAMGB: 8, recommendedRAMGB: 10, diskGB: 5.3),
+        "mlx-community/gemma-3-1b-it-4bit":
+            ModelRequirements(minRAMGB: 2, recommendedRAMGB: 4, diskGB: 0.8),
+        "mlx-community/gemma-3-4b-it-4bit":
+            ModelRequirements(minRAMGB: 6, recommendedRAMGB: 8, diskGB: 3.5),
+        "mlx-community/gemma-3-12b-it-4bit":
+            ModelRequirements(minRAMGB: 12, recommendedRAMGB: 16, diskGB: 8.1),
+        "mlx-community/Llama-4-Scout-17B-16E-Instruct-4bit":
+            ModelRequirements(minRAMGB: 48, recommendedRAMGB: 64, diskGB: 61.2),
+        "mlx-community/Llama-4-Maverick-17B-128E-Instruct-4bit":
+            ModelRequirements(minRAMGB: 128, recommendedRAMGB: 192, diskGB: 226.0),
+    ]
+}
+
+extension DeviceCapability.Info {
+    var chipDisplayName: String { chipName }
+    var ramDisplayText: String { String(format: "%.0f GB", totalRAMGB) }
+    var gpuDisplayText: String { "\(gpuCoreCount)-core GPU" }
+    var diskAvailableText: String { String(format: "%.1f GB", availableDiskGB) }
+}
+
+extension DeviceCapability.Compatibility {
+    var isCompatible: Bool {
+        if case .compatible = self { return true }
+        return false
+    }
+
+    var isMarginal: Bool {
+        if case .marginal = self { return true }
+        return false
+    }
+
+    var isIncompatible: Bool {
+        if case .incompatible = self { return true }
+        return false
+    }
+
+    var message: String? {
+        switch self {
+        case .compatible: return nil
+        case .marginal(let msg), .incompatible(let msg): return msg
+        }
+    }
+}
diff --git a/Sources/Config/ModelCatalog.swift b/Sources/Config/ModelCatalog.swift
index eb6f54e..3daab96 100644
--- a/Sources/Config/ModelCatalog.swift
+++ b/Sources/Config/ModelCatalog.swift
@@ -54,6 +54,7 @@ final class ModelCatalog: ObservableObject {
         var downloadDetail: String = ""
         var benchmarkTPS: Double?
         var isBenchmarking: Bool = false
+        var compatibility: DeviceCapability.Compatibility = .compatible
 
         static func == (lhs: Self, rhs: Self) -> Bool {
             lhs.id == rhs.id && lhs.status == rhs.status &&
@@ -210,6 +211,10 @@ final class ModelCatalog: ObservableObject {
                         : (size > 0 ? .error(L("model.download_incomplete")) : .notDownloaded)
                 }
             }
+            llmModels[i].compatibility = DeviceCapability.check(
+                modelID: id,
+                downloadSizeBytes: Self.defaultDownloadEstimateBytes(for: id)
+            )
         }
         refreshASRStatus(recheckingErrors: recheckingErrors)
     }
diff --git a/Sources/Resources/en.lproj/Localizable.strings b/Sources/Resources/en.lproj/Localizable.strings
index 89bf4d3..a4b09b8 100644
--- a/Sources/Resources/en.lproj/Localizable.strings
+++ b/Sources/Resources/en.lproj/Localizable.strings
@@ -485,3 +485,14 @@
 "error.llm_not_downloaded" = "The model files have not been downloaded. Open Settings → Models and confirm the data usage first.";
 "onboarding.download_notice" = "No download starts automatically. Review the size, then confirm if you want this local model.";
 "onboarding.download_size" = "Download (%@)";
+
+/* ── Device Capability ── */
+"device.title" = "This Mac";
+"device.chip" = "Chip";
+"device.ram" = "Memory";
+"device.disk_available" = "Disk Available";
+"device.insufficient_disk" = "Not enough free disk space to download this model. Free up storage or choose a smaller model.";
+"device.insufficient_ram" = "This Mac does not have enough memory to run this model. Choose a smaller model or use a remote LLM.";
+"device.marginal_ram" = "This model may run slowly on this Mac due to limited memory. A smaller model is recommended for better performance.";
+"device.badge_marginal" = "May Be Slow";
+"device.badge_incompatible" = "Too Large";
diff --git a/Sources/Resources/zh-Hans.lproj/Localizable.strings b/Sources/Resources/zh-Hans.lproj/Localizable.strings
index a1fc182..d0ec235 100644
--- a/Sources/Resources/zh-Hans.lproj/Localizable.strings
+++ b/Sources/Resources/zh-Hans.lproj/Localizable.strings
@@ -485,3 +485,14 @@
 "error.llm_not_downloaded" = "模型文件尚未下载。请前往 设置 → 模型，确认流量后下载";
 "onboarding.download_notice" = "这里不会自动下载。请先确认体积，再决定是否下载这个本地模型。";
 "onboarding.download_size" = "下载（%@）";
+
+/* ── 设备性能检查 ── */
+"device.title" = "本机信息";
+"device.chip" = "芯片";
+"device.ram" = "内存";
+"device.disk_available" = "可用磁盘";
+"device.insufficient_disk" = "磁盘空间不足，无法下载此模型。请清理空间或选择更小的模型。";
+"device.insufficient_ram" = "本机内存不足以运行此模型。请选择更小的模型，或使用远程 LLM。";
+"device.marginal_ram" = "本机内存有限，运行此模型可能较慢。建议选择更小的模型以获得更好的性能。";
+"device.badge_marginal" = "可能较慢";
+"device.badge_incompatible" = "超出配置";
diff --git a/Sources/UI/ModelManagementActions.swift b/Sources/UI/ModelManagementActions.swift
index 8d6f272..b11b8ee 100644
--- a/Sources/UI/ModelManagementActions.swift
+++ b/Sources/UI/ModelManagementActions.swift
@@ -161,12 +161,16 @@ extension ModelManagementView {
         let estimate = estimatedDownloadBytes(for: action.model, type: action.type)
         let remaining = estimate.map { max($0 - action.model.cacheSize, 0) }
         let sizeText = remaining.map(ModelCatalog.formatBytes) ?? L("download.unknown")
-        return String(
+        var message = String(
             format: L("model.download_confirm_message"),
             action.model.displayName,
             sizeText,
             ModelStorage.root.path
         )
+        if let warning = action.model.compatibility.message {
+            message += "\n\n⚠️ " + warning
+        }
+        return message
     }
 
     private func estimatedDownloadBytes(
diff --git a/Sources/UI/ModelManagementRows.swift b/Sources/UI/ModelManagementRows.swift
index aba2252..6cb9394 100644
--- a/Sources/UI/ModelManagementRows.swift
+++ b/Sources/UI/ModelManagementRows.swift
@@ -53,6 +53,7 @@ extension ModelManagementView {
                                 .foregroundStyle(Color.accentColor)
                                 .clipShape(Capsule())
                         }
+                        compatibilityBadge(model.compatibility)
                     }
                     HStack(spacing: 6) {
                         Text(secondaryText(for: model))
@@ -92,6 +93,32 @@ extension ModelManagementView {
         .background(isActive ? Color.accentColor.opacity(0.04) : .clear)
     }
 
+    @ViewBuilder
+    func compatibilityBadge(_ compat: DeviceCapability.Compatibility) -> some View {
+        switch compat {
+        case .compatible:
+            EmptyView()
+        case .marginal(let msg):
+            Text(L("device.badge_marginal"))
+                .font(.system(size: 9, weight: .medium))
+                .padding(.horizontal, 5)
+                .padding(.vertical, 1)
+                .background(Color.orange.opacity(0.15))
+                .foregroundStyle(.orange)
+                .clipShape(Capsule())
+                .help(msg)
+        case .incompatible(let msg):
+            Text(L("device.badge_incompatible"))
+                .font(.system(size: 9, weight: .medium))
+                .padding(.horizontal, 5)
+                .padding(.vertical, 1)
+                .background(Color.red.opacity(0.15))
+                .foregroundStyle(.red)
+                .clipShape(Capsule())
+                .help(msg)
+        }
+    }
+
     @ViewBuilder
     func busyModelStatus(_ model: ModelCatalog.ModelEntry) -> some View {
         VStack(alignment: .leading, spacing: 4) {
diff --git a/Sources/UI/ModelManagementSections.swift b/Sources/UI/ModelManagementSections.swift
index 51d3793..4a34ac2 100644
--- a/Sources/UI/ModelManagementSections.swift
+++ b/Sources/UI/ModelManagementSections.swift
@@ -2,6 +2,55 @@ import AppKit
 import SwiftUI
 
 extension ModelManagementView {
+    var deviceInfoSection: some View {
+        let info = DeviceCapability.current
+        return VStack(alignment: .leading, spacing: 10) {
+            Label(L("device.title"), systemImage: "cpu")
+                .font(.headline)
+
+            HStack(spacing: 16) {
+                deviceInfoItem(
+                    icon: "memorychip",
+                    label: L("device.chip"),
+                    value: info.chipDisplayName
+                )
+                Divider().frame(height: 28)
+                deviceInfoItem(
+                    icon: "memorychip.fill",
+                    label: L("device.ram"),
+                    value: info.ramDisplayText
+                )
+                Divider().frame(height: 28)
+                deviceInfoItem(
+                    icon: "gpu",
+                    label: "GPU",
+                    value: info.gpuDisplayText
+                )
+                Divider().frame(height: 28)
+                deviceInfoItem(
+                    icon: "internaldrive",
+                    label: L("device.disk_available"),
+                    value: info.diskAvailableText
+                )
+            }
+        }
+    }
+
+    private func deviceInfoItem(icon: String, label: String, value: String) -> some View {
+        HStack(spacing: 6) {
+            Image(systemName: icon)
+                .font(.system(size: 12))
+                .foregroundStyle(.secondary)
+            VStack(alignment: .leading, spacing: 1) {
+                Text(label)
+                    .font(.system(size: 9))
+                    .foregroundStyle(.tertiary)
+                Text(value)
+                    .font(.system(size: 11, weight: .medium, design: .monospaced))
+            }
+        }
+    }
+
     var storageSection: some View {
         VStack(alignment: .leading, spacing: 8) {
             Label(L("model.storage.title"), systemImage: "externaldrive")
diff --git a/Sources/UI/ModelManagementView.swift b/Sources/UI/ModelManagementView.swift
index 6af2ae7..322859e 100644
--- a/Sources/UI/ModelManagementView.swift
+++ b/Sources/UI/ModelManagementView.swift
@@ -36,6 +36,8 @@ struct ModelManagementView: View {
                         SettingsPanel { activeDownloadsSection }
                     }
 
+                    SettingsPanel { deviceInfoSection }
+
                     SettingsPanel {
                         enginePickerSection
                         Divider().padding(.vertical, 4)
diff --git a/docs/index.html b/docs/index.html
index ee4700d..507c2d8 100644
--- a/docs/index.html
+++ b/docs/index.html
@@ -37,10 +37,10 @@
   <link rel="stylesheet" href="assets/styles.css?v=20260819-terminal" />
 
   <script type="application/ld+json">
-  {"@context":"https://schema.org","@type":"SoftwareApplication","name":"Utter","applicationCategory":"UtilitiesApplication","applicationSubCategory":"Voice Input & Dictation","operatingSystem":"macOS 26+","runtimePlatform":"Apple Silicon","url":"https://utter.idevlab.dev/","downloadUrl":"https://github.com/IchenDEV/utter/releases/latest","codeRepository":"https://github.com/IchenDEV/utter","license":"https://github.com/IchenDEV/utter/blob/main/LICENSE","offers":{"@type":"Offer","price":"0","priceCurrency":"USD"},"description":"Utter is a native macOS menu bar app for AI-powered voice input and dictation. Hold a hotkey, speak, and polished text appears at your cursor. WhisperKit and Apple Speech run on-device, with MLX LLM cleanup on Apple Silicon. Optional remote LLM providers.","featureList":["On-device speech recognition with WhisperKit, Apple Speech, or Qwen3-ASR","LLM text cleanup with local MLX Qwen2.5/Qwen3 or remote providers","Screen context OCR via ScreenCaptureKit and Vision for homophone correction","Voice command mode that acts on what is on screen","Global hotkey works in any macOS app","Bilingual UI in English and Simplified Chinese","Input history, memory, and personal edit rules"],"author":{"@type":"Organization","name":"Utter","url":"https://github.com/IchenDEV/utter"}}
+  {"@context":"https://schema.org","@type":"SoftwareApplication","name":"Utter","applicationCategory":"UtilitiesApplication","applicationSubCategory":"Voice Input & Dictation","operatingSystem":"macOS 26+","runtimePlatform":"Apple Silicon","url":"https://utter.idevlab.dev/","downloadUrl":"https://github.com/IchenDEV/utter/releases/latest","codeRepository":"https://github.com/IchenDEV/utter","license":"https://github.com/IchenDEV/utter/blob/main/LICENSE","offers":{"@type":"Offer","price":"0","priceCurrency":"USD"},"description":"Utter is a native macOS menu bar app for AI-powered voice input and dictation. Hold a hotkey, speak, and polished text appears at your cursor. WhisperKit and Apple Speech run on-device, with MLX LLM cleanup on Apple Silicon. Optional remote LLM providers.","featureList":["On-device speech recognition with WhisperKit, Apple Speech, or Qwen3-ASR","LLM text cleanup with local MLX Qwen2.5/Qwen3 or remote providers","Screen context OCR via ScreenCaptureKit and Vision for homophone correction","Voice command mode that acts on what is on screen","Global hotkey works in any macOS app","Bilingual UI in English and Simplified Chinese","Input history, memory, and personal edit rules","Device performance check with model compatibility recommendations"],"author":{"@type":"Organization","name":"Utter","url":"https://github.com/IchenDEV/utter"}}
   </script>
   <script type="application/ld+json">
-  {"@context":"https://schema.org","@type":"FAQPage","mainEntity":[{"@type":"Question","name":"What is Utter?","acceptedAnswer":{"@type":"Answer","text":"Utter is a native macOS menu bar app that turns your voice into polished text in any app. Hold a hotkey, speak, and the transcribed, cleaned-up text is typed directly at your cursor."}},{"@type":"Question","name":"Does Utter work offline?","acceptedAnswer":{"@type":"Answer","text":"Yes. With the WhisperKit or Apple Speech engine plus a local MLX model, Utter can transcribe and refine text entirely on-device. Remote LLM providers are optional."}},{"@type":"Question","name":"Which macOS version and hardware does Utter require?","acceptedAnswer":{"@type":"Answer","text":"macOS 26 or later on Apple Silicon. Intel Macs are not supported."}},{"@type":"Question","name":"How is Utter different from Apple's built-in dictation?","acceptedAnswer":{"@type":"Answer","text":"Utter adds LLM cleanup, screen context via OCR, voice command mode, and global hotkey insertion into any app."}},{"@type":"Question","name":"Which apps does Utter work in?","acceptedAnswer":{"@type":"Answer","text":"Any macOS app that accepts text input. Utter uses the Accessibility API to insert text at your cursor, with a clipboard-paste fallback."}},{"@type":"Question","name":"Is Utter free and open source?","acceptedAnswer":{"@type":"Answer","text":"Yes. Utter is MIT-licensed and the full source is on GitHub."}},{"@type":"Question","name":"Which languages does Utter support?","acceptedAnswer":{"@type":"Answer","text":"Speech recognition supports many languages via WhisperKit and Apple Speech. The app UI is bilingual in English and Simplified Chinese."}}]}
+  {"@context":"https://schema.org","@type":"FAQPage","mainEntity":[{"@type":"Question","name":"What is Utter?","acceptedAnswer":{"@type":"Answer","text":"Utter is a native macOS menu bar app that turns your voice into polished text in any app. Hold a hotkey, speak, and the transcribed, cleaned-up text is typed directly at your cursor."}},{"@type":"Question","name":"Does Utter work offline?","acceptedAnswer":{"@type":"Answer","text":"Yes. With the WhisperKit or Apple Speech engine plus a local MLX model, Utter can transcribe and refine text entirely on-device. Remote LLM providers are optional."}},{"@type":"Question","name":"Which macOS version and hardware does Utter require?","acceptedAnswer":{"@type":"Answer","text":"macOS 26 or later on Apple Silicon. Intel Macs are not supported. Utter checks your chip, RAM, GPU, and disk space and shows which models are compatible with your Mac."}},{"@type":"Question","name":"How does Utter decide which models I can run?","acceptedAnswer":{"@type":"Answer","text":"Utter reads your Mac's chip type, unified memory, GPU core count, and available disk space. Models that exceed your hardware are flagged with a warning badge. Smaller models are recommended for Macs with 8 GB of RAM, while larger MoE models require 16 GB or more."}},{"@type":"Question","name":"How is Utter different from Apple's built-in dictation?","acceptedAnswer":{"@type":"Answer","text":"Utter adds LLM cleanup, screen context via OCR, voice command mode, and global hotkey insertion into any app."}},{"@type":"Question","name":"Which apps does Utter work in?","acceptedAnswer":{"@type":"Answer","text":"Any macOS app that accepts text input. Utter uses the Accessibility API to insert text at your cursor, with a clipboard-paste fallback."}},{"@type":"Question","name":"Is Utter free and open source?","acceptedAnswer":{"@type":"Answer","text":"Yes. Utter is MIT-licensed and the full source is on GitHub."}},{"@type":"Question","name":"Which languages does Utter support?","acceptedAnswer":{"@type":"Answer","text":"Speech recognition supports many languages via WhisperKit and Apple Speech. The app UI is bilingual in English and Simplified Chinese."}}]}
   </script>
   <script type="application/ld+json">
   {"@context":"https://schema.org","@type":"HowTo","name":"Dictate text with Utter","description":"Hold a key, speak, and release to insert polished text at your cursor in any macOS app.","totalTime":"PT5S","step":[{"@type":"HowToStep","position":1,"name":"Hold your hotkey","text":"Press and hold your configured key. Recording starts instantly."},{"@type":"HowToStep","position":2,"name":"Speak naturally","text":"Talk the way you normally would. Audio is captured on your Mac."},{"@type":"HowToStep","position":3,"name":"Release to insert","text":"Release the key. Utter transcribes, cleans up the text with the LLM, and types the result at your cursor."}]}
@@ -57,7 +57,7 @@
 
     <nav class="site-nav" aria-label="Primary navigation">
       <a class="nav-brand" href="#hero" aria-label="Utter home"><img src="assets/opentype-icon.png" alt="" /><strong>Utter</strong></a>
-      <div class="nav-links"><a href="#pipeline">Pipeline</a><a href="#features">Features</a><a href="#privacy">Privacy</a><a href="#faq">FAQ</a><a href="https://github.com/IchenDEV/utter">GitHub</a></div>
+      <div class="nav-links"><a href="#pipeline">Pipeline</a><a href="#features">Features</a><a href="#hardware">Hardware</a><a href="#privacy">Privacy</a><a href="#faq">FAQ</a><a href="https://github.com/IchenDEV/utter">GitHub</a></div>
       <a class="nav-download" href="https://github.com/IchenDEV/utter/releases/latest">Download for macOS</a>
     </nav>
 
@@ -137,9 +137,25 @@ <h2>The features are the pipeline.</h2>
             <li><span>⌘</span><h3>Voice command mode</h3><p>Edit, summarize, translate, or reply without switching away from the current app.</p></li>
             <li><span>◇</span><h3>Local and remote models</h3><p>Run fully on-device, or choose a remote LLM provider when it suits your workflow.</p></li>
             <li><span>◎</span><h3>Bilingual UI</h3><p>Switch the interface between English and Simplified Chinese.</p></li>
+            <li><span>⚙</span><h3>Smart model recommendations</h3><p>Utter checks your Mac's chip, memory, GPU, and disk space, then highlights which models run well and warns about those that won't.</p></li>
           </ul>
         </section>
 
+        <section id="hardware" class="content-section">
+          <p class="command">$ utter hardware --check</p>
+          <h2>Pick the right model for your Mac.</h2>
+          <p class="hero-sub" style="margin-bottom:24px">Utter pre-checks CPU, GPU, RAM, and disk space so you always know which models fit your hardware before downloading.</p>
+          <div class="tech-list" style="border-top:1px solid var(--line);border-left:1px solid var(--line)">
+            <div><dt>Qwen3.5 0.8B</dt><dd>2 GB RAM · 0.7 GB disk · Any Apple Silicon</dd></div>
+            <div><dt>Qwen3.5 2B</dt><dd>4 GB RAM · 1.8 GB disk · M1 or later</dd></div>
+            <div><dt>Qwen3.5 9B</dt><dd>10 GB RAM · 7.1 GB disk · 16 GB Mac recommended</dd></div>
+            <div><dt>Qwen3 30B-A3B</dt><dd>12 GB RAM · 17.2 GB disk · 18 GB+ Mac recommended</dd></div>
+            <div><dt>Gemma 4 E2B</dt><dd>6 GB RAM · 3.6 GB disk · 8 GB Mac or above</dd></div>
+            <div><dt>Llama 4 Scout</dt><dd>48 GB RAM · 61.2 GB disk · M-series Max/Ultra only</dd></div>
+          </div>
+          <p style="margin-top:14px;color:var(--muted);font-size:11px">Models that exceed your hardware are flagged in Settings → Models. You can always use a remote LLM provider instead.</p>
+        </section>
+
         <section id="privacy" class="content-section privacy-section">
           <p class="command">$ utter privacy --local</p>
           <div class="privacy-grid"><div><h2>Private by default.<br />Local when you want it.</h2><p>WhisperKit transcribes on-device. MLX-Swift-LM can refine text locally. Remote providers are optional.</p></div>
@@ -160,7 +176,8 @@ <h2>The features are the pipeline.</h2>
           <div class="faq-list">
             <details open><summary>What is Utter?</summary><p>Utter is a native macOS menu bar app that turns your voice into polished text in any app. Hold a hotkey, speak, and the result is typed directly at your cursor.</p></details>
             <details><summary>Does Utter work offline?</summary><p>Yes. With WhisperKit or Apple Speech plus a local MLX model, Utter can transcribe and refine text entirely on-device. Remote LLM providers are optional.</p></details>
-            <details><summary>Which macOS version and hardware does Utter require?</summary><p>macOS 26 or later on Apple Silicon. Intel Macs are not supported.</p></details>
+            <details><summary>Which macOS version and hardware does Utter require?</summary><p>macOS 26 or later on Apple Silicon. Intel Macs are not supported. Utter checks your chip, RAM, GPU, and disk space and shows which models are compatible with your Mac.</p></details>
+            <details><summary>How does Utter decide which models I can run?</summary><p>Utter reads your Mac's chip type, unified memory, GPU core count, and available disk space. Models that exceed your hardware are flagged with a warning badge. Smaller models are recommended for Macs with 8 GB of RAM, while larger MoE models require 16 GB or more.</p></details>
             <details><summary>Which apps does Utter work in?</summary><p>Any macOS app that accepts text input. Utter uses the Accessibility API, with a clipboard-paste fallback.</p></details>
             <details><summary>Is Utter free and open source?</summary><p>Yes. Utter is MIT-licensed and the full source is on GitHub.</p></details>
           </div>