Download
Aegis AI is a desktop application you download and run on your own computer. No Docker, no terminal commands, no server setup — download, install, open.
Get Aegis AI
Download Aegis AI from the official SharpAI website:
Available for macOS (Apple Silicon + Intel), Windows (x64), and Linux (.deb and AppImage).
System Requirements
| Minimum | Recommended | Ideal | |
|---|---|---|---|
| RAM | 8 GB | 16 GB+ | 32 GB+ |
| Storage | 10 GB free | 50 GB+ free | 200 GB+ (external/NAS) |
| GPU | Any (CPU works) | Apple Silicon or NVIDIA RTX | M2 Pro+ or RTX 4070+ |
| CPU | 4 cores | 8+ cores | 12+ cores |
| OS | macOS 13+, Windows 10+, Ubuntu 22.04+ | Latest version | Latest version |
| Internet | Required for download and cloud APIs | Optional after setup with local models | Optional |
Understanding the Hardware Impact
The most important factor in Aegis AI performance is whether you have a compatible GPU:
| Hardware | Experience |
|---|---|
| Apple Silicon (M1–M4) | Best experience on Mac — Metal GPU acceleration makes local models very responsive |
| NVIDIA RTX GPU | Best experience on Windows/Linux — CUDA acceleration for fast local inference |
| Modern CPU (no GPU) | Works, but noticeably slower — consider using a cloud VLM for better responsiveness |
| Intel Mac (CPU only) | Functional but limited — cloud VLM recommended for practical use |
With a GPU, you can run multiple cameras with continuous AI analysis. On CPU, you'll want to lower the analysis frame rate or use a cloud VLM provider for responsive performance.
Storage Breakdown
Aegis AI uses disk space for several purposes:
| Category | Notes |
|---|---|
| Application | The installed desktop application |
| AI Engine | Inference engine, auto-downloaded on first launch |
| VLM models | Varies by model — smaller models under 1 GB, larger models several GB each |
| LLM models | Varies by model size and quantization level |
| Recorded clips | Depends on camera count and retention settings — configurable in Storage settings |
| Skill models | Each skill may download its own ML model |
Total disk usage depends entirely on which models you download and how long you retain clips. Start small — you can always download more models or extend retention later.
Platform Details
macOS (Apple Silicon)
Best experience. Metal GPU acceleration runs vision models at near-realtime speeds. The AI Engine automatically selects the Metal-optimized binary for your chip.
What you get:
- Metal-accelerated inference for VLM and LLM models
- CoreML support for skills like Depth Estimation — runs on the Neural Engine, leaving GPU free for other tasks
- Automatic camera and network device discovery
- RTSP stream proxying for IP cameras
Intel Macs are supported but will run models slower — AI inference falls back to CPU. If you have an Intel Mac, consider using a cloud VLM provider (OpenAI, Anthropic, or Google) for faster analysis.
Windows (x64)
Full NVIDIA CUDA support for GPU-accelerated inference. The AI Engine detects your GPU and selects the appropriate CUDA-enabled binary automatically.
What you get:
- CUDA-accelerated inference when an NVIDIA GPU is detected
- CPU-optimized fallback for systems without NVIDIA GPUs
- Full camera discovery and RTSP proxying support
- Automatic NVIDIA driver detection
Without an NVIDIA GPU, Aegis uses CPU inference. It works but is significantly slower — a GPU provides a dramatic speedup for AI model inference.
NVIDIA driver note: Install the latest NVIDIA drivers for best performance. The application includes its own CUDA runtime, so a separate CUDA toolkit installation is not needed.
Linux
Linux builds are available as .deb and AppImage packages. Linux support is experimental — macOS and Windows are the primary supported platforms.
What Happens After Installation
- First launch — Aegis opens and presents the guided walkthrough with a friendly greeting
- AI Engine setup — the app downloads the correct inference engine for your platform. This happens automatically in the background.
- Add a camera — connect a webcam, IP camera, or cloud camera (Blink/Ring)
- Download a model — pick a vision model from the Staff Picks list (recommended: SmolVLM2 256M for a quick start, or LFM2.5-VL 1.6B for the best quality-to-size ratio)
- Start monitoring — the AI begins analyzing your camera feeds immediately
Full walkthrough: Getting Started guide
While Aegis works on CPU, a GPU dramatically speeds up AI inference. Apple Silicon Macs and NVIDIA RTX cards deliver the best experience — local vision models run significantly faster with GPU acceleration, making real-time multi-camera analysis practical.
Choosing the Right Setup for Your Hardware
Budget Setup (8 GB RAM, no GPU)
- Use a small vision model like SmolVLM2-256M
- Set analysis frame rate to 0.5 fps (1 frame every 2 seconds)
- Monitor 1–2 cameras
- Consider a cloud VLM (OpenAI Vision) for faster responses
Mid-Range Setup (16 GB RAM, Apple M1/M2 or RTX 3060)
- Use a mid-size vision model like LFM2.5-VL 1.6B
- Set analysis frame rate to 1–3 fps
- Monitor 4–8 cameras simultaneously
- Run 1–2 skills alongside monitoring (e.g., YOLO detection)
High-End Setup (32+ GB RAM, M2 Pro+ or RTX 4070+)
- Use the Gemma 3 4B or LLaVA 7B vision model
- Set analysis frame rate to 5 fps for detailed coverage
- Monitor 8–12+ cameras
- Run multiple skills simultaneously
- Use a local LLM (7B–13B) for fully offline, high-quality reasoning
Uninstalling
macOS
Move Aegis AI.app to the Trash. To also remove stored data, delete ~/.aegis-ai/.
Windows
Use Add or Remove Programs in Windows Settings. To also remove stored data, delete the Aegis AI folder in your AppData directory.
Linux
For .deb: sudo dpkg -r aegis-ai. For AppImage: delete the AppImage file. To also remove stored data, delete ~/.aegis-ai/.
Uninstalling removes the application only. Downloaded models and recorded clips are stored separately in the data directory and persist until you delete them manually.