agent-cli 0.98.0__tar.gz → 0.98.2__tar.gz

{agent_cli-0.98.0 → agent_cli-0.98.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-cli
-Version: 0.98.0
+Version: 0.98.2
 Summary: A suite of AI-powered command-line tools for text correction, audio transcription, and voice assistance.
 Project-URL: Homepage, https://github.com/basnijholt/agent-cli
 Author-email: Bas Nijholt <bas@nijho.lt>

{agent_cli-0.98.0 → agent_cli-0.98.2}/agent_cli/core/deps.py

@@ -283,7 +283,41 @@ def _is_uv_cmd(cmd: list[str]) -> bool:
 
 def _uv_override_path(extra: str) -> Path:
     """Return the override file path for an extra."""
-    return _OVERRIDES_DIR / f"{extra}.txt"
+    source = _OVERRIDES_DIR / f"{extra}.txt"
+    if not _path_contains_whitespace(source):
+        return source
+    # uv mishandles --overrides paths with spaces (for example, app bundles
+    # installed under "Application Support"), so hand it a copied no-space path.
+    return _materialize_uv_override(source)
+
+
+def _path_contains_whitespace(path: Path) -> bool:
+    return any(character.isspace() for character in str(path))
+
+
+def _materialized_uv_overrides_dir() -> Path:
+    configured_dir = os.environ.get("AGENTCLI_UV_OVERRIDES_DIR")
+    if configured_dir:
+        configured_path = Path(configured_dir).expanduser()
+        if not _path_contains_whitespace(configured_path):
+            return configured_path
+
+    # Keep this outside AGENTCLI_APP_SUPPORT_DIR; that path usually contains a
+    # space on macOS and would reintroduce the launchd/uv parse failure.
+    return Path.home() / ".cache" / "agent-cli" / "uv-overrides"
+
+
+def _materialize_uv_override(source: Path) -> Path:
+    """Copy a bundled override to a no-space path that uv can parse."""
+    target = _materialized_uv_overrides_dir() / source.name
+    try:
+        source_bytes = source.read_bytes()
+        target.parent.mkdir(parents=True, exist_ok=True)
+        if not target.exists() or target.read_bytes() != source_bytes:
+            target.write_bytes(source_bytes)
+    except OSError:
+        return source
+    return target
 
 
 def _supports_uv_runtime_override(

{agent_cli-0.98.0 → agent_cli-0.98.2}/agent_cli/install/launchd.py

@@ -88,16 +88,18 @@ def _get_recent_logs(service_name: str, num_lines: int = 10) -> list[str]:
 
     lines: list[str] = []
 
-    # Prefer stdout as it has structured output (startup message, usage examples)
-    # stderr often has noisy warnings from libraries like PyTorch/Kokoro
     for log_file in [stdout_log, stderr_log]:
         if log_file.exists():
             try:
                 with log_file.open() as f:
                     all_lines = f.readlines()
-                    lines = [line.rstrip() for line in all_lines[-num_lines:]]
+                    recent_lines = [line.rstrip() for line in all_lines[-num_lines:]]
+                    if not recent_lines:
+                        continue
                     if lines:
-                        break
+                        lines.append("")
+                    lines.append(f"==> {log_file.name} <==")
+                    lines.extend(recent_lines)
             except OSError:
                 continue
 

{agent_cli-0.98.0 → agent_cli-0.98.2}/agent_cli/server/cli.py

@@ -351,7 +351,7 @@ def whisper_cmd(  # noqa: C901, PLR0912, PLR0915
             "--device",
             "-d",
             help=(
-                "Compute device: `auto` (detect GPU), `cuda`, `cuda:0`, `cpu`. "
+                "Compute device: `auto` (detect GPU), `cuda`, `cuda:0`, `mps`, `cpu`. "
                 "MLX backend always uses Apple Silicon"
             ),
         ),

{agent_cli-0.98.0 → agent_cli-0.98.2}/agent_cli/server/whisper/backends/nemo.py

@@ -6,6 +6,8 @@ import asyncio
 import inspect
 import io
 import logging
+import os
+import sys
 import tempfile
 import time
 import wave
@@ -89,18 +91,35 @@ _state = _SubprocessState()
 
 def _resolve_device(device: str) -> str:
     """Resolve the runtime device for NeMo inference."""
+    if device in {"auto", "mps"} and sys.platform == "darwin":
+        os.environ.setdefault("PYTORCH_ENABLE_MPS_FALLBACK", "1")
+
     import torch  # noqa: PLC0415
 
     if device == "auto":
-        return "cuda" if torch.cuda.is_available() else "cpu"
+        if torch.cuda.is_available():
+            return "cuda"
+        if _mps_available(torch):
+            return "mps"
+        return "cpu"
 
     if device.startswith("cuda") and not torch.cuda.is_available():
         msg = f"CUDA device requested ({device}) but CUDA is not available"
         raise RuntimeError(msg)
 
+    if device == "mps" and not _mps_available(torch):
+        msg = "MPS device requested (mps) but MPS is not available"
+        raise RuntimeError(msg)
+
     return device
 
 
+def _mps_available(torch: Any) -> bool:
+    """Return True when PyTorch exposes an available MPS backend."""
+    mps = getattr(getattr(torch, "backends", None), "mps", None)
+    return bool(mps is not None and mps.is_available())
+
+
 def _extract_text(hypothesis: Any) -> str:
     """Extract a transcript string from a NeMo hypothesis-like object."""
     if isinstance(hypothesis, str):

{agent_cli-0.98.0 → agent_cli-0.98.2}/docs/commands/server/whisper.md

@@ -112,7 +112,7 @@ agent-cli server whisper \
 |--------|---------|-------------|
 | `--model, -m` | - | Whisper model(s) to load. Common models: `tiny`, `base`, `small`, `medium`, `large-v3`, `distil-large-v3`, `parakeet-tdt-0.6b-v3`, `parakeet-unified-en-0.6b` (NeMo backend). Can specify multiple for different accuracy/speed tradeoffs. Default: `large-v3` (`parakeet-unified-en-0.6b` with `--backend nemo`) |
 | `--default-model` | - | Model to use when client doesn't specify one. Must be in the `--model` list |
-| `--device, -d` | `auto` | Compute device: `auto` (detect GPU), `cuda`, `cuda:0`, `cpu`. MLX backend always uses Apple Silicon |
+| `--device, -d` | `auto` | Compute device: `auto` (detect GPU), `cuda`, `cuda:0`, `mps`, `cpu`. MLX backend always uses Apple Silicon |
 | `--compute-type` | `auto` | Precision for faster-whisper: `auto`, `float16`, `int8`, `int8_float16`. Lower precision = faster + less VRAM |
 | `--cache-dir` | - | Custom directory for downloaded models (default: HuggingFace cache) |
 | `--default-language` | - | Fallback language code for requests that omit `language`. Required for models that do not support language auto-detection (for example Cohere Transcribe). |
@@ -269,6 +269,13 @@ pip install "agent-cli[mlx-whisper]"
 
 The server will automatically detect and use the MLX backend when available.
 
+For NeMo/Parakeet models on Apple Silicon, `--device auto` uses PyTorch MPS when
+available. You can also request it explicitly:
+
+```bash
+agent-cli server whisper --backend nemo --model parakeet-unified-en-0.6b --device mps
+```
+
 ### HuggingFace Transformers
 
 For loading models in safetensors format (instead of CTranslate2 `.bin` files):

{agent_cli-0.98.0 → agent_cli-0.98.2}/macos/AgentCLI/Sources/AgentCLI/AgentRuntime.swift

@@ -360,8 +360,10 @@ struct AgentRuntime {
         var contents = "runtimeMode=\(modeName)\npackageSource=\(agentCLIPackageSource)\n"
         if runtimeMode == .bundled {
             let backend = TranscriptionSettings.selectedBackend(userDefaults: userDefaults)
+            let ttlSeconds = TranscriptionSettings.selectedModelTTLSeconds(userDefaults: userDefaults)
             contents += "transcriptionBackend=\(backend.rawValue)\n"
             contents += "transcriptionModel=\(TranscriptionSettings.selectedModelName(userDefaults: userDefaults))\n"
+            contents += "transcriptionModelTTLSeconds=\(ttlSeconds)\n"
         }
         return contents
     }

{agent_cli-0.98.0 → agent_cli-0.98.2}/macos/AgentCLI/Sources/AgentCLI/SettingsWindowController.swift

@@ -5,7 +5,7 @@ import SwiftUI
 final class SettingsWindowController {
     static let shared = SettingsWindowController()
 
-    private static let contentSize = NSSize(width: 460, height: 640)
+    private static let contentSize = NSSize(width: 460, height: 680)
 
     private var window: NSWindow?
 

{agent_cli-0.98.0 → agent_cli-0.98.2}/macos/AgentCLI/Sources/AgentCLI/Shortcuts.swift

@@ -285,6 +285,8 @@ struct SettingsView: View {
     private var transcriptionBackend = TranscriptionBackend.whisper.rawValue
     @AppStorage(TranscriptionSettings.transcriptionModelKey)
     private var transcriptionModel = TranscriptionBackend.whisper.defaultModelName
+    @AppStorage(TranscriptionSettings.transcriptionModelTTLSecondsKey)
+    private var transcriptionModelTTLSeconds = TranscriptionSettings.defaultModelTTLSeconds
     @AppStorage(TranscriptionSettings.transcriptionExtraInstructionsKey)
     private var transcriptionExtraInstructions = ""
     @State private var shortcutRevision = 0
@@ -335,13 +337,31 @@ struct SettingsView: View {
                 }
                 .pickerStyle(.menu)
                 .disabled(useUserInstalledAgentCLI)
+
+                Stepper(
+                    value: Binding(
+                        get: { max(0, transcriptionModelTTLSeconds) },
+                        set: { transcriptionModelTTLSeconds = max(0, $0) }
+                    ),
+                    in: 0...86_400,
+                    step: 60
+                ) {
+                    HStack {
+                        Text("Model TTL")
+                        Spacer()
+                        Text(Self.formatTTLSeconds(transcriptionModelTTLSeconds))
+                            .foregroundStyle(.secondary)
+                            .monospacedDigit()
+                    }
+                }
+                .disabled(useUserInstalledAgentCLI)
             } header: {
                 Text("Voice Service")
             } footer: {
                 Text(
                     useUserInstalledAgentCLI
                         ? "Disabled while User-Installed agent-cli is active."
-                        : "Changing these settings restarts the bundled voice service on next transcription."
+                        : "TTL is how long the selected model lives in memory after the last transcription. Set to 0 to keep it loaded until the voice service restarts."
                 )
             }
 
@@ -426,6 +446,26 @@ struct SettingsView: View {
         guard backend.modelOption(named: transcriptionModel) == nil else { return }
         transcriptionModel = backend.defaultModelName
     }
+
+    private static func formatTTLSeconds(_ seconds: Int) -> String {
+        let clampedSeconds = max(0, seconds)
+        if clampedSeconds == 0 {
+            return "Never"
+        }
+        if clampedSeconds < 60 {
+            return "\(clampedSeconds)s"
+        }
+        if clampedSeconds < 3_600 {
+            return "\(clampedSeconds / 60)m"
+        }
+
+        let hours = clampedSeconds / 3_600
+        let minutes = (clampedSeconds % 3_600) / 60
+        if minutes == 0 {
+            return "\(hours)h"
+        }
+        return "\(hours)h \(minutes)m"
+    }
 }
 
 struct ShortcutRecorderRow: View {

{agent_cli-0.98.0 → agent_cli-0.98.2}/macos/AgentCLI/Sources/AgentCLI/TranscriptionSettings.swift

@@ -61,6 +61,8 @@ enum TranscriptionSettings {
     static let transcriptionExtraInstructionsKey = "transcriptionExtraInstructions"
     static let transcriptionBackendKey = "transcriptionBackend"
     static let transcriptionModelKey = "transcriptionModel"
+    static let transcriptionModelTTLSecondsKey = "transcriptionModelTTLSeconds"
+    static let defaultModelTTLSeconds = 300
 
     static var extraInstructions: String {
         UserDefaults.standard.string(forKey: transcriptionExtraInstructionsKey) ?? ""
@@ -77,6 +79,13 @@ enum TranscriptionSettings {
         return backend.modelOption(named: storedModelName)?.id ?? backend.defaultModelName
     }
 
+    static func selectedModelTTLSeconds(userDefaults: UserDefaults = .standard) -> Int {
+        guard let storedValue = userDefaults.object(forKey: transcriptionModelTTLSecondsKey) as? NSNumber else {
+            return defaultModelTTLSeconds
+        }
+        return max(0, storedValue.intValue)
+    }
+
     static func whisperDaemonInstallArguments(userDefaults: UserDefaults = .standard) -> [String] {
         let backend = selectedBackend(userDefaults: userDefaults)
         return [
@@ -89,6 +98,8 @@ enum TranscriptionSettings {
             backend.cliBackend,
             "--model",
             selectedModelName(userDefaults: userDefaults),
+            "--ttl",
+            "\(selectedModelTTLSeconds(userDefaults: userDefaults))",
         ]
     }
 }

{agent_cli-0.98.0 → agent_cli-0.98.2}/macos/AgentCLI/Tests/AgentCLITests/AgentCommandTests.swift

@@ -64,12 +64,32 @@ final class AgentCommandTests: XCTestCase {
                     "daemon", "install", "whisper", "-y", "--",
                     "--backend", "nemo",
                     "--model", "parakeet-tdt-0.6b-v3",
+                    "--ttl", "86400",
                 ]
             ),
             [
                 "daemon", "install", "whisper", "-y", "--",
                 "--backend", "nemo",
                 "--model", "parakeet-tdt-0.6b-v3",
+                "--ttl", "86400",
+            ]
+        )
+    }
+
+    func testTranscriptionDaemonArgumentsIncludeConfiguredTTL() {
+        let defaults = UserDefaults(suiteName: "AgentCLITests.transcription-ttl")!
+        defaults.removePersistentDomain(forName: "AgentCLITests.transcription-ttl")
+        defaults.set("nemo", forKey: TranscriptionSettings.transcriptionBackendKey)
+        defaults.set("parakeet-unified-en-0.6b", forKey: TranscriptionSettings.transcriptionModelKey)
+        defaults.set(86400, forKey: TranscriptionSettings.transcriptionModelTTLSecondsKey)
+
+        XCTAssertEqual(
+            TranscriptionSettings.whisperDaemonInstallArguments(userDefaults: defaults),
+            [
+                "daemon", "install", "whisper", "-y", "--",
+                "--backend", "nemo",
+                "--model", "parakeet-unified-en-0.6b",
+                "--ttl", "86400",
             ]
         )
     }
@@ -220,7 +240,12 @@ final class AgentCommandTests: XCTestCase {
             userDefaults: defaults,
             processRunner: { _, arguments, _ in
                 processArguments.append(arguments)
-                if arguments == ["daemon", "install", "whisper", "-y", "--", "--backend", "auto", "--model", "large-v3"] {
+                if arguments == [
+                    "daemon", "install", "whisper", "-y", "--",
+                    "--backend", "auto",
+                    "--model", "large-v3",
+                    "--ttl", "300",
+                ] {
                     installedWhisper = true
                     return CommandResult(exitCode: 0, output: "Installed and started")
                 }
@@ -241,7 +266,12 @@ final class AgentCommandTests: XCTestCase {
         XCTAssertEqual(
             processArguments,
             [
-                ["daemon", "install", "whisper", "-y", "--", "--backend", "auto", "--model", "large-v3"],
+                [
+                    "daemon", "install", "whisper", "-y", "--",
+                    "--backend", "auto",
+                    "--model", "large-v3",
+                    "--ttl", "300",
+                ],
             ]
         )
         XCTAssertEqual(phases, [.installingVoiceService, .waitingForVoiceService])
@@ -277,7 +307,12 @@ final class AgentCommandTests: XCTestCase {
             userDefaults: defaults,
             processRunner: { _, arguments, _ in
                 processArguments.append(arguments)
-                if arguments == ["daemon", "install", "whisper", "-y", "--", "--backend", "auto", "--model", "large-v3"] {
+                if arguments == [
+                    "daemon", "install", "whisper", "-y", "--",
+                    "--backend", "auto",
+                    "--model", "large-v3",
+                    "--ttl", "300",
+                ] {
                     installedWhisper = true
                     return CommandResult(exitCode: 0, output: "Installed and started")
                 }
@@ -297,7 +332,12 @@ final class AgentCommandTests: XCTestCase {
         XCTAssertEqual(
             processArguments,
             [
-                ["daemon", "install", "whisper", "-y", "--", "--backend", "auto", "--model", "large-v3"],
+                [
+                    "daemon", "install", "whisper", "-y", "--",
+                    "--backend", "auto",
+                    "--model", "large-v3",
+                    "--ttl", "300",
+                ],
             ]
         )
     }
@@ -339,6 +379,7 @@ final class AgentCommandTests: XCTestCase {
                     "daemon", "install", "whisper", "-y", "--",
                     "--backend", "nemo",
                     "--model", "parakeet-tdt-0.6b-v3",
+                    "--ttl", "300",
                 ] {
                     installedWhisper = true
                     return CommandResult(exitCode: 0, output: "Installed and started")
@@ -363,6 +404,7 @@ final class AgentCommandTests: XCTestCase {
                     "daemon", "install", "whisper", "-y", "--",
                     "--backend", "nemo",
                     "--model", "parakeet-tdt-0.6b-v3",
+                    "--ttl", "300",
                 ],
             ]
         )
@@ -413,6 +455,7 @@ final class AgentCommandTests: XCTestCase {
                     "daemon", "install", "whisper", "-y", "--",
                     "--backend", "nemo",
                     "--model", "parakeet-tdt-0.6b-v3",
+                    "--ttl", "300",
                 ] {
                     installedWhisper = true
                     return CommandResult(exitCode: 0, output: "Installed and started")
@@ -438,6 +481,7 @@ final class AgentCommandTests: XCTestCase {
                     "daemon", "install", "whisper", "-y", "--",
                     "--backend", "nemo",
                     "--model", "parakeet-tdt-0.6b-v3",
+                    "--ttl", "300",
                 ],
             ]
         )

{agent_cli-0.98.0 → agent_cli-0.98.2}/macos/appcast.xml

@@ -4,6 +4,38 @@
         <title>Agent CLI Updates</title>
         <link>https://github.com/basnijholt/agent-cli</link>
         <description>Signed Agent CLI macOS app updates.</description>
+        <item>
+            <title>0.98.1</title>
+            <pubDate>Thu, 04 Jun 2026 16:50:48 +0000</pubDate>
+            <link>https://github.com/basnijholt/agent-cli</link>
+            <sparkle:version>189</sparkle:version>
+            <sparkle:shortVersionString>0.98.1</sparkle:shortVersionString>
+            <sparkle:minimumSystemVersion>13.0</sparkle:minimumSystemVersion>
+            <sparkle:hardwareRequirements>arm64</sparkle:hardwareRequirements>
+            <description sparkle:format="markdown"><![CDATA[## What’s Changed
+
+* Add macOS transcription TTL setting (#578) @basnijholt
+
+]]></description>
+            <enclosure url="https://github.com/basnijholt/agent-cli/releases/download/v0.98.1/AgentCLI.zip" length="22896191" type="application/octet-stream" sparkle:edSignature="4g9PD/is1i4Htq5eM2TfAUjxEaXSb7xJJDjnliSfX5TQfqeLLsItUcaCLbhpXaOKqzq0YByVg5SLSpqHK08vAw=="/>
+        </item>
+        <item>
+            <title>0.98.0</title>
+            <pubDate>Thu, 04 Jun 2026 16:25:34 +0000</pubDate>
+            <link>https://github.com/basnijholt/agent-cli</link>
+            <sparkle:version>188</sparkle:version>
+            <sparkle:shortVersionString>0.98.0</sparkle:shortVersionString>
+            <sparkle:minimumSystemVersion>13.0</sparkle:minimumSystemVersion>
+            <sparkle:hardwareRequirements>arm64</sparkle:hardwareRequirements>
+            <description sparkle:format="markdown"><![CDATA[## What’s Changed
+
+* Document NeMo Parakeet setup (#573) @basnijholt
+* Add macOS transcription backend picker (#574) @basnijholt
+* Reference upstream NeMo validation_ds fix (#577) @basnijholt
+
+]]></description>
+            <enclosure url="https://github.com/basnijholt/agent-cli/releases/download/v0.98.0/AgentCLI.zip" length="22891739" type="application/octet-stream" sparkle:edSignature="tC7V5Z+r9fyFe5GvISxKB2ExZJoyx8G7Qt/Zqnb1ELJcU6lm+0KSN3tx7d/BBm1Dbl8eKRiCorOATkTKE2HSCw=="/>
+        </item>
         <item>
             <title>0.97.2</title>
             <pubDate>Thu, 04 Jun 2026 07:34:52 +0000</pubDate>
@@ -137,39 +169,5 @@
 ]]></description>
             <enclosure url="https://github.com/basnijholt/agent-cli/releases/download/v0.96.6/AgentCLI.zip" length="22742441" type="application/octet-stream" sparkle:edSignature="DriNk09uSrq9l20Pu7S+y34r4LvO+rfTHGOm1gsPDOd94FQZXqgDZxbVYYfTGgJ6eWSb0l3ljQokW6VQhJ97DQ=="/>
         </item>
-        <item>
-            <title>0.96.5</title>
-            <pubDate>Fri, 29 May 2026 16:40:33 +0000</pubDate>
-            <link>https://github.com/basnijholt/agent-cli</link>
-            <sparkle:version>179</sparkle:version>
-            <sparkle:shortVersionString>0.96.5</sparkle:shortVersionString>
-            <sparkle:minimumSystemVersion>13.0</sparkle:minimumSystemVersion>
-            <sparkle:hardwareRequirements>arm64</sparkle:hardwareRequirements>
-            <description sparkle:format="markdown"><![CDATA[## What’s Changed
-
-* Fix bare Fn shortcut detection on macOS (#554) @marcellaholm
-* Fix macOS setup menu bar icon (#555) @marcellaholm
-* Use purr for recording release sound (#556) @basnijholt
-* Boost quiet waveform levels (#553) @basnijholt
-
-]]></description>
-            <enclosure url="https://github.com/basnijholt/agent-cli/releases/download/v0.96.5/AgentCLI.zip" length="22738995" type="application/octet-stream" sparkle:edSignature="Jf0aoWxUGvOR6idZhELnlGf61IHrf8Kw92tcH5Xk4wC8chn3mO3OgeE+8+P2orbHC0VhszHgjMNzsGN3bZPwCg=="/>
-        </item>
-        <item>
-            <title>0.96.4</title>
-            <pubDate>Fri, 29 May 2026 14:59:58 +0000</pubDate>
-            <link>https://github.com/basnijholt/agent-cli</link>
-            <sparkle:version>178</sparkle:version>
-            <sparkle:shortVersionString>0.96.4</sparkle:shortVersionString>
-            <sparkle:minimumSystemVersion>13.0</sparkle:minimumSystemVersion>
-            <sparkle:hardwareRequirements>arm64</sparkle:hardwareRequirements>
-            <description sparkle:format="markdown"><![CDATA[## What’s Changed
-
-* Cache user CLI runtime checks (#552) @basnijholt
-* Clean up macOS menu activity abstractions (#551) @basnijholt
-
-]]></description>
-            <enclosure url="https://github.com/basnijholt/agent-cli/releases/download/v0.96.4/AgentCLI.zip" length="22738926" type="application/octet-stream" sparkle:edSignature="Ro8WhN/AVGWxCvG/C6//ISayy5lo5KiOJhvpVQdJKLBsPdZv96HmtZx7G2qeVG87sAmcyIR/84pW4zwzCR7lAA=="/>
-        </item>
     </channel>
 </rss>

{agent_cli-0.98.0 → agent_cli-0.98.2}/tests/test_daemon.py

@@ -12,6 +12,8 @@ import pytest
 from typer.testing import CliRunner
 
 from agent_cli.cli import app
+from agent_cli.core import deps
+from agent_cli.install import launchd as launchd_module
 from agent_cli.install.launchd import _generate_plist as launchd_generate_plist
 from agent_cli.install.launchd import _get_log_command as launchd_get_log_command
 from agent_cli.install.launchd import _get_service_status as launchd_get_service_status
@@ -158,6 +160,34 @@ class TestServiceConfig:
         assert "agent-cli[server,nemo-whisper,wyoming]" in cmd
         assert cmd[cmd.index("--python") + 1] == "3.13"
 
+    def test_build_service_command_materializes_nemo_override_without_spaces(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """UV --overrides receives a copied no-space path for bundled app installs."""
+        uv_path = tmp_path / "uv"
+        uv_path.touch()
+        source_dir = tmp_path / "Application Support" / "AgentCLI" / "_overrides"
+        source_dir.mkdir(parents=True)
+        source_override = source_dir / "nemo-whisper.txt"
+        source_override.write_text("kaldialign==0.9.3\n")
+        materialized_dir = tmp_path / "agentcli-overrides"
+        monkeypatch.setattr(deps, "_OVERRIDES_DIR", source_dir)
+        monkeypatch.setattr(deps, "_find_runtime_uv", lambda: str(uv_path))
+        monkeypatch.setenv("AGENTCLI_UV_OVERRIDES_DIR", str(materialized_dir))
+
+        cmd = build_service_command(
+            SERVICES["whisper"],
+            uv_path,
+            use_macos_extra=True,
+            extra_command_args=["--backend", "nemo"],
+        )
+
+        override_path = Path(cmd[cmd.index("--overrides") + 1])
+        assert override_path == materialized_dir / "nemo-whisper.txt"
+        assert " " not in override_path.as_posix()
+        assert override_path.read_text() == source_override.read_text()
+        assert "agent-cli[server,nemo-whisper,wyoming]" in cmd
+
     def test_build_service_command_uses_app_package_source(
         self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
     ) -> None:
@@ -763,6 +793,21 @@ class TestLaunchdModule:
 
         assert plist["ProgramArguments"][-4:] == ["--model", "small", "--port", "10311"]
 
+    def test_launchd_recent_logs_include_stdout_and_stderr(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """Status output should include stderr when stdout has stale startup logs."""
+        (tmp_path / "stdout.log").write_text("old large-v3 startup\n")
+        (tmp_path / "stderr.log").write_text("error: File not found: `Library/Application`\n")
+        monkeypatch.setattr(launchd_module, "_get_log_dir", lambda _service_name: tmp_path)
+
+        lines = launchd_module._get_recent_logs("whisper", 5)
+
+        assert "==> stdout.log <==" in lines
+        assert "old large-v3 startup" in lines
+        assert "==> stderr.log <==" in lines
+        assert "error: File not found: `Library/Application`" in lines
+
     @patch("subprocess.run")
     def test_launchd_get_service_status_not_installed(
         self,

{agent_cli-0.98.0 → agent_cli-0.98.2}/tests/test_nemo_backend.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import io
+import os
 import sys
 import wave
 from concurrent.futures import ThreadPoolExecutor
@@ -47,16 +48,27 @@ def _install_mock_nemo(
     monkeypatch.setitem(sys.modules, "nemo.collections.asr", asr_module)
 
 
-def _install_mock_torch(monkeypatch: pytest.MonkeyPatch, *, cuda_available: bool) -> None:
-    """Install a minimal mock torch module exposing cuda.is_available()."""
+def _install_mock_torch(
+    monkeypatch: pytest.MonkeyPatch,
+    *,
+    cuda_available: bool,
+    mps_available: bool = False,
+) -> None:
+    """Install a minimal mock torch module exposing device availability."""
 
     class _Cuda:
         @staticmethod
         def is_available() -> bool:
             return cuda_available
 
+    class _Mps:
+        @staticmethod
+        def is_available() -> bool:
+            return mps_available
+
     torch_module: Any = ModuleType("torch")
     torch_module.cuda = _Cuda()
+    torch_module.backends = SimpleNamespace(mps=_Mps())
     monkeypatch.setitem(sys.modules, "torch", torch_module)
 
 
@@ -207,15 +219,27 @@ def test_resolve_device_auto_uses_cuda_when_available(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
     """Ensure auto device selection prefers CUDA when available."""
-    _install_mock_torch(monkeypatch, cuda_available=True)
+    _install_mock_torch(monkeypatch, cuda_available=True, mps_available=True)
     assert backend._resolve_device("auto") == "cuda"
 
 
+def test_resolve_device_auto_uses_mps_when_cuda_unavailable(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """Ensure auto device selection uses Apple GPU when CUDA is unavailable."""
+    monkeypatch.delenv("PYTORCH_ENABLE_MPS_FALLBACK", raising=False)
+    monkeypatch.setattr(backend.sys, "platform", "darwin")
+    _install_mock_torch(monkeypatch, cuda_available=False, mps_available=True)
+
+    assert backend._resolve_device("auto") == "mps"
+    assert os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] == "1"
+
+
 def test_resolve_device_auto_falls_back_to_cpu(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
     """Ensure auto device selection falls back to CPU."""
-    _install_mock_torch(monkeypatch, cuda_available=False)
+    _install_mock_torch(monkeypatch, cuda_available=False, mps_available=False)
     assert backend._resolve_device("auto") == "cpu"
 
 
@@ -229,6 +253,16 @@ def test_resolve_device_explicit_cuda_requires_available_cuda(
         backend._resolve_device("cuda:0")
 
 
+def test_resolve_device_explicit_mps_requires_available_mps(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """Ensure explicit MPS requests fail fast when MPS is unavailable."""
+    _install_mock_torch(monkeypatch, cuda_available=False, mps_available=False)
+
+    with pytest.raises(RuntimeError, match="MPS device requested"):
+        backend._resolve_device("mps")
+
+
 def test_resolve_device_preserves_explicit_cpu(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None: