"I am simply one hell of a digital butler."
Sebastian is a local, multi-agent desktop concierge designed to manage the "household chores" of a developer's digital life. While you focus on high-value creative work, Sebastian handles background maintenance, file organization, debugging, and session context management.
Submitted for the Google AI Agents Capstone (Freestyle / Concierge Agents Track).
Developers lose hours weekly to digital entropy:
- Context Switching: Breaking flow to copy-paste errors into a browser AI.
- File Clutter: Downloads and temp folders becoming unmanageable dumps.
- Micro-Friction: Repetitive tasks like formatting JSON or writing boilerplate emails.
Sebastian functions as an autonomous agentic OS overlay that lives on your machine:
- 👀 Vision-Based: Monitors your screen pixel-by-pixel to detect when you are stuck.
- 🛹 Auto-Sorting: Automatically organizes files into semantic categories (Finance, Dev, Personal) using
sorting_rules.json. - 🛠️ Proactive Help: Analyzes code, detects bugs, and provides fixes via a non-intrusive HUD.
Sebastian reduces cognitive load and gives you a maintenance-free digital environment, saving 5–10 hours of "digital housekeeping" per week.
Sebastian uses a Local Loop-based Agentic Architecture powered by the multimodal capabilities of Google Gemini 2.0 Flash.
- Uses Gemini's multimodal vision capabilities to "see" screen state directly.
- Dynamic system prompts inject task-specific personas.
- Vision:
pyautoguicaptures screen state;ImageChopsdetects motion/idleness. - File System:
watchdogmonitors OS events (Created, Moved, Modified) in real-time.
- Short-Term: Thread-safe real-time state for HUD updates.
- Long-Term: JSON-based persistence (
sebas_long_term_memory.json).
- HUD: Custom tkinter overlay offers contextual assistance.
- File Ops: Autonomous file moving, renaming, and content generation.
- ✅ Multimodality (Screen Images + Text)
- ✅ Custom Tools for code analysis & file manipulation
- ✅ Long-Term Memory with MemoryManager
- ✅ Observability via
sebas.log
Sebastian/
│
├── Sebas.py # Main entry point, UI logic, and Agent definitions
│
├── sebas_config.json # Configuration for cooldowns, thresholds, and UI position
│
├── sorting_rules.json # Taxonomy for the file organizer (User Customizable)
│
├── sebas_long_term_memory.json # Persistent brain for agent memories
│
└── sebas.log # Telemetry and event logs
Sebastian is designed as a Local Desktop Agent for privacy and OS-level control.
git clone https://github.com/No-Reed/Sebas-AI-Concierge.git
cd Sebas-AI-Concierge
pip install -r requirements.txt
Create a .env file:
GEMINI_API_KEY=...
python Sebas.py
- Auto-Mode: Simply work; Sebastian analyzes when you're idle or stuck.
- Manual: Use the HUD buttons to "Analyze" or "Chat".
https://drive.google.com/drive/folders/1dgYN27Sk4zpFhy6gUf4uxAOzUsfFXnx3?usp=sharing
| Hotkey | Action |
|---|---|
| Ctrl+Shift+A | Auto-Analyze: Triggers the Multi-Agent workflow on the current screen/file. |
| Ctrl+Shift+S | Apply Fixes: Automatically applies generated code fixes. |
| Ctrl+Shift+M | Metrics: Exports usage data to metrics.json. |
| Ctrl+Shift+R | Restore: Reopens the GUI if minimized. |
- Local-First: No files are uploaded except specific snippets for Gemini API calls.
- Encrypted Keys: API tokens are stored in
.env. - No Cloud Storage: All memory and logs are stored locally.
- GUI Dashboard for configuring sorting rules.
- Voice Interaction mode.
- Multi-Device Sync for memory sharing.