Aegis Home (Chat)
Aegis Home is your command center. It's a conversational interface where you talk to your AI security agent in natural language — ask questions, get summaries, give commands, and explore what's happening across your property.
Click the Aegis Home icon in the sidebar to open it.
What Makes This Different From a Generic Chatbot
Aegis Home isn't ChatGPT with a camera skin. The agent has real-time awareness of your entire security system:
| Context Source | What It Gives the Agent |
|---|---|
| Camera states | Which cameras are online, what they last saw, current stream status |
| VLM analysis | Latest AI descriptions from every camera — the agent knows what's happening right now |
| Clip history | Access to all recorded clips with timestamps, camera names, and AI descriptions |
| Memory | Everything it's learned — family members, routines, vehicles, preferences |
| Event handlers | Active alert rules and their trigger history |
| Tools | Web search, weather, camera control, memory writing, and any tools installed by skills |
| Soul personality | The name, tone, and behavioral instructions you've configured |
This means when you ask "was anyone on the porch today?", the agent doesn't guess — it searches actual clip data and VLM descriptions to give you a factual answer.
What You Can Ask
Camera Queries
| Question | What Happens |
|---|---|
| "What's happening on the front door camera right now?" | Agent retrieves the latest VLM analysis for that camera |
| "Was anyone on the driveway between 2pm and 4pm?" | Agent searches clip descriptions for that camera and time range |
| "Show me the last 5 clips from the backyard" | Agent retrieves and summarizes recent clips |
| "Which cameras saw a delivery today?" | Agent searches all cameras' descriptions for delivery-related events |
| "Is Mom home?" | Agent checks recent descriptions and memory (e.g., "Mom drives a blue Honda") |
Summaries and Reports
| Question | What Happens |
|---|---|
| "What happened last night?" | Agent compiles a summary of notable events across all cameras |
| "Give me a daily security report" | Agent generates a structured report of all activity |
| "How many people came to the door today?" | Agent counts matching events and provides details |
| "Was there anything unusual this week?" | Agent identifies anomalies compared to learned routines |
System Control
| Command | What Happens |
|---|---|
| "Remember that Max is our golden retriever" | Agent writes to memory — future observations will reference Max by name |
| "Create an alert for packages at the front door" | Agent creates an event handler for you |
| "What time does the mail usually arrive?" | Agent searches observations and memory for mail delivery patterns |
| "Turn off alerts for the backyard camera" | Agent disables event handlers for that camera |
General Conversation
The agent is also a general-purpose AI assistant. You can ask about:
- Weather (uses weather tool)
- Web searches (uses web search tool)
- General knowledge questions
- Coding, math, or research tasks
- Anything a capable LLM can answer
Chat Interface
Message Input
Type your message in the text box at the bottom. Press Enter to send. The agent responds in the conversation thread with formatted text, and (if Auto-TTS is enabled) speaks the response aloud.
Push-to-Talk
Click and hold the microphone button to speak a question or command. Release to submit. Your speech is transcribed to text and sent as a regular message. This enables fully hands-free interaction.
Message History
The conversation persists across sessions. Scroll up to see previous messages. The agent maintains context from recent messages, so follow-up questions work naturally:
- "What happened on the porch at 3pm?"
- Agent responds with a description
- "Was it the same person as yesterday?"
- Agent compares today's description with yesterday's observations
Response Formatting
The agent formats responses with:
- Bold text for emphasis
- Bulleted lists for multi-item responses
- Timestamps for event reports
- Camera references so you know which camera observed what
- Inline suggestions when it recommends an action (e.g., "want me to create an alert for this?")
Agent Context: How It Knows What It Knows
Every time you send a message, the agent constructs its context from multiple sources in a specific priority order:
- Soul personality — highest priority, shapes tone and behavior
- System state — camera statuses, active tools, current time
- Relevant memories — facts about your family, routines, and environment
- Conversation history — recent messages for continuity
- Your message — what you just asked
This layered context is why the agent can answer questions like "is that the same delivery driver from Tuesday?" — it combines current VLM analysis, historical clip descriptions, and memory to reason about the answer.
Tips for Effective Interaction
- Be specific about cameras: "What happened on the front porch?" is better than "What happened?"
- Use time ranges: "between 2pm and 4pm" helps the agent narrow its search
- Teach through conversation: Tell the agent things and ask it to remember — this builds memory over time
- Use follow-up questions: The agent maintains context, so "What about the backyard?" naturally follows a previous camera query
- Ask for summaries: "Give me a summary of today's activity" is a powerful way to catch up
- Let the agent suggest: If it notices unusual patterns, it may proactively suggest creating an alert or updating memory