OutroKit 1.8.0
See the version list below for details.
dotnet tool install --global OutroKit --version 1.8.0
dotnet new tool-manifest
dotnet tool install --local OutroKit --version 1.8.0
#tool dotnet:?package=OutroKit&version=1.8.0
nuke :add-package OutroKit --version 1.8.0
๐ผ OutroKit
OutroKit writes the titles, descriptions, chapters, and SRT subtitles for your episode from its transcript, video, or audio, using AI powered by the GitHub Copilot SDK. It works for podcasts and videos alike.
OutroKit was called Podcast Metadata Generator, and its NuGet package is now
OutroKit. The oldPodcastMetadataGeneratorpackage still works and tells you how to switch: rundnx OutroKit, or if you installed it as a tool, rundotnet tool uninstall -g PodcastMetadataGeneratorand thendotnet tool install -g OutroKit. Your settings carry over.
New here? The step-by-step install and usage guide at outrokit.com walks through everything, starting from opening a terminal.
<img width="720" alt="The OutroKit main menu in a terminal, with Burn Captions into Video selected" src="https://raw.githubusercontent.com/jamesmontemagno/outrokit/main/docs/images/outrokit-menu.png" />
<img width="720" alt="Ten title suggestions generated by OutroKit, with a prompt to select one" src="https://raw.githubusercontent.com/jamesmontemagno/outrokit/main/docs/images/outrokit-titles.png" />
โจ Features
- ๐ฏ Title Generation - Get multiple creative title suggestions for your episode
- ๐จ Title Styles - Pick a style (balanced, descriptive, curiosity hook, question, how-to, playful, professional, SEO, or mixed) each time you generate titles
- ๐ Description Generation - Create short, medium, and long descriptions optimized for different platforms, picking which lengths to generate each time
- ๐ Chapter Generation - Auto-generate YouTube-compatible chapter markers with timestamps
- ๐ฌ SRT Conversion - Convert transcripts to valid SRT subtitle format
- ๐งน Auto Fix SRT - Check an
.srtfile for readability issues, apply safe fixes, and save a corrected copy without changing the original - ๐ Multiple Transcript Formats - Support for Zencastr, time-range, SRT formats, and plain text
- ๐ง Video & Audio Transcription - Start from a video or audio file (
.mp3,.wav) and generate an SRT transcript locally with ffmpeg and Whisper.net - ๐ฅ Burn Captions into Video - Draw an
.srt,.vtt,.ass, or.ssacaptions file into the picture of a video with four appearance choices. For other caption formats, or captions that need more work first, use CaptionStack - ๐ File Browser - Built-in file browser or drag-and-drop support
- โก Streaming Responses - Watch AI responses generate in real-time
- ๐ Copy to Clipboard - Copy titles, descriptions, chapters, or everything at once; finished output is printed unwrapped so it also selects cleanly in the terminal
- ๐พ Choose what to save - Save selected titles, descriptions, chapters, SRT subtitles, and the manifest, or everything at once; the suggested output folder is next to the transcript unless you configured another folder in Settings
- ๐ค Model Selection - Choose from multiple AI models (GPT-5, Claude, Gemini)
- โ๏ธ Configurable Settings - Customize generation parameters and save preferences
- โจ What's New - See full details for the latest release and compact, title-only updates for older releases from the main menu
- Agent-callable commands - Generate metadata, transcribe, convert or fix captions, burn captions, and extract or replace audio without interactive menus, with JSON results and safe explicit output paths
๐ Prerequisites
- .NET 10.0 SDK or later
- A GitHub Copilot account and authentication, unless using BYOK, for the interactive app and AI metadata commands. The free Copilot plan works
- GitHub Copilot CLI installed for the interactive app and AI metadata commands (see below)
- ffmpeg available on
PATH, or its executable path configured in Settings (only needed to transcribe video or audio files and to burn captions into video)- Burning captions needs an ffmpeg built with libass, which provides the
subtitlesfilter. Homebrew's standardffmpegformula leaves it out;brew install ffmpeg-fullincludes it
- Burning captions needs an ffmpeg built with libass, which provides the
The app reaches Copilot through the Copilot CLI installed on your computer, on every system and whether you run the NuGet tool or build from source. It does not bundle a Copilot runtime. Make sure copilot is on your PATH, or set COPILOT_CLI_PATH to the executable. If the app cannot find it, it shows how to install it. The noninteractive transcription, caption, subtitle conversion, audio, and Whisper model commands run without a Copilot readiness check or authentication.
Installing GitHub Copilot CLI
macOS and Linux (Homebrew):
brew install copilot-cli
Windows (WinGet):
winget install GitHub.Copilot
All platforms (npm, requires Node.js 22+):
npm install -g @github/copilot
macOS and Linux (install script):
curl -fsSL https://gh.io/copilot-install | bash
Then authenticate:
copilot
# Type /login and follow the prompts
๐ Installation
Just run it with .NET 10+ (Recommended)
dnx OutroKit
As a .NET Tool
dotnet tool install -g OutroKit
Then start it with outrokit.
From Source
git clone https://github.com/jamesmontemagno/outrokit.git
cd outrokit
dotnet build PodcastMetadataGenerator.sln
๐ Additional Docs
Agent skills in the NuGet package
OutroKit bundles focused Agent Skills alongside the
tool, following the NuGet skill-delivery convention.
The OutroKit and legacy PodcastMetadataGenerator packages carry the same files
at skills/<name>/SKILL.md, matched to the app version:
| Skill | Helps your agent with |
|---|---|
outrokit-episode-metadata |
Agent-callable titles, descriptions, and publishing-kit JSON |
outrokit-youtube-chapters |
Real topic-boundary timestamps and YouTube chapter publishing checks |
outrokit-captions |
Local transcription, SRT readability fixes, and safe styled caption burns |
outrokit-audio |
Original-audio extraction and processed-track replacement without re-encoding the picture |
These skills tell your agent how to invoke the noninteractive commands below, interpret their JSON responses, and escalate decisions to you. Starting OutroKit without a subcommand still opens the interactive app. Installing the tool does not automatically register skills with your editor or load them into the app's Copilot SDK sessions.
For a published version containing skills, the article's dotnet-inspect tool
can discover them directly from NuGet:
dotnet tool install -g dotnet-inspect
dotnet-inspect package OutroKit -S Skills
To make them available to Copilot in your own repository, download the .nupkg
for the version you use from NuGet and extract its skills/ directory, then copy
the desired skill folders into your repository's .github/skills/. For example,
with unzip on macOS or Linux (replace <version> with your package version):
unzip "OutroKit.<version>.nupkg" 'skills/*' -d ./outrokit-package
mkdir -p .github/skills
cp -R ./outrokit-package/skills/outrokit-* .github/skills/
Review the instructions before enabling them, check for existing folders before
copying, and do not overwrite customized skills without review. Use only packages
from maintainers you trust. Refresh the copied skills when updating OutroKit.
Because this is a tool package, do not add it as a PackageReference just to
obtain skills; project-dependency skill installers are not the installation path
for this package. From a source checkout, the same documents live in skills/.
๐ Usage
Agent CLI
Use outrokit --help and outrokit <command> --help to discover the installed
version's options. Replace outrokit with dnx OutroKit to run without a global
installation. Existing outrokit and outrokit /path/to/transcript.txt usage
still opens the interactive app.
| Command | Result |
|---|---|
metadata <transcript> |
Requested titles, description lengths, and/or checked chapters as JSON |
transcribe <media> --output <file.srt> |
Local Whisper transcription; requires an existing model and ffmpeg |
srt convert <transcript> --output <file.srt> |
SRT from existing timestamps, with conversion warnings |
captions check <captions> |
Readability report for SRT, WebVTT, ASS, or SSA |
captions fix <captions.srt> --output <fixed.srt> |
Separate corrected SRT and remaining findings |
captions burn <video> --captions <file> --output <video.mp4> |
Captioned H.264 video, with optional style and safe fixes |
audio extract <video> --output <audio.wav> |
Original audio; WAV, MP3, M4A, or FLAC output |
audio replace <video> --audio <track> --output <video.mp4> |
Original picture with AAC replacement audio, padded or trimmed to the video length |
models list / models download --whisper-model Base |
Whisper model catalog / explicitly downloaded and initialized model path |
For example, start with a timed transcript and select only the metadata needed:
outrokit metadata "/path/to/episode.srt" \
--artifacts Titles Descriptions Chapters --lengths Short Medium \
--style Seo --context "This episode's guest and subject" \
--output "/path/to/episode-metadata.json"
outrokit captions check "/path/to/episode.srt"
outrokit captions fix "/path/to/episode.srt" --output "/path/to/episode-fixed.srt"
outrokit captions burn "/path/to/episode.mp4" \
--captions "/path/to/episode-fixed.srt" --output "/path/to/episode-captioned.mp4" \
--appearance ContrastPanel --size Large --position Bottom --fix
Metadata defaults to Titles and Descriptions, with all three description
lengths. Use --artifacts to select Titles, Descriptions, and/or Chapters,
and --lengths to select Short, Medium, and/or Long. Titles are suggestions:
the command never silently picks one. To condition descriptions on your choice,
pass --title "Your selected title" in a descriptions invocation.
Metadata accepts transcripts, not media. Chapters and SRT conversion require real timestamps; untimed chapter requests fail before contacting AI. Chapters are checked for valid times, a zero start, at least three markers, 10-second minimum durations including the final chapter, transcript-supported starts, and the configured title word limit. Review whether the boundaries actually match the topics and final edited video.
Download a Whisper model explicitly if one is not already installed:
outrokit models download --whisper-model Base
outrokit transcribe "/path/to/episode.mp4" --output "/path/to/episode.srt" \
--whisper-model-path "/path/returned/by/download/ggml-base.bin"
Downloads do not save app preferences. Transcription never implicitly downloads a missing model. Local commands do not need Copilot authentication; only metadata uses AI and sends transcript content to the configured provider.
Response contract. Each command emits one JSON object on stdout:
{"success":true,"command":"...","result":{...},"error":null}. Failures set
success:false and provide error.code, error.message, and optional
error.details; diagnostics also go to stderr. Help and version are plain text.
Metadata is in result.metadata (titles, descriptions, chapters);
descriptions are keyed by Short, Medium, and Long. Saved metadata paths
are in result.savedFiles; other file-producing commands return result.output.
Metadata's optional .json output contains input, model, and metadata,
not the response envelope and not separate text exports.
| Exit code | Meaning |
|---|---|
0 |
Operation completed; inspect any report or warnings before publishing |
1 |
Operation failed |
2 |
Invalid arguments, input, or settings |
3 |
SDK permission needs a user decision |
4 |
Caption quality findings remain |
130 |
Cancelled or timeout expired |
captions fix can save a corrected file and still return 4 if findings
remain; inspect its result.output and result.report. Burns with findings
are blocked without writing video: use --fix for safe SRT/WebVTT fixes and
only add --allow-issues after reviewing and explicitly accepting remaining
findings. ASS/SSA keep their embedded styling; appearance overrides and --fix
are rejected for those formats.
Safety and settings. Every writer requires a separate destination; metadata
can instead return JSON only on stdout. Existing outputs require explicit
--overwrite, and inputs cannot be overwritten even through an output symlink.
Failed or cancelled writes leave existing files unchanged. --settings reads
an optional settings file; invalid files fail rather than silently falling back.
CLI overrides do not save preferences. Use --ffmpeg, --model (metadata), and
--timeout-seconds as needed; the default timeout is 1800 seconds and the maximum
is 86400. Ctrl+C cancels an active command.
Metadata sessions disable tools for transcript-only generation. If the SDK
still requests permission, exit 3 returns the exact request description and
ID in error.details. The agent must ask you to approve or deny it. Retry with
--approve-permission <id> only after approval, or --deny-permission <id> after
denial if you want to continue with that action rejected. A decision is consumed
once for that matching request, never applied to other actions; stopping without
retrying is also safe.
Console App - Interactive Mode
# If installed as a tool:
outrokit
# Or from source:
cd src/Console
dotnet run
Console App - With a Transcript File
outrokit /path/to/transcript.txt
# Or from source:
dotnet run -- /path/to/transcript.txt
Console App - With a Video or Audio File
Open Settings โ Transcription Settings, choose a Whisper GGML model, then download and initialize it. After setup, select Load Transcript, Video, or Audio โ Video or audio file, or pass the file on the command line:
outrokit /path/to/episode.mp4
outrokit /path/to/episode.mp3
Supported video formats are .mp4, .mov, .mkv, .avi, .webm, .m4v, .wmv, .mpeg, and .mpg. Supported audio formats are .mp3, .wav, .wave, .m4a, .aac, .flac, .ogg, .opus, and .mp2.
The app checks the file for an audio track with ffmpeg, asks where to save the .srt, transcribes the audio locally, and loads that SRT into the existing metadata flow. Model downloads and transcription can take a while; press Ctrl+C to cancel either one and return to the menu.
Models are downloaded from the Whisper.net Hugging Face repository into ~/.podcast-metadata-generator/models. Smaller models are faster and use less memory; larger models generally improve accuracy. English (.en) variants only transcribe English, while the other models are multilingual.
Console App - Auto Fix SRT
Select Auto Fix SRT from the main menu and choose an .srt file. OutroKit checks caption timing and readability, applies safe fixes, and suggests a <filename>-fixed.srt file next to the original. The original is never changed; issues that cannot be safely fixed are reported for manual review.
Console App - Save Results
Choose Save Results from the main menu to select the generated titles, individual descriptions, chapters, SRT subtitles, and manifest to save. Choose Everything to save all available results. The output folder defaults to the folder containing the loaded transcript; a custom Output Directory set under Settings takes precedence, and you can edit the folder for each save. If the generated SRT would overwrite the source transcript, OutroKit saves it as a separate *_subtitles.srt file.
Console App - Burn Captions into a Video
Select Burn Captions into Video from the main menu, then choose a video and a captions file (.srt, .vtt, .ass, or .ssa). No transcript needs to be loaded, and an SRT made by the app from the same video works as the captions file.
Before burning, the app checks caption timing and readability and reports invalid times, overlaps, empty cues, short durations, long lines, too many lines, and fast reading speeds. For SRT and WebVTT files, it can safely clean whitespace, remove empty cues, rewrap text, trim suitable overlaps, and extend captions into available time. Fixes are applied only to a temporary copy used for the burn; the original caption file is never changed. Remaining findings can be reviewed manually or left unchanged. The app then asks you to confirm the caption appearance, text size (Small, Medium, or Large), and position (Bottom or Top), then where to save the result. Choose Classic Outline, Contrast Panel, Bold Impact, or Clean Shadow; Classic Outline is the default. The default output is <video name>-captioned next to the original, and the original video is never overwritten. Captions are sized to the video, so they look the same at 720p and 4K and stay proportionate on portrait video. An .ass or .ssa file keeps its own fonts, sizes, and positions.
Burning captions re-encodes the picture as H.264, so it takes a while on long videos. The result is saved as .mp4, .mov, .mkv, or .m4v; a source in another format is saved as .mp4. Audio is copied unchanged when the format stays the same and converted to AAC otherwise. Press Ctrl+C to cancel and return to the menu; a cancelled or failed burn saves nothing and leaves an existing file as it was.
Captions in another format, such as YouTube's .sbv or .ttml? CaptionStack converts caption files for free in your browser, and is the place to go when captions need more work before you burn them in.
This feature needs an ffmpeg that includes libass. If yours does not, the app says so and offers to switch to another one. With Homebrew:
brew install ffmpeg-full
# It is installed beside the standard ffmpeg, not on PATH. Use this as the ffmpeg path:
echo "$(brew --prefix ffmpeg-full)/bin/ffmpeg"
๐ Blazor Demo (Local Only)
A web UI demo is included for local development and presentations.
โ ๏ธ Local Use Only: The Blazor app uses your local Copilot authentication and is not designed for deployment or multi-user access.
Running the Blazor Demo
# From repository root
cd src/Blazor
dotnet run
Then open https://localhost:5001 in your browser.
Features
- Drag-and-drop file upload
- Real-time streaming AI output
- Tabbed results view
- Settings persistence via localStorage
Title Styles
| Style | What you get |
|---|---|
| Balanced | Engaging, descriptive, and SEO-friendly |
| Descriptive | Clear and straightforward, no hype |
| Curiosity Hook | Teases a hook that makes people want to listen |
| Question | Phrased as a question the episode answers |
| How-To / Educational | Outcome-focused: how-to, guide, lessons learned |
| Playful | Witty, with wordplay or humor |
| Professional | Polished and authoritative |
| SEO Keywords | Leads with the terms people search for |
| Mixed Variety | A different style for each title |
Set the default under Settings โ Generation Settings โ Title Settings (console) or on the Settings page (Blazor). The default is preselected whenever you generate titles, and you can choose a different style for that run.
Copying Results (Console)
When a generation finishes, the console offers a Copy to clipboard menu for the selected title, all titles, each description, chapters, or everything. The same menu is available from the main menu once you have results. On Linux, clipboard access requires xsel.
๐ Website
The guide at outrokit.com lives in site/ as plain HTML, CSS, and JavaScript with no build step. Pushing changes under site/ to main deploys it through the Pages workflow.
To preview it locally:
cd site
python3 -m http.server 4173
Then open http://localhost:4173.
The site is served at outrokit.com through the custom domain in the repository's Pages settings.
The image shown when the link is shared is site/og.png. It is a 1200ร630 screenshot of design/og-image.html, which uses the site's fonts and colors. To regenerate it after an edit, with Chrome or Edge:
"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" --headless --hide-scrollbars \
--force-device-scale-factor=1 --window-size=1200,630 \
--screenshot="$PWD/site/og.png" "file://$PWD/design/og-image.html"
The hype video that announces a release is made in the same style from design/hype-video/. Its README covers how to render it and how to update it for the next release.
The browser can keep running after the file is written; stop it with Ctrl+C.
๐ฏ Supported Transcript Formats
Zencastr Format
00:00.00 Speaker 1: Hello and welcome to the show.
00:15.50 Speaker 2: Thanks for having me!
Time-Range Format
00:00:00 - 00:00:15
Hello and welcome to the show.
00:00:15 - 00:00:30
Thanks for having me!
SRT Format
1
00:00:00,000 --> 00:00:15,000
Hello and welcome to the show.
2
00:00:15,000 --> 00:00:30,000
Thanks for having me!
Plain Text
Any text file without timestamps will be processed as plain text. Note: Chapter generation and SRT conversion require timestamps.
๐ Output Files
When you save results, the following files are generated:
| File | Description |
|---|---|
titles.txt |
List of generated title suggestions |
description-short.txt |
Short description (~50 words) |
description-medium.txt |
Medium description (~150 words) |
description-long.txt |
Long description (~300 words) |
chapters.txt |
YouTube-compatible chapter markers |
subtitles.srt |
SRT subtitle file |
manifest.json |
JSON manifest with all metadata |
โ๏ธ Configuration
Access settings from the main menu to configure:
General Settings
- AI Model - Select from available Copilot models (dynamically fetched from CLI)
- ffmpeg Path - Executable path or command used to check video and audio files, extract 16 kHz mono audio, and burn captions into video
- Whisper GGML Model - Select, download, and initialize a local Whisper.net model for transcription
- Caption Appearance, Size, and Position - Default appearance (default: Classic Outline), text size (default: Medium), and position (default: Bottom) for captions burned into video. You are asked to confirm or change them each time you burn captions
- Output Directory - Default location for saved files
- Show Name - The name of your show (used in prompts for better context)
- Host Names - Host names (used in prompts)
- Episode Context - Add guest names, topics, or other context to improve generation
Generation Settings
- Title Count - Number of title suggestions to generate (default: 5)
- Title Max Words - Maximum words per title (default: 10)
- Title Style - Default style for generated titles (default: Balanced). You are asked to confirm or change it each time you generate titles or all metadata
- Description Lengths - Word counts for short/medium/long descriptions (default: 50/150/300)
- Chapter Range - Min/max chapters to generate (default: 3-12)
- Chapters per 30 min - Target density of chapters (default: 5)
- Chapter Title Words - Max words per chapter title (default: 8)
Settings are automatically saved to ~/.podcast-metadata-generator/settings.json.
๐๏ธ Project Structure
outrokit/
โโโ PodcastMetadataGenerator.sln # Solution file
โโโ src/
โ โโโ Core/ # Shared class library
โ โ โโโ Models/
โ โ โ โโโ AppSettings.cs # Configuration and generation settings
โ โ โ โโโ CaptionStyle.cs # Caption size and position options
โ โ โ โโโ GenerationResult.cs # Results container
โ โ โ โโโ Manifest.cs # JSON manifest structure
โ โ โ โโโ TitleStyle.cs # Title style options and prompt guidance
โ โ โ โโโ Transcript.cs # Transcript model
โ โ โ โโโ TranscriptSegment.cs # Segment model
โ โ โโโ Services/
โ โ โ โโโ CaptionBurnService.cs # Burns captions into video with ffmpeg
โ โ โ โโโ CopilotAuthService.cs # CLI authentication checks
โ โ โ โโโ FfmpegRunner.cs # Shared ffmpeg process launcher
โ โ โ โโโ MetadataGenerator.cs # AI generation via Copilot SDK
โ โ โ โโโ OutputService.cs # File output handling
โ โ โ โโโ SettingsService.cs # Settings persistence
โ โ โ โโโ SrtConverter.cs # SRT format conversion
โ โ โ โโโ TranscriptParser.cs # Multi-format transcript parsing
โ โ โโโ Prompts/
โ โ โโโ PromptTemplates.cs # AI prompt templates
โ โโโ Console/ # Console application
โ โ โโโ UI/
โ โ โ โโโ AppWorkflow.cs # Main application workflow
โ โ โ โโโ ConsoleUI.cs # Spectre.Console UI helpers
โ โ โ โโโ WhatsNew.cs # Entries for the What's New screen
โ โ โโโ Program.cs # Console entry point
โ โโโ Blazor/ # Blazor Server demo (local only)
โ โโโ Components/
โ โ โโโ Layout/ # MainLayout, NavMenu
โ โ โโโ Pages/ # Home, Generate, Settings
โ โโโ wwwroot/
โ โ โโโ css/app.css
โ โโโ Program.cs # Blazor entry point
โโโ tests/ # xUnit tests, run on Linux, Windows, and macOS in CI
โโโ data/ # Sample transcripts
๐ค Contributing
Contributions are welcome! Please see CONTRIBUTING.md for guidelines.
- Fork the repository
- Create your feature branch (
git checkout -b feature/amazing-feature) - Commit your changes (
git commit -m 'Add some amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Open a Pull Request
๐ License
This project is licensed under the MIT License - see the LICENSE file for details.
๐ Acknowledgments
- GitHub Copilot SDK for AI integration
- Spectre.Console for the beautiful terminal UI
- Inspired by jamesmontemagno/app-podcast-assistant
| Product | Versions Compatible and additional computed target framework versions. |
|---|---|
| .NET | net10.0 is compatible. net10.0-android was computed. net10.0-browser was computed. net10.0-ios was computed. net10.0-maccatalyst was computed. net10.0-macos was computed. net10.0-tvos was computed. net10.0-windows was computed. |
This package has no dependencies.