Documentation

Complete getting started: from zero to your first useful alert

Complete these eight steps in order. After account approval and license issuance, validate one camera before expanding; this usually takes 10–30 minutes. Apple Silicon Macs can use the bundled local AI engine. Windows and Linux currently require a cloud AI provider or a reachable BYO-GPU endpoint.

0. Sign in and obtain product access

First create an account or sign in. Both magic link and email/password are supported. Open Account and submit your organization, deployment, and camera requirements. Access is currently approved by Tensor Dynamics. After approval, the same Account page shows the license key and Downloads; you do not wait for a second email containing the key. Website sign-ins do not consume server activations.

The identity chain: the website account identifies the person → organization approval controls product access → the license key activates one Sentinel server → a local API key authenticates integrations. A Docker account, Sentinel API key, and website sign-in are not App user logins.

Prepare these four things

A powered camera or NVR visible on the local site network.
Its RTSP URL, username, and password. Keep these in your own password manager; never send them to support.
Tensor Dynamics Watch installer and license key.
Optional: a phone with Telegram for notification testing.

Keep three keys separate: the website-issued license key activates Sentinel; a Gemini/OpenAI AI provider key is supplied by the customer and stays in the customer environment; a Sentinel API/MCP token is generated locally by the App. They are not interchangeable.

1. Install and activate

Install and open the App, then paste the license key in the activation window. On the first start, wait for the top-right backend status to read READY. If it still says LOADING, do not add 64 cameras yet—wait or review license and startup help.

2. Connect the first camera

Choose Add camera in the sidebar. Try ONVIF discovery first; if it does not find the device, enter RTSP manually. An RTSP URL normally looks like:

rtsp://username:password@192.168.1.50:554/stream1

Save and connect only this camera. Success means the tile shows video and its state becomes online/connected. If it fails, check power, IP, credentials, and same LAN/VPN first, then see the camera brand guide.

3. Choose inference and test it

On Apple Silicon, you can select the bundled local engine. On Windows/Linux, choose a cloud provider with your own provider API key, or enter an endpoint for a GPU you manage. Press Test Connection in Settings; green OK is the success signal.

4. Assign a mission

Choose a mission preset in Settings, or write a natural-language instruction on the camera tile that states area + behavior + exception. Example: Watch the rear-door area for loitering; cleaners between 09:00 and 10:00 are expected. Do not write only “watch for anything unusual.”

5. Start monitoring and verify

Press Initialise System/connect. After one inference window, confirm the right-side log contains analysis for that camera. Run a safe real-world test, such as someone pausing at the entrance, and compare the description with your mission. Refine the mission if needed.

6. Enable alerts and Telegram (optional but recommended)

Open Agent → Channels → Telegram. The App includes a two-minute guide: create a bot through @BotFather, obtain the Bot Token and Chat ID, save them, then send /status to the bot. Success means it replies with the current system state. See notifications for commands and roles.

7. Confirm incidents are reviewable

Open the Database page and filter to the time of your test. Your delivery standard is: video is stable, the AI mission matches the site, the incident is recorded, and the right people receive alerts when required. Only then add the second camera.

Next: for different missions by booking time, see Calendar missions. For external systems, use Developer Center. Do not turn on every advanced integration on day one.

Overview