agent-cli 0.102.1__tar.gz → 0.102.2__tar.gz

{agent_cli-0.102.1 → agent_cli-0.102.2}/.github/workflows/release.yml

@@ -126,17 +126,3 @@ jobs:
           git commit -m "Update AgentCLI release metadata to ${TAG_NAME}"
           git pull --rebase origin main
           git push origin HEAD:main
-      - name: Trigger Homebrew tap update
-        env:
-          GH_TOKEN: ${{ secrets.HOMEBREW_TAP_DISPATCH_TOKEN }}
-          TAG_NAME: ${{ github.event.release.tag_name }}
-        run: |
-          if [[ -z "$GH_TOKEN" ]]; then
-            echo "HOMEBREW_TAP_DISPATCH_TOKEN is required to update basnijholt/homebrew-tap." >&2
-            exit 1
-          fi
-
-          gh api repos/basnijholt/homebrew-tap/dispatches \
-            -f event_type=agent-cli-release \
-            -F "client_payload[tag_name]=$TAG_NAME" \
-            -F "client_payload[asset_url]=https://github.com/basnijholt/agent-cli/releases/download/${TAG_NAME}/AgentCLI.dmg"

{agent_cli-0.102.1 → agent_cli-0.102.2}/.pre-commit-config.yaml

@@ -48,14 +48,17 @@ repos:
         name: uv lock
         entry: uv lock
         language: python
-        additional_dependencies: ["uv"]
+        # Pinned: uv export's marker output differs between releases, so an
+        # unpinned uv makes agent_cli/_requirements/*.txt depend on whoever's
+        # hook cache is freshest.
+        additional_dependencies: ["uv==0.11.32"]
         pass_filenames: false
         files: pyproject\.toml
       - id: sync-requirements
         name: sync extra requirements
         entry: python .github/scripts/sync_requirements.py
         language: python
-        additional_dependencies: ["uv"]
+        additional_dependencies: ["uv==0.11.32"]
         pass_filenames: false
         files: (pyproject\.toml|uv\.lock)
       - id: check-extras-sync

{agent_cli-0.102.1 → agent_cli-0.102.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-cli
-Version: 0.102.1
+Version: 0.102.2
 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>
@@ -61,6 +61,7 @@ Requires-Dist: watchfiles>=0.21.0; extra == 'memory'
 Provides-Extra: mlx-whisper
 Requires-Dist: fastapi[standard]; extra == 'mlx-whisper'
 Requires-Dist: mlx-whisper>=0.4.0; (sys_platform == 'darwin' and platform_machine == 'arm64') and extra == 'mlx-whisper'
+Requires-Dist: mlx>=0.24; (sys_platform == 'darwin' and platform_machine == 'arm64') and extra == 'mlx-whisper'
 Provides-Extra: nemo-whisper
 Requires-Dist: fastapi[standard]; extra == 'nemo-whisper'
 Requires-Dist: nemo-toolkit[asr]>=2.2.0; (python_version < '3.14') and extra == 'nemo-whisper'

{agent_cli-0.102.1 → agent_cli-0.102.2}/agent_cli/_requirements/diarization.txt

@@ -1,10 +1,10 @@
 # This file was autogenerated by uv via the following command:
 #    uv export --extra diarization --no-dev --no-emit-project --no-hashes
-aiohappyeyeballs==2.6.2
+aiohappyeyeballs==2.6.2 ; python_full_version < '3.13'
     # via aiohttp
-aiohttp==3.13.5
+aiohttp==3.13.5 ; python_full_version < '3.13'
     # via fsspec
-aiosignal==1.4.0
+aiosignal==1.4.0 ; python_full_version < '3.13'
     # via aiohttp
 alembic==1.18.4
     # via optuna
@@ -18,7 +18,7 @@ anyio==4.13.0
     # via httpx
 asteroid-filterbanks==0.4.0
     # via pyannote-audio
-attrs==26.1.0
+attrs==26.1.0 ; python_full_version < '3.13'
     # via aiohttp
 certifi==2026.5.20
     # via
@@ -61,7 +61,7 @@ filelock==3.29.0
     #   torch
 fonttools==4.63.0
     # via matplotlib
-frozenlist==1.8.0
+frozenlist==1.8.0 ; python_full_version < '3.13'
     # via
     #   aiohttp
     #   aiosignal
@@ -136,7 +136,7 @@ mdurl==0.1.2
     # via markdown-it-py
 mpmath==1.3.0
     # via sympy
-multidict==6.7.1
+multidict==6.7.1 ; python_full_version < '3.13'
     # via
     #   aiohttp
     #   yarl
@@ -267,7 +267,7 @@ pillow==12.2.0
     # via matplotlib
 primepy==1.3
     # via torch-pitch-shift
-propcache==0.5.2
+propcache==0.5.2 ; python_full_version < '3.13'
     # via
     #   aiohttp
     #   yarl
@@ -480,5 +480,5 @@ tzdata==2026.2 ; sys_platform == 'emscripten' or sys_platform == 'win32'
     # via pandas
 urllib3==2.7.0
     # via requests
-yarl==1.24.2
+yarl==1.24.2 ; python_full_version < '3.13'
     # via aiohttp

{agent_cli-0.102.1 → agent_cli-0.102.2}/agent_cli/_requirements/mlx-whisper.txt

@@ -98,7 +98,9 @@ markupsafe==3.0.3
 mdurl==0.1.2
     # via markdown-it-py
 mlx==0.31.2 ; platform_machine == 'arm64' and sys_platform == 'darwin'
-    # via mlx-whisper
+    # via
+    #   agent-cli
+    #   mlx-whisper
 mlx-metal==0.31.2 ; platform_machine == 'arm64' and sys_platform == 'darwin'
     # via mlx
 mlx-whisper==0.4.3 ; platform_machine == 'arm64' and sys_platform == 'darwin'

{agent_cli-0.102.1 → agent_cli-0.102.2}/agent_cli/_requirements/nemo-whisper.txt

@@ -6,11 +6,11 @@ absl-py==2.4.0 ; python_full_version < '3.14'
     #   tensorboard
 accelerate==1.13.0 ; python_full_version < '3.14'
     # via peft
-aiohappyeyeballs==2.6.2 ; python_full_version < '3.14'
+aiohappyeyeballs==2.6.2 ; python_full_version < '3.13'
     # via aiohttp
-aiohttp==3.13.5 ; python_full_version < '3.14'
+aiohttp==3.13.5 ; python_full_version < '3.13'
     # via fsspec
-aiosignal==1.4.0 ; python_full_version < '3.14'
+aiosignal==1.4.0 ; python_full_version < '3.13'
     # via aiohttp
 alembic==1.18.4 ; python_full_version < '3.14'
     # via optuna
@@ -31,7 +31,7 @@ anyio==4.13.0
     #   watchfiles
 asttokens==3.0.1 ; python_full_version < '3.14'
     # via stack-data
-attrs==26.1.0 ; python_full_version < '3.14'
+attrs==26.1.0 ; python_full_version < '3.13'
     # via aiohttp
 audioop-lts==0.2.2 ; python_full_version == '3.13.*'
     # via
@@ -146,7 +146,7 @@ filelock==3.29.0 ; python_full_version < '3.14'
     #   transformers
 fonttools==4.63.0 ; python_full_version < '3.14'
     # via matplotlib
-frozenlist==1.8.0 ; python_full_version < '3.14'
+frozenlist==1.8.0 ; python_full_version < '3.13'
     # via
     #   aiohttp
     #   aiosignal
@@ -308,7 +308,7 @@ mpmath==1.3.0 ; python_full_version < '3.14'
     # via sympy
 msgpack==1.1.2 ; python_full_version < '3.14'
     # via librosa
-multidict==6.7.1 ; python_full_version < '3.14'
+multidict==6.7.1 ; python_full_version < '3.13'
     # via
     #   aiohttp
     #   yarl
@@ -515,7 +515,7 @@ portalocker==3.2.0 ; python_full_version == '3.13.*'
     # via sacrebleu
 prompt-toolkit==3.0.52 ; python_full_version < '3.14'
     # via ipython
-propcache==0.5.2 ; python_full_version < '3.14'
+propcache==0.5.2 ; python_full_version < '3.13'
     # via
     #   aiohttp
     #   yarl
@@ -888,5 +888,5 @@ wrapt==2.2.1 ; python_full_version < '3.14'
     # via nemo-toolkit
 xxhash==3.7.0 ; python_full_version < '3.14'
     # via datasets
-yarl==1.24.2 ; python_full_version < '3.14'
+yarl==1.24.2 ; python_full_version < '3.13'
     # via aiohttp

{agent_cli-0.102.1 → agent_cli-0.102.2}/agent_cli/server/whisper/backends/mlx.py

@@ -135,11 +135,24 @@ def _transcribe_in_subprocess(
     kwargs: dict[str, Any],
 ) -> dict[str, Any]:
     """Run transcription in subprocess. Model stays loaded between calls."""
+    import mlx.core as mx  # noqa: PLC0415
     import mlx_whisper  # noqa: PLC0415
     import numpy as np  # noqa: PLC0415
 
     audio_array = np.frombuffer(audio_bytes, dtype=audio_dtype).reshape(audio_shape)
-    result = mlx_whisper.transcribe(audio_array, path_or_hf_repo=model_name, **kwargs)
+    try:
+        result = mlx_whisper.transcribe(audio_array, path_or_hf_repo=model_name, **kwargs)
+    finally:
+        # Release MLX's reusable Metal buffer cache after each request. With --ttl 0
+        # the subprocess is never recycled, so without this the cache grows to the
+        # largest working set ever seen and gets pushed to swap, degrading latency.
+        # clear_cache() frees only the cache, NOT live arrays, so the loaded model
+        # weights stay resident and inference stays hot (no cold reload).
+        # In finally so a failed transcription also releases the buffers it
+        # allocated; otherwise the long-lived worker would retain them.
+        # Top-level clear_cache() needs mlx>=0.24 (pinned in the mlx-whisper
+        # extra); older releases only expose mx.metal.clear_cache().
+        mx.clear_cache()
 
     return {
         "text": result.get("text", ""),

{agent_cli-0.102.1 → agent_cli-0.102.2}/macos/AgentCLI/README.md

@@ -111,8 +111,11 @@ workflow expects these repository secrets:
 - `APPLE_APP_SPECIFIC_PASSWORD`: app-specific password for `notarytool`
 - `APPLE_TEAM_ID`: Apple Developer Team ID
 - `SPARKLE_PRIVATE_ED_KEY`: private Sparkle EdDSA key for signing updates
-- `HOMEBREW_TAP_DISPATCH_TOKEN`: fine-grained token with Contents write access
-  to `basnijholt/homebrew-tap`, used to trigger the tap cask update workflow
+
+The Homebrew cask is not published from here. `basnijholt/homebrew-tap` runs a
+daily cron that reads the latest release and bumps `Casks/agent-cli.rb` itself;
+run it on demand with
+`gh workflow run update-agent-cli-cask.yml --repo basnijholt/homebrew-tap`.
 
 The workflow also expects the repository variable `SPARKLE_PUBLIC_ED_KEY`.
 Generate the key pair with Sparkle's `generate_keys` tool and store only the

{agent_cli-0.102.1 → agent_cli-0.102.2}/macos/appcast.xml

@@ -4,6 +4,33 @@
         <title>Agent CLI Updates</title>
         <link>https://github.com/basnijholt/agent-cli</link>
         <description>Signed Agent CLI macOS app updates.</description>
+        <item>
+            <title>0.102.1</title>
+            <pubDate>Sat, 25 Jul 2026 18:58:28 +0000</pubDate>
+            <link>https://github.com/basnijholt/agent-cli</link>
+            <sparkle:version>198</sparkle:version>
+            <sparkle:shortVersionString>0.102.1</sparkle:shortVersionString>
+            <sparkle:minimumSystemVersion>13.0</sparkle:minimumSystemVersion>
+            <sparkle:hardwareRequirements>arm64</sparkle:hardwareRequirements>
+            <description sparkle:format="markdown"><![CDATA[## Bug fixes
+
+* **Transcription now starts ~1.3s sooner after pressing the hotkey.** The macOS app spawned a login + interactive shell (`zsh -lic`) to discover `PATH` on *every* command it ran, costing 1.1–1.4s with a typical dotfiles setup. The recording pill appeared immediately but the microphone did not open until that shell finished, so speaking right away lost the first words. The login shell `PATH` is now resolved once per app launch, during the existing startup warm-up. Measured hotkey to first audio chunk: ~1.8s → ~0.4–0.6s. (#622)
+  * Only successful lookups are cached, so a login shell that fails once does not leave the app on a degraded `PATH` for the rest of the session.
+  * *Install or Update CLI* still re-probes, so a relocated `agent-cli` binary is picked up without restarting the app.
+
+## Performance
+
+* `agent-cli` no longer reads its own distribution metadata on startup. `__version__` resolves lazily, saving ~29ms on every command. (#622)
+
+## Maintenance
+
+* ⬆️ Update docker/login-action to v4.5.1 (#621, #619)
+* ⬆️ Update actions/checkout to v7.0.1 (#617)
+* ⬆️ Update release-drafter/release-drafter to v7.6.0 (#614)
+* [pre-commit.ci] pre-commit autoupdate (#616)
+]]></description>
+            <enclosure url="https://github.com/basnijholt/agent-cli/releases/download/v0.102.1/AgentCLI.zip" length="24329452" type="application/octet-stream" sparkle:edSignature="MQizz/9FfhxJGKQc6I/XXtIUBLVPh20yWi/kvfE7xBpoQGPF4yE4C0T1Ex1WpU3im/vRLrMIbZMv0+yHYTifBQ=="/>
+        </item>
         <item>
             <title>0.102.0</title>
             <pubDate>Tue, 14 Jul 2026 05:57:01 +0000</pubDate>
@@ -163,22 +190,5 @@
 ]]></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>
     </channel>
 </rss>

{agent_cli-0.102.1 → agent_cli-0.102.2}/pyproject.toml

@@ -60,6 +60,10 @@ faster-whisper = ["agent-cli[server]", "faster-whisper>=1.0.0", "ctranslate2>=4.
 mlx-whisper = [
     "agent-cli[server]; sys_platform == 'darwin' and platform_machine == 'arm64'",
     "mlx-whisper>=0.4.0; sys_platform == 'darwin' and platform_machine == 'arm64'",
+    # mlx-whisper only requires mlx>=0.11, but the MLX backend calls the
+    # top-level mx.clear_cache(), which does not exist before mlx 0.24
+    # (0.23.2 and older only expose the deprecated mx.metal.clear_cache).
+    "mlx>=0.24; sys_platform == 'darwin' and platform_machine == 'arm64'",
 ]
 whisper-transformers = ["agent-cli[server]", "transformers>=4.30.0", "torch>=2.0.0"]
 nemo-whisper = [

{agent_cli-0.102.1 → agent_cli-0.102.2}/tests/conftest.py

@@ -55,11 +55,31 @@ def mock_console() -> Console:
     return Console(file=io.StringIO(), width=80, force_terminal=True)
 
 
+def force_plain_console(console: Console) -> None:
+    """Make a console audible and strip every escape sequence from its output.
+
+    `no_color` alone still emits bold and dim, so the colour system is switched
+    off outright.
+    """
+    console.quiet = False
+    console.no_color = True
+    console._color_system = None
+
+
 @pytest.fixture(autouse=True)
-def reset_rich_console_quiet() -> None:
-    """Prevent JSON-mode CLI tests from silencing later command output."""
-    utils.console.quiet = False
-    utils.err_console.quiet = False
+def reset_rich_console_state() -> None:
+    """Keep the shared consoles audible and plain for every test.
+
+    `quiet` is reset so JSON-mode CLI tests cannot silence later command output.
+
+    Colour is disabled because Rich honours `FORCE_COLOR` even when stdout is
+    captured, so a developer shell exporting it makes 29 tests fail on ANSI
+    escapes interleaved into otherwise-matching output. The consoles are mutated
+    in place rather than rebound: 17 modules hold a direct
+    `from agent_cli.core.utils import console` reference.
+    """
+    for console in (utils.console, utils.err_console):
+        force_plain_console(console)
 
 
 @pytest.fixture

{agent_cli-0.102.1 → agent_cli-0.102.2}/tests/test_macos_app.py

@@ -346,12 +346,11 @@ def test_release_workflow_publishes_macos_app_asset() -> None:
     assert "NOTARIZE=1" in workflow
     assert "gh release upload" in workflow
     assert "dist/macos/AgentCLI.dmg" in workflow
-    assert "name: Trigger Homebrew tap update" in workflow
-    assert "HOMEBREW_TAP_DISPATCH_TOKEN" in workflow
-    assert "repos/basnijholt/homebrew-tap/dispatches" in workflow
-    assert "event_type=agent-cli-release" in workflow
-    assert "client_payload[tag_name]" in workflow
-    assert "client_payload[asset_url]" in workflow
+    # The tap owns its own updates: basnijholt/homebrew-tap runs a daily cron that
+    # reads the latest release. Dispatching from here only duplicated that with an
+    # expiring cross-repo token, whose 401 turned every release run red.
+    assert "HOMEBREW_TAP_DISPATCH_TOKEN" not in workflow
+    assert "repos/basnijholt/homebrew-tap/dispatches" not in workflow
     assert "python3 .github/scripts/normalize_appcast.py macos/appcast.xml" in workflow
     assert workflow.index(
         "python3 .github/scripts/normalize_appcast.py macos/appcast.xml"

{agent_cli-0.102.1 → agent_cli-0.102.2}/tests/test_mlx_backend.py

@@ -7,10 +7,14 @@ import wave
 from typing import Any
 from unittest.mock import MagicMock, patch
 
+import numpy as np
 import pytest
 
 from agent_cli.server.whisper.backends.base import BackendConfig
-from agent_cli.server.whisper.backends.mlx import MLXWhisperBackend
+from agent_cli.server.whisper.backends.mlx import (
+    MLXWhisperBackend,
+    _transcribe_in_subprocess,
+)
 
 
 def _make_wav_bytes(
@@ -78,3 +82,65 @@ async def test_mlx_transcribe_accepts_matching_wav() -> None:
 
     mock_convert.assert_not_called()
     assert result.text == "hello"
+
+
+def test_transcribe_in_subprocess_clears_metal_cache() -> None:
+    """_transcribe_in_subprocess must release MLX's Metal buffer cache.
+
+    Regression test for the --ttl 0 memory leak: without mx.clear_cache() the
+    long-lived subprocess never releases the Metal buffer cache, so it grows to
+    the largest working set ever seen and gets pushed to swap.
+    """
+    audio = np.zeros(160, dtype=np.float32)
+    fake_result = {"text": "hi", "language": "en", "segments": []}
+
+    fake_mx = MagicMock()
+    fake_mlx_whisper = MagicMock()
+    fake_mlx_whisper.transcribe.return_value = fake_result
+
+    with patch.dict(
+        "sys.modules",
+        {"mlx": MagicMock(core=fake_mx), "mlx.core": fake_mx, "mlx_whisper": fake_mlx_whisper},
+    ):
+        result = _transcribe_in_subprocess(
+            "mlx-community/whisper-large-v3-mlx",
+            audio.tobytes(),
+            audio.shape,
+            str(audio.dtype),
+            {"temperature": 0.0},
+        )
+
+    fake_mlx_whisper.transcribe.assert_called_once()
+    fake_mx.clear_cache.assert_called_once_with()
+    assert result["text"] == "hi"
+
+
+def test_transcribe_in_subprocess_clears_cache_on_failure() -> None:
+    """A failed transcription must still release the Metal buffer cache.
+
+    The worker subprocess is long-lived under --ttl 0, so buffers allocated by a
+    transcription that raises would otherwise leak. clear_cache() runs in a
+    finally block, so it must fire even when mlx_whisper.transcribe raises.
+    """
+    audio = np.zeros(160, dtype=np.float32)
+
+    fake_mx = MagicMock()
+    fake_mlx_whisper = MagicMock()
+    fake_mlx_whisper.transcribe.side_effect = RuntimeError("boom")
+
+    with (
+        patch.dict(
+            "sys.modules",
+            {"mlx": MagicMock(core=fake_mx), "mlx.core": fake_mx, "mlx_whisper": fake_mlx_whisper},
+        ),
+        pytest.raises(RuntimeError, match="boom"),
+    ):
+        _transcribe_in_subprocess(
+            "mlx-community/whisper-large-v3-mlx",
+            audio.tobytes(),
+            audio.shape,
+            str(audio.dtype),
+            {"temperature": 0.0},
+        )
+
+    fake_mx.clear_cache.assert_called_once_with()

{agent_cli-0.102.1 → agent_cli-0.102.2}/tests/test_utils.py

@@ -7,8 +7,10 @@ from datetime import timedelta
 from unittest.mock import Mock, patch
 
 import pytest
+from rich.console import ColorSystem
 
 from agent_cli.core import process, utils
+from tests.conftest import force_plain_console
 
 
 @pytest.mark.parametrize(
@@ -208,3 +210,43 @@ async def test_signal_handling_context_uses_sync_handlers_on_windows(
     assert signal.getsignal(signal.SIGINT) is prev_sigint
     assert signal.getsignal(signal.SIGTERM) is prev_sigterm
     logger.debug.assert_called()
+
+
+def test_shared_consoles_render_without_ansi(capsys: pytest.CaptureFixture[str]) -> None:
+    """Rich honours FORCE_COLOR even into captured output, breaking substring asserts.
+
+    The autouse `reset_rich_console_state` fixture must keep both shared consoles
+    colourless so the suite passes regardless of the developer's shell.
+    """
+    assert utils.console.color_system is None
+    assert utils.err_console.color_system is None
+
+    utils.console.print("[bold red]stdout text[/bold red]")
+    utils.err_console.print("[dim]stderr text[/dim]")
+
+    captured = capsys.readouterr()
+    assert captured.out == "stdout text\n"
+    assert captured.err == "stderr text\n"
+
+
+@pytest.mark.parametrize("stream", ["console", "err_console"])
+def test_force_plain_console_strips_forced_colour(
+    stream: str,
+    capsys: pytest.CaptureFixture[str],
+) -> None:
+    """Reproduce a FORCE_COLOR shell rather than trusting the ambient environment.
+
+    Without this the coverage above is vacuous wherever Rich already detects a
+    colourless stream, which is every CI runner. Mutating the shared console is
+    safe: the autouse fixture restores it before the next test.
+    """
+    console = getattr(utils, stream)
+    console.no_color = False
+    console._color_system = ColorSystem.TRUECOLOR
+    assert console.color_system == "truecolor"
+
+    force_plain_console(console)
+
+    console.print("[bold red]coloured[/bold red]")
+    captured = capsys.readouterr()
+    assert (captured.err if stream == "err_console" else captured.out) == "coloured\n"

{agent_cli-0.102.1 → agent_cli-0.102.2}/uv.lock

@@ -138,6 +138,7 @@ memory = [
 ]
 mlx-whisper = [
     { name = "fastapi", extra = ["standard"] },
+    { name = "mlx", marker = "platform_machine == 'arm64' and sys_platform == 'darwin'" },
     { name = "mlx-whisper", marker = "platform_machine == 'arm64' and sys_platform == 'darwin'" },
 ]
 nemo-whisper = [
@@ -246,6 +247,7 @@ requires-dist = [
     { name = "markdown-code-runner", marker = "extra == 'dev'", specifier = ">=2.7.0" },
     { name = "markdown-gfm-admonition", marker = "extra == 'dev'" },
     { name = "markitdown", extras = ["docx", "pdf", "pptx"], marker = "extra == 'rag'", specifier = ">=0.1.3" },
+    { name = "mlx", marker = "platform_machine == 'arm64' and sys_platform == 'darwin' and extra == 'mlx-whisper'", specifier = ">=0.24" },
     { name = "mlx-whisper", marker = "platform_machine == 'arm64' and sys_platform == 'darwin' and extra == 'mlx-whisper'", specifier = ">=0.4.0" },
     { name = "nemo-toolkit", extras = ["asr"], marker = "python_full_version < '3.14' and extra == 'nemo-whisper'", specifier = ">=2.2.0" },
     { name = "notebook", marker = "extra == 'dev'" },