Ready-to-Run Demo#

Run Smart Community end to end with an open-source video.

All files are in examples/venue-monitor.

Prerequisites#

Finish the Prerequisites and Step 1 - Start all services in Get Started.

Start#

From the component root (agentic-smart-community):

# Optional, in China: use the Hugging Face mirror.
export HF_ENDPOINT=https://hf-mirror.com

bash examples/venue-monitor/start-demo.sh

If the demo is already running, it is stopped first and then started again. This downloads the video, pushes it as rtsp://localhost:8010/live/venue, starts the stack, and registers the venue_monitor use case and the cam_venue monitor.

Note

For a Chinese prompt (descriptions and reports in Chinese), run PROMPT_LANG=zh bash examples/venue-monitor/start-demo.sh instead.

Verify#

ffprobe -rtsp_transport tcp rtsp://localhost:8010/live/venue
curl -fsS http://localhost:3101/health

Open http://localhost:3100/ and select cam_venue to see the live preview.

Connect with OpenClaw agent#

Follow Get Started - Step 3 to install the OpenClaw on host, and then register the MCP server with OpenClaw.

Create the agent and enable alerts#

bash examples/venue-monitor/openclaw-adapter/install.sh

This creates venue-monitor-agent and routes cam_venue alerts to it. Run openclaw dashboard to open the Control UI (http://localhost:18789), where alerts appear in the agent’s session.

Schedule a daily report#

Cron job

Schedule

Agent

Session

Behavior

venue-daily-report-22

Daily at 22:30

venue-monitor-agent

daily_report

Generates the daily report for cam_venue: foot-traffic statistics and abnormal events.

Replace Asia/Shanghai with your timezone:

openclaw cron add --name venue-daily-report-22 --cron "30 22 * * *" --tz Asia/Shanghai \
  --agent venue-monitor-agent --session "session:daily_report" --session-key agent:venue-monitor-agent:daily_report \
  --no-deliver --message "Generate today's daily report for cam_venue."

Talk with the agent#

In the Control UI, select venue-monitor-agent and ask, for example:

  • “What is happening on cam_venue right now?”

  • “Were there any alerts on cam_venue today?”

  • “Generate today’s report for cam_venue.”

Stop#

bash examples/venue-monitor/stop-demo.sh

This stops the stream and the app tier, and keeps the model serving warm. To stop everything, run bash setup_docker.sh --down.

Learn More#

Use case#

venue_monitor classifies each clip and counts the distinct people in it (people_count, stored in the database and used for the daily report’s foot-traffic statistics); warn events raise alerts.

Event

Severity

fight

warn

fall

warn

running

warn

multiple_people

info

person_present

info

empty

info

The prompts are in prompts. Re-running with a different PROMPT_LANG updates the registered prompt.

Video source#

By default, people-detection.mp4 is downloaded from Intel’s open-source sample-videos repository on the first run.

Use your own local video (any file ffmpeg can read, H.264 recommended):

VIDEO_FILE=/absolute/path/to/video.mp4 bash examples/venue-monitor/start-demo.sh

Use an existing RTSP stream (nothing is pushed; cam_venue reads your URL):

RTSP_URL=rtsp://<host>:<port>/<path> bash examples/venue-monitor/start-demo.sh

Stream settings#

PLAYS_PER_HOUR controls the pushed video (default 6; not used with RTSP_URL):

Value

Behavior

N

Play N times per hour; the last frame is held between plays

-1

Loop continuously, no hold (a very large N, 72+ for this video, does the same)

0

Play once

PLAYS_PER_HOUR=12 bash examples/venue-monitor/start-demo.sh

To change only the stream, run examples/venue-monitor/stop-stream.sh, then examples/venue-monitor/start-stream.sh.