A voice-driven personal task-tracking assistant that helps you manage context switches throughout your workday. Speak naturally to create tasks, switch between them, take notes, and get reminders — all hands-free.

Built with .NET 8, the GitHub Copilot SDK, and Vosk for offline speech recognition.

Focus Assistant runs continuously in the background, listening for the wake word "Hey Focus". When activated, it captures your speech, sends it to a Copilot-powered AI agent, and speaks the response back to you.

• Voice-based task tracking — Create, switch, pause, complete, rename, delete, and merge tasks by speaking naturally. Time spent on each task is tracked automatically.
• Proactive check-ins — If you've been idle for a while, the assistant asks what you're working on. Paused tasks trigger reminders at configurable intervals.
• Task notes — Dictate notes attached to the current task. When you resume a paused task, the assistant reads back your last notes so you remember where you left off.
• End-of-day reflection — Summarizes your day (tasks completed, time spent, what's still open) and lets you set priorities for tomorrow.
• Morning briefing — On a new session, the assistant greets you with yesterday's carry-over tasks and the priorities you set.
• First-use onboarding — Guides you through setting preferences on first launch (reminder intervals, idle threshold, reflection time).

Microphone → Wake Word ("Hey Focus") → Speech-to-Text (Vosk) → raw text
  → Copilot AI Agent (interprets intent, calls tools)
    → Application Use Cases → Domain Entities → JSON File Persistence
  → AI formats response → Text-to-Speech (espeak-ng / SAPI) → Speaker

The Copilot SDK acts as the "brain" — it interprets free-form voice commands and routes them to the correct tool (e.g., create_task, switch_task, add_note). Tools are thin wrappers that delegate to application-layer use cases.

The setup scripts install all dependencies and download the Vosk speech model:

Linux:

cd scripts
./setup.sh

Windows (PowerShell):

cd scripts
.\setup.ps1

The assistant requires an authenticated GitHub account with a Copilot subscription:

gh auth login

Follow the prompts to authenticate. The assistant verifies authentication on startup and will show a clear error if it's missing.

From the scripts/ directory:

Linux:

./run.sh

Windows:

.\run.ps1

The run scripts auto-detect your audio hardware. If a microphone is found, the assistant starts in voice mode (wake word listening). Otherwise, it falls back to text mode (type commands in the terminal).

dotnet run --project src/FocusAssistant.Cli -- [options]

Options:
  --text       Run in text mode (keyboard input instead of voice)
  --verbose    Enable debug-level logging
  --quiet      Suppress all logs except errors
  --test       Run a diagnostic test (checks auth, SDK, model connectivity)

You can also set FOCUS_ASSISTANT_LOG_LEVEL=Debug (or Information, Warning, Error) as an environment variable.

Build a self-contained single-file executable:

cd scripts

./publish.sh              # Build for current OS
./publish.sh linux        # Build for Linux x64
./publish.sh windows      # Build for Windows x64
./publish.sh both         # Build for both platforms

Executables are output to the publish/ directory.

Once running, say "Hey Focus" followed by a command:

If running with --text, type your commands directly (no wake word needed):

> I'm starting work on the API refactor
Created and started 'API refactor'.

> What are my current tasks?
You have 1 task in progress:
  • API refactor (in-progress, 12 min today)

All data is stored locally as JSON files under ~/.focus-assistant/data/ by default:

~/.focus-assistant/
├── data/
│   ├── tasks.json
│   ├── sessions.json
│   ├── daily-plans.json
│   └── preferences.json
└── models/
    └── vosk-model-small-en-us-0.15/

Override the data directory with the FOCUS_ASSISTANT_DATA_DIR environment variable.

Clean Architecture with Domain-Driven Design (DDD):

src/
├── FocusAssistant.Domain/          # Entities, value objects, aggregates, repository interfaces
├── FocusAssistant.Application/     # Use cases, services, application interfaces
├── FocusAssistant.Infrastructure/  # File persistence, voice I/O (Vosk, espeak-ng)
└── FocusAssistant.Cli/             # Composition root, Copilot agent, hosted services

All dependencies point inward. Domain has zero external references. Infrastructure and CLI implement abstractions defined in Domain and Application.

• CopilotAgentSession — Manages the Copilot SDK client/session lifecycle. Serializes concurrent SendCommandAsync calls via semaphore.
• VoiceListenerService — Background service: wake word → STT → Copilot → TTS loop.
• ReminderBackgroundService — Background service: checks idle time and paused-task durations every 30 seconds.
• ToolDefinitions — Maps AI tool calls to application use cases (create_task, switch_task, etc.).
• WakeWordDetector — Vosk-based continuous audio monitoring for the wake phrase.

dotnet test

The test suite includes 338 tests across three projects:

This is a personal project — no license specified.