Hardware and Performance

Aegis detects your hardware automatically and shows detailed diagnostics. This page explains everything you see in the Hardware settings panel.

Open the Hardware View

⚙️ Settings → System → Hardware

System Hardware Overview

The top section shows three cards with live-updating data:

CPU Card

→ Model name — e.g., "Apple M2 Pro" or "AMD Ryzen 9 7950X" → Architecture — ARM64, x86_64, etc. → Core count — physical cores / logical threads (e.g., "8C/16T") → Live usage — real-time CPU percentage with a color-coded progress bar (green < 60%, yellow < 80%, red ≥ 80%)

GPU Card

→ Model name — e.g., "NVIDIA RTX 4090" or "Apple M2 Pro GPU" → Vendor — NVIDIA, Apple, AMD, etc. → VRAM — total GPU memory in GB → Acceleration badges — colored tags showing supported APIs:

• METAL (purple) — Apple Silicon GPU acceleration
• CUDA (green) — NVIDIA GPU acceleration

If no dedicated GPU is detected, the card shows "No dedicated GPU detected."

Memory Card

→ Total RAM — e.g., "32.0 GB RAM" → Available — free memory right now → Unified (purple badge) — shown on Apple Silicon where GPU and CPU share memory → Live usage — real-time memory percentage with color-coded bar

Operating System Info

Below the hardware cards, two panels show:

OS Details

Field	Example
System	macOS 14.2 / Windows 11 23H2 / Ubuntu 24.04
Build	Build number or "N/A"
Kernel	Kernel version string
Hostname	Your computer's network name

Framework Compatibility

Shows which ML frameworks are available on your system:

Framework	What It Does
llama.cpp	Powers local LLM inference (the brain)
ONNX Runtime	Alternative inference engine for some models
PyTorch	Used by many skills and VLM models
TensorFlow	Legacy framework support

Each framework shows a green checkmark (available) or gray circle (not available). Below that, you'll see optimization tags like AVX2, NEON, or Metal — these indicate hardware-specific speedups your system supports.

ML Intelligence Layer

The bottom section details your system's machine learning capabilities:

Precision Support

→ FP16 — half-precision floating point (faster inference, slightly less accurate) → INT8 — 8-bit integer quantization (much faster, good for compact models) → BF16 — bfloat16 (used by newer models, good balance of speed and accuracy)

Supported precisions appear in bright text; unsupported ones are dimmed.

Acceleration Frameworks

Tags showing which GPU acceleration APIs are available — Metal, CUDA, ROCm, etc.

Key Metrics

→ Max Context — estimated maximum token context window your hardware can handle → Ideal Model Classes — recommended model sizes based on your available memory (e.g., "7B, 13B")

AI Engine

⚙️ Settings → System → AI Engine

The AI Engine is the local inference server that powers your vision and language models. Controls include:

→ Start / Stop — manually control the engine → Status — shows whether the engine is running, loading, or stopped → Loaded models — which LLM and VLM models are currently in memory → Performance metrics — inference speed, tokens per second

Performance Tips

Maximizing Speed

Action	Impact
Use Apple Silicon or NVIDIA RTX GPU	Dramatically faster AI analysis vs CPU
Match VLM model size to your VRAM	Prevents memory swapping, which kills performance
Close GPU-heavy apps while Aegis runs	More VRAM available for inference
Use INT8 or FP16 quantized models	Faster inference with minimal quality loss

For Lower-Spec Machines

→ Use SmolVLM2-256M — smallest available model, runs well even on CPU → Use cloud VLM (OpenAI Vision, Google) instead of downloading a local model → Reduce simultaneous camera feeds in Monitor view → Set shorter clip retention in Storage settings → Close the Skills console when not debugging (it buffers log output)

Recommendations

At the bottom of the Hardware page, Aegis may show personalized recommendations based on your system — for example, suggesting a specific model size that fits your VRAM, or recommending driver updates for better CUDA performance.