FAQ
Privacy & Data
Is my camera footage sent to the cloud?
No. All video processing happens locally on your hardware. Your footage never leaves your machine unless you explicitly send clips through a messaging channel (Telegram, Discord, Slack) when an alert fires — and even then, only a snapshot image is sent, not full video.
Can SharpAI see my cameras?
No. Aegis runs entirely on your computer. There is no server-side component that accesses your feeds. There is no SharpAI account, no login, no telemetry that transmits footage. You are the sole controller of your data.
Does Aegis work offline?
Yes. With local models (built-in LLM + downloaded VLM), Aegis runs fully offline after initial setup. Cloud API keys are optional. Skills that use locally downloaded models also work offline. The only features that require internet are:
- Downloading new models from HuggingFace
- Using cloud VLM/LLM providers (OpenAI, Anthropic, Google)
- Sending alerts through messaging channels (Telegram, Discord, Slack)
- Checking for software updates
What data does crash reporting send?
When enabled, anonymous crash reports include only stack traces and error metadata — never footage, AI descriptions, camera data, agent conversations, or memory contents. You can disable crash reporting entirely in ⚙️ Settings → System → Privacy.
Where is my data stored?
All data is stored locally on your machine:
| Data | Default Location |
|---|---|
| Recorded clips | ~/.aegis-ai/media/ (configurable) |
| AI models | Application data folder |
| Agent memory | User data directory |
| Configuration | User data directory |
| Chat history | User data directory |
You can change the clip storage path in ⚙️ Settings → System → Storage to any directory, including external drives or NAS mounts.
Cameras
What cameras work with Aegis?
Any camera that outputs an RTSP stream, supports ONVIF, or is a Blink/Ring camera. Plus webcams and mobile devices. Specifically:
| Camera Type | Examples |
|---|---|
| Blink | Blink Mini, Blink Outdoor, Blink Indoor, Blink XT2 |
| Ring | Ring Doorbell, Ring Stick Up Cam, Ring Floodlight Cam, Ring Indoor Cam |
| ONVIF | DaHua, Hikvision, Lorex, Reolink, Amcrest, Axis, TP-Link |
| RTSP | Any camera with an RTSP URL (virtually all IP cameras) |
| Webcam | Built-in laptop cameras, USB webcams |
| Mobile | iPhone or Android with an RTSP streaming app |
Can I use cameras from different brands together?
Yes. You can mix and match — for example, two Blink cameras, one HikVision RTSP camera, and a USB webcam all running simultaneously. All cameras feed into the same AI pipeline.
How many cameras can I connect?
There's no hard limit. Performance depends on your hardware:
| Setup | Recommended Cameras |
|---|---|
| CPU only, ≤8 GB RAM | 1–4 cameras |
| GPU (NVIDIA RTX), 16+ GB RAM | 4–12 cameras |
| Apple Silicon (M1–M4), 16+ GB | 4–12 cameras |
| Cloud VLM (OpenAI/Anthropic) | Limited by API rate, not hardware |
My ONVIF camera wasn't discovered — what should I do?
- Verify your camera supports ONVIF (check manufacturer specs)
- Ensure your computer and camera are on the same subnet
- Some cameras require ONVIF to be explicitly enabled in their web interface settings
- Try the Network Scanner as an alternative to ONVIF discovery
- If all else fails, add the camera manually using its RTSP URL
AI Models
Can I use my own API keys?
Yes. Bring your own OpenAI, Google, or Anthropic API key. You pay the API provider directly — Aegis doesn't add any markup or intermediate fees.
Do I need a GPU?
No, but it helps enormously. Aegis works on CPU, but AI inference is dramatically faster with a GPU. Apple Silicon Macs (Metal acceleration) and NVIDIA RTX cards (CUDA acceleration) provide the best experience for local model inference, making real-time multi-camera analysis practical. On CPU-only systems, consider using a smaller model or a cloud VLM provider.
Which VLM model should I start with?
- SmolVLM2-256M for a quick start — it's the smallest model, fast, and gives you a feel for how Aegis sees
- LFM2.5-VL 1.6B for the best quality-to-size ratio — our top pick for most users
- Gemma 3 4B for detailed descriptions if you have 16+ GB RAM
- Cloud VLM (OpenAI Vision, Google Gemini) if you want maximum quality without local compute
Can I run both a local LLM and a local VLM at the same time?
Yes, but they share your hardware resources. Your ability to run both depends on your RAM and the sizes of models you choose. Larger models need more memory — if you run into issues, consider using a cloud provider for one of them.
What's the difference between a VLM and an LLM?
A VLM (Vision Language Model) can see images and describe them. An LLM (Language Model) can only process text. Aegis uses the VLM to watch cameras and the LLM to reason about descriptions, answer your questions, and evaluate event handlers.
Skills
What are skills?
Pluggable AI capabilities that extend Aegis — like apps for your security system. Each skill adds a specific ability: object detection, depth estimation, segmentation, etc. Learn more →
Can I create my own skills?
Yes — skills are built on the DeepCamera framework. You can write and publish your own. Fork an existing skill, modify it, host it in a Git repo, and add the repo URL in the Skills panel.
Do skills slow down the main app?
No. Skills run as independent processes with their own memory space. Starting a skill doesn't affect the core application's performance. However, skills do use system resources (CPU, GPU, RAM), so running many simultaneously on limited hardware may reduce overall performance.
Can I run multiple skills at once?
Yes. You can run object detection and depth estimation simultaneously, for example. Each skill has its own console tab showing live output.
Alerts
Why am I getting too many alerts?
Your event handler condition is probably too broad. Narrow it:
- Too broad: "person" → fires for any person, anywhere
- Better: "person at the front door after sunset" → fires only for relevant events
Check the trigger count on your event handlers — if a handler fires 100+ times per day, the condition needs tightening.
Can I set alerts only for certain times?
Yes. Include time-based language in your event handler conditions: "person on the driveway after 10pm" or "activity in the backyard before sunrise."
Do I need messaging channels for alerts to work?
No. Alerts still appear in the Aegis AI interface and are logged in the timeline. Messaging channels (Telegram, Discord, Slack) send notifications to your phone — they're optional but recommended for real-time awareness when you're away from the computer.
General
Where are my recordings stored?
In ~/.aegis-ai/media/ on your local machine by default. You can change this in ⚙️ Settings → System → Storage. You can point this to an external drive or NAS mount for more space.
How do I update Aegis?
Aegis checks for updates automatically by default. You can also check manually in the Settings Modal → Updates tab. Update modes: Automatic (downloads and installs on launch), Manual (notifies you, you click Download), or Disabled. Learn more →
Does Aegis work on Raspberry Pi?
The ARM64 Linux build can run on Raspberry Pi 4/5, but performance will be limited. Use the smallest VLM model (SmolVLM2-256M) and consider a cloud VLM provider for better speed. Skills that require substantial compute (SAM2, model training) may not be practical on Pi hardware.
Can I access Aegis remotely?
Aegis is a desktop application, so the interface runs on the machine where it's installed. However, you can interact with the agent remotely through your configured messaging channels — ask questions, get status reports, and receive alerts on Telegram, Discord, or Slack from anywhere.
How much does Aegis cost?
Aegis AI is free to download and use. There are no subscriptions, cloud fees, or hidden costs. The only potential costs are:
- Cloud API keys (OpenAI, Anthropic, Google) — optional, you pay the provider directly
- Hardware — Aegis runs on your existing computer