shortcast/project.yml at main · mutonby/shortcast · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
# XcodeGen project definition for Shortcast.
# Regenerate the .xcodeproj with:  xcodegen generate
name: Shortcast
options:
  bundleIdPrefix: app.shortcast
  deploymentTarget:
    macOS: "15.0"
  createIntermediateGroups: true
  groupSortPosition: top

packages:
  # Vendored: native Gemma 4 multimodal inference (MIT). Source lives in Vendor/.
  # Pulls in mlx-swift, mlx-swift-lm, swift-transformers transitively.
  Gemma4Swift:
    path: Vendor/gemma-4-swift-mlx
  # Qwen 3.5 9B (the "Director" that finds viral moments from a transcript).
  # Same MLX stack Gemma4Swift already resolves — declaring the products here
  # just makes them linkable from the app target. Do NOT pin mlx-swift directly:
  # let it stay transitive so it respects Gemma's 0.31.3 floor (single source).
  mlx-swift-lm:
    url: https://github.com/ml-explore/mlx-swift-lm
    branch: main
  swift-huggingface:
    url: https://github.com/huggingface/swift-huggingface
    from: "0.9.0"
  # WhisperKit caps swift-transformers at <1.2.0; Gemma only needs >=1.1.6, so
  # 1.1.x satisfies the whole graph. (mlx-swift-lm doesn't depend on it at all.)
  swift-transformers:
    url: https://github.com/huggingface/swift-transformers
    from: "1.1.6"
  # On-device transcription, used only as a fallback when a long video has no
  # .srt/.vtt alongside it. Lazy: the Whisper model never downloads otherwise.
  WhisperKit:
    url: https://github.com/argmaxinc/WhisperKit
    from: "0.9.0"

targets:
  Shortcast:
    type: application
    platform: macOS
    sources:
      - path: Shortcast
    dependencies:
      - package: Gemma4Swift
        product: Gemma4Swift
      - package: mlx-swift-lm
        product: MLXVLM         # VLMModelFactory — Qwen 3.5 9B is a VLM package
      - package: mlx-swift-lm
        product: MLXLMCommon     # ChatSession, GenerateParameters, ModelContainer
      - package: mlx-swift-lm
        product: MLXHuggingFace  # #hubDownloader(), #huggingFaceTokenizerLoader()
      - package: swift-huggingface
        product: HuggingFace
      - package: swift-transformers
        product: Tokenizers
      - package: WhisperKit
        product: WhisperKit
    settings:
      base:
        PRODUCT_BUNDLE_IDENTIFIER: app.shortcast.Shortcast
        PRODUCT_NAME: Shortcast
        MARKETING_VERSION: "0.1.0"
        CURRENT_PROJECT_VERSION: "1"
        GENERATE_INFOPLIST_FILE: NO
        INFOPLIST_FILE: Shortcast/Info.plist
        CODE_SIGN_IDENTITY: "-"
        CODE_SIGN_STYLE: Manual
        ENABLE_HARDENED_RUNTIME: NO
        SWIFT_VERSION: "6.0"
        SWIFT_STRICT_CONCURRENCY: complete
        DEAD_CODE_STRIPPING: YES
        ASSETCATALOG_COMPILER_APPICON_NAME: AppIcon
        ASSETCATALOG_COMPILER_GLOBAL_ACCENT_COLOR_NAME: AccentColor
        COMBINE_HIDPI_IMAGES: YES
        ENABLE_PREVIEWS: YES
        MACOSX_DEPLOYMENT_TARGET: "15.0"

  # Developer-only command-line probe. Runs the real generation path
  # (MediaExtractor → Gemma4Engine → JSONVariantParser) headlessly.
  shortcast-probe:
    type: tool
    platform: macOS
    sources:
      - path: Probe
      - path: Shortcast/Models/PostVariant.swift
      - path: Shortcast/Models/VideoJob.swift
      - path: Shortcast/Models/String+Trimmed.swift
      - path: Shortcast/Services/MediaExtractor.swift
      - path: Shortcast/Services/PromptBuilder.swift
      - path: Shortcast/Services/JSONVariantParser.swift
    dependencies:
      - package: Gemma4Swift
        product: Gemma4Swift
    settings:
      base:
        PRODUCT_NAME: shortcast-probe
        SWIFT_VERSION: "6.0"
        MACOSX_DEPLOYMENT_TARGET: "15.0"
        GENERATE_INFOPLIST_FILE: YES