Aurix: Project Scope Document

Core Purpose

Aurix is a local-first desktop application that transforms spoken ideas into structured documentation. It helps users offload their thoughts frictionlessly, capturing valuable insights before they are lost. As a secondary function, it analyzes the complexity of the captured content to provide a unique Cognitive Load Index (θ), a measure of cognitive load.

Core Features

1. Voice Capture & Transcription

• Recording Modes: Support for Push-to-Talk (PTT), Click-to-Toggle, and Voice Activity Detection (VAD).
• Local Transcription: Integration with a local whisper.cpp engine for fast, private, and offline speech-to-text.
• Audio Management: Secure local storage of audio recordings (WAV format) linked to their transcriptions.
• Real-time Feedback: Display a live, streaming transcript in the UI as the user speaks.

2. AI-Powered Document Generation (LangGraph Workflow)

• Content Analysis: An AI node to classify the document type (e.g., technical spec, meeting notes), extract topics, and identify key entities.
• Structured Document Generation: An AI node that takes the raw transcript and generates a well-structured Markdown document with headings and logical sections.
• Mermaid Diagram Generation: An AI node that detects when a user is describing a process or system and automatically generates the corresponding Mermaid.js diagram syntax.
• Document Assembly: A final node to combine all generated text and diagrams into a single, coherent document.

3. Cognitive Load Index (θ)

• Complexity Analysis: The AI workflow will calculate a complexityScore for the transcribed content based on jargon, concept density, and structure.
• Index Calculation: The Cognitive Load Index will be derived from the complexityScore, the length of the transcript, and the number of diagrams generated.
• User Feedback Loop: A UI slider will allow users to provide subjective feedback ("How mentally taxing was that session?"), which will be used to personalize the weighting of the index calculation over time.

4. Core UI Components

• Main Recording Interface: A clean, simple UI with clear controls to start/stop recording and provide real-time visual feedback (e.g., a waveform).
• Live Transcript View: A panel that displays the streaming transcript as it's generated.
• Document Preview Panel: A side-by-side view showing the final Markdown document as it's being assembled by the AI workflow.
• Dashboard: A view to browse, search, and manage previously generated documents and view historical Cognitive Load Index trends.

5. Local-First Infrastructure

• Local Storage: All data (audio, transcripts, documents, settings) will be stored in a local SQLite database.
• Local AI: All core transcription and LLM processing will happen locally via whisper.cpp and Ollama. Internet connectivity is not required for core functionality.
• Privacy by Default: No user data leaves the machine unless the user explicitly opts into cloud features.

Stretch Goals (Optional Features)

• Cloud Sync & Backup: Optional, end-to-end encrypted synchronization of documents between multiple devices using PouchDB/CouchDB.
• Cloud Model Support: Allow users to provide their own API keys for cloud-based LLMs (GPT-4, Claude) for potentially higher-quality document generation.
• Advanced Voice Commands: Implement commands like "Create a sequence diagram for..." or "Summarize the last five minutes."
• Multiple Export Formats: Add support for exporting documents to PDF, HTML, or DOCX.
• Team Collaboration: Allow sharing and collaborative editing of documents for authenticated users.

Out of Scope (for MVP)

• Real-time collaborative editing.
• Mobile or web versions.
• Integration with third-party project management tools.
• Any feature that requires a cloud connection to function.

Technical Stack

Layer	Tool
UI	React (via Vite) + Tailwind + shadcn/ui
App Shell	Electron Forge
Language	TypeScript
AI Workflow	LangGraph
Data Sync	Motion API (OAuth2)
Storage	Local SQLite or JSON DB
Auth	OAuth2 + local token vault
Optional	N8N, OpenAI, Ollama

MVP Definition

Required:

• Motion OAuth login
• Pull tasks and projects
• Simple θ index calculator
• Desktop dashboard (gauge, breakdown, graph)
• One notification when threshold is exceeded
• User feedback slider
• LangGraph workflow that computes θ

Optional but Ideal:

• Task detail breakdown panel
• Cache and refresh logic
• Daily "morning summary" notification

Next Steps

1. Finalize this scope
2. Break into phases (e.g., Phase 0: Auth, Phase 1: Motion Sync)
3. Write best-practices docs per subsystem
4. Build MVP core in Electron and LangGraph
5. Add AI enhancements after baseline stability