AI Engine
The AI Engine page manages the core inference server that powers all local AI features — both the LLM (language) and VLM (vision) models run through this engine.
Open via ⚙️ Settings → System → AI Engine.
Runtime Card
The top section shows the current state of the llama-server binary:
Platform Detection
Aegis automatically detects your hardware capabilities:
| Badge | Meaning |
|---|---|
| CUDA | NVIDIA GPU detected — GPU-accelerated inference available |
| METAL | Apple Metal detected — optimized for Apple Silicon |
| Apple Silicon | Running on M1/M2/M3/M4 chip |
| CPU | No GPU acceleration — inference runs on CPU only |
Feature tags (e.g. CUDA, AVX2, METAL) appear as badges showing what the installed binary supports.
Binary Status
| State | Description |
|---|---|
| BUILT-IN | Using the binary bundled with the app (green badge) |
| Binary Not Found | No binary installed — click Install to set up |
| Preferred binary | Shows the recommended binary for your platform and acceleration type |
| Build number | Current build version (e.g. b8235) |
Actions
| Button | What It Does |
|---|---|
| Install | Downloads and installs the appropriate llama-server binary for your platform. Appears only when no binary is found. |
| Update | Re-downloads the binary to get the latest version. Shows "UP TO DATE" if already current. |
| Check | Checks for newer binary versions without installing. Shows update availability. |
| Remove | Deletes the installed binary. Requires a confirmation click (auto-cancels after 3 seconds). |
Console
Below the runtime card, a VS Code-style terminal shows live server output with two tabs:
Vision Tab (VLM)
Shows real-time output from the VLM (llama-server) process:
- Streams model loading progress, inference logs, and health checks
- Connection indicator: green WiFi icon (connected), gray (disconnected), yellow pulsing (connecting)
Language Tab (LLM)
Shows output from the LLM server process:
- Streams chat inference logs, token generation, and system messages
Console Features
| Feature | Description |
|---|---|
| 1000-line buffer | Each tab stores up to 1000 log lines. Older entries are automatically dropped. |
| Auto-scroll | Follows new output automatically. Scroll up to pause; shows "Paused" in the status bar. |
| Clear | Trash icon clears all logs in the active tab |
| Download | Download icon exports the current tab's logs as a .log text file |
| Status bar | Shows which server is active, connection state, line count (N/1000), and scroll mode |
Troubleshooting
| Issue | Solution |
|---|---|
| "Binary Not Found" | Click Install. Aegis downloads the correct binary for your OS and GPU. |
| Install fails | Check your internet connection. If behind a firewall, the binary download URL may be blocked. |
| Console shows "Disconnected" | The VLM server may not be running. Load a model from the VLM settings. |
| CUDA badge missing | Ensure NVIDIA drivers are installed. On Windows, update via GeForce Experience or the NVIDIA website. |
| Metal badge missing on Mac | Verify you're running macOS 12+ on Apple Silicon or a supported Intel Mac with a discrete GPU. |