Agents

What are agents?

Think of an agent as your on-duty assistant that “watches” video events and decides what matters. You tell it what to look for (instructions), who to notify (contacts), and you can give it documents or images for extra context.

Create an agent (step‑by‑step)

1. Open Agents and click “Add Agent”.
2. Name: a descriptive label (e.g., “Loading dock safety monitor”).
3. Instructions: describe what to watch for, in plain language. Be specific (who, where, when, severity).
4. Agent Rule (optional): add concise criteria used for alerting/decision logic.
5. Alert contacts: pick who should be notified. If you have none, create them in Directory.
6. Context files (optional): upload PDFs/images to give site policies, maps, signage, or examples.
7. Save.

Assign an agent to streams

1. Go to Streams and edit a camera.
2. Set the Agent field to the agent you created.
3. Save. New events from that stream will be analyzed by the agent.

Alerts and contacts

Alerts are sent to the contacts you assign to an agent. Keep the list focused to reduce noise. Use the Directory to add or update contacts.

Provide context files

Upload site maps, SOPs, or example images to help the agent understand your environment and policies. Files are tied to the agent and used during analysis.

Best practices

• One agent per distinct task (e.g., “PPE compliance”, “After-hours entry”).
• Keep instructions explicit and scoped; avoid vague goals.
• Start with fewer streams per agent, validate alerts, then scale up.
• Use context files for site-specific nuance (restricted zones, signage).

Troubleshooting

• No alerts? Ensure the agent is assigned on the stream and contacts are set.
• Too many alerts? Tighten instructions, add clearer rules, or reduce contact list.
• Missed events? Add context files, clarify times/locations, and verify stream quality.