Visual DAG Automated Workflow Engine

DuRT’s most revolutionary productivity feature is its modular Directed Acyclic Graph (DAG) visual workflow engine.

Instead of treating audio recording, speech recognition, LLM text processing, translation, and speech synthesis as fragmented standalone tools, DuRT 3.0 allows you to wire nodes together on an interactive 2D canvas via bezier connections, creating fully automated end-to-end media production pipelines.

---

Core Paradigm: From Linear Chains to DAG Topologies

Traditional tools restrict you to rigid Step A ➔ Step B ➔ Step C linear sequences. DuRT 3.0’s DAG architecture brings two generational breakthroughs:

1. Parallel Branch Fan-out

A single upstream step can branch into multiple downstream steps simultaneously. For example, once transcription generates a subtitle track, the engine can execute three parallel branches at once:

• Branch 1 (Publishing): Subtitle refinement ➔ Export formatted SRT;
• Branch 2 (Localization): Multilingual translation ➔ TTS character voiceover ➔ Export localized audio track;
• Branch 3 (Knowledge Base): AI meeting digest & action items extraction ➔ Synchronize to Notion and Obsidian.

2. Strongly-Typed Data Contracts (DuWorkflowDataType)

To ensure total runtime safety and predictability, DuRT enforces a strict data type contract between node ports across 6 distinct types:

Data Type Identifier	Name	Icon	Typical Producer Step	Typical Consumer Step
audioStream	Audio Sample Stream	🎙️	Audio / Screen capture (monitorCapture)	Real-time transcription (realtimeASR)
mediaFile	Media File Entity	📁	Media / URL input (mediaInput)	Media transcription (mediaASR)
subtitleStream	Live Subtitle Stream	💬	Real-time transcription (realtimeASR)	Live interpretation (realtimeTranslation)
subtitleTrack	Timed Subtitle Track	⏱️	Media transcription (mediaASR), Subtitle input	Refinement, Translation, Artifact export
document	Structured Document	📄	Content creation (contentCreation)	Artifact export, Notion / Obsidian sync
deliverable	Delivery Package	📦	Artifact export (exportArtifact)	Local storage, S3 / WebDAV upload

NOTE

The built-in rule engine automatically checks port compatibility and prevents cycles: only ports with compatible data types can be wired together, eliminating runtime crashes.

---

The 11 Standard Workflow Steps

DuRT 3.0 provides 11 standardized atomic steps spanning capture to multi-cloud delivery:

┌────────────────────────┐      ┌────────────────────────┐      ┌────────────────────────┐
│   1. Inputs            │ ───► │  2. ASR & Interpreting │ ───► │ 3. AI Polish & Delivery│
│ Audio Capture / Media  │      │ Live Dictation / Files │      │ Translate / TTS / S3   │
└────────────────────────┘      └────────────────────────┘      └────────────────────────┘

1. Inputs:
   • monitorCapture: macOS system audio & microphone capture
   • mediaInput: Local audio/video file import and web stream URL extraction
   • subtitleInput: Existing SRT/VTT subtitle and text import
2. ASR & Live Interpretation:
   • realtimeASR: Millisecond-level live streaming dictation
   • mediaASR: File transcription, word timestamps, and unsupervised speaker diarization
   • realtimeTranslation: Sentence-by-sentence streaming bilingual live subtitles
3. AI Post-Processing, Synthesis & Delivery:
   • subtitleRefinement: Formatting, filler word removal, and segment optimization
   • fullTranslation: High-fidelity multi-language contextual translation
   • contentCreation: Meeting minutes, action items, and social media post drafting
   • textToSpeech: Multi-role natural AI voiceover and dubbing
   • exportArtifact: Local multi-format saving and 6 automated cloud delivery channels

For detailed parameters and port specs, see the 11 Standard Workflow Steps Guide.

---

Workflow Hub (Template Center)

DuRT comes with pre-built workflow templates for high-frequency daily routines:

• 🎙️ Live Meeting Scribe: System audio capture ➔ Live dictation ➔ AI meeting summary & action items ➔ Auto-sync to Notion database;
• 🌏 Video Localization Pipeline: Import MP4 video ➔ Media transcription ➔ Full translation ➔ Multi-voice TTS dubbing ➔ Export bilingual SRT and dubbed audio;
• 🧹 Podcast Cleaner: Audio transcription (word timestamps) ➔ LLM filler-word scrubbing ➔ Export cleaned SRT and Markdown notes;
• ☁️ Multi-Cloud Archival: Media transcription ➔ Auto-upload to S3-compatible storage ➔ Trigger downstream webhook automation.

---

Workflow Studio Experience

• Persistent 2D World Coordinates: Dragged node coordinates and bezier paths are saved natively. Canvas layouts never shift when you reopen the app;
• Single-Step Test Runs: Run and fine-tune any individual node directly on the canvas without executing the whole pipeline;
• Pin to Dashboard: Pin your most frequent workflows to the top row of the Dashboard for one-click drag-and-drop execution.