Free no-code AI automation for existing CCTV
Demo: https://youtu.be/FNZ72Bq5KGI
Hey all, I'm the developer of Grablo Vision and wanted to share it here. It's free for personal, non-commercial use.
The idea is simple: point AI detection at cameras you already own, then build rules with no code. When something is detected, run these actions.
It scales the way you'd expect: add as many cameras as you want, run multiple AI analyzers on a single camera, and attach multiple rules (automations) to each analyzer.
Detection (4 types)
- Object Detection: pick a class (e.g. person), set a confidence threshold and a region of interest
- Face Recognition: enroll faces, then trigger on known vs. unknown
- License Plate (LPR): read plates and match against a registered list
- Fire Detection
Camera input: RTSP, ONVIF, or a directly connected USB camera.
Triggers & conditions
- Detected / not detected (or known/unknown, registered/unregistered)
- Sustained-for duration, so a brief flicker doesn't set it off
- Schedule: restrict a rule to certain days and hours
- Ask LLM: a plain-language confirmation step before acting, e.g. "only if the person is doing something suspicious or dangerous" (Object detection)
Actions, chained freely in sequence. Stack as many as you want, in any order, with delays between them. For example: announce over a speaker, wait 5s, send a photo, then turn on a light. Available actions:
- Notify: Push, Telegram, Email
- Log the event with a snapshot
- Text-to-Speech announcement, or Play Audio/Video
- Control external devices via Zigbee or I/O relays (open a gate, flip a light or siren)
- MQTT publish or HTTP request (webhook)
- Delay between steps
Home Assistant integration: separate from the rules above, each AI analyzer can expose its detection state directly as a Home Assistant sensor, so your existing HA automations can react to it.
Runs on hardware you already have: a Raspberry Pi, Jetson, or any Linux / Windows / macOS machine (x86 or ARM). It's also available as a Docker image or a Home Assistant add-on.
Detection runs on your own machine and streams peer-to-peer, so footage isn't sitting in someone else's cloud.
A few things people set up with it
- Face: send a photo alert when an unknown face appears, and log known faces silently
- License plate: raise the parking barrier for a registered plate, and keep a log of every plate that pulls in with a timestamp and snapshot
- Fire: sound a siren and send an immediate alert if fire appears in the garage
- Person, at night only: announce "someone is at the door" over a speaker between 10pm and 6am
- Suspicious behavior (LLM check): if a person is loitering or acting oddly, flip a relay to turn on a light and push a notification
It's still evolving, so I'd genuinely welcome feedback and feature ideas.
Link: https://vision.grablo.co
