AtmovioAtmovio

Atmovio 5 · Open source · Raspberry Pi 5

The sky keeps changing.
Keep its story.

Atmovio records your IP cameras, uses AI to follow the changing sky and turns interesting moments into video. From the first image to YouTube, in one administration.

Get started Explore the administration

Free software. Recordings at home. Your choice of AI provider.

ATMOVIO 5.3.4 · DEMO
Administration · image collection and AI evaluation
Administration · image collection and AI evaluation

One path from camera to video

  1. Camera recording
  2. Image collection
  3. AI assessment
  4. Saved clip
  5. Studio and YouTube
Recordings stay at home

Recording works without cloud AI.

44 phenomenon categories

Clouds, light, storms and changes over time.

AI under your control

Your provider, rules and daily limit.

An open project

MIT licence, REST API and webhooks.

What it does

Record without transcoding

RTSP cameras, H.264/H.265, recording retention and export through Frigate.

Your cameras in one place

ONVIF discovery, live previews and remote cameras over WireGuard.

Completed AI films

Image collection and full-sequence evaluation run separately. Inspect the images AI received.

Meaningful notifications

Selected phenomena, camera-specific thresholds and notification intervals. Email and webhooks.

Clips that follow the event

Connect adjacent AI films, set a fixed clip up to 120 minutes or select an exact time range.

Time-lapse, Studio, YouTube

A separate time-lapse score. Speed, intro, music and editable titles and descriptions.

Know how your device is doing

Storage status, diagnostics, newest-first logs and updates with a completion result.

AI sees the changes. Not just a moment.

An AI film is a time-ordered set of photographs. Choose the sampling interval and collection duration; AI evaluates the whole sequence when it is complete.

  1. 1

    Collect images

    Each camera has its own film. Set the interval, duration and image limit.

  2. 2

    Assess the whole sequence

    AI describes phenomena, development and time-lapse interest. You can inspect its source images.

  3. 3

    Apply your rules

    Phenomena and time-lapses have separate criteria. A failed notification does not block a clip from an AI film.

  4. 4

    Follow a continuing event

    An event at the end of a film can continue into the next. Bounds depend on available images and recordings.

Providers

  • Gemini · OpenAI · Anthropiccloud models with your own key
  • Groq · OpenRouteradditional model choices
  • Ollamalocal or self-hosted
A real AI input example · 13 images across an hour
A real AI input example · 13 images across an hour

44 categories of sky phenomena and development

  • Clouds and structures
  • Storms and precipitation
  • Light and the night sky
  • Fog and other phenomena
  • Motion and development
  • Custom phenomena

AI FILM

Settings you can understand

Sampling interval, image count and film duration side by side. The preview shows what AI will receive.

Example: one image every 5 minutes → 9 images across 40 minutes → one evaluation per camera.

A shorter interval helps capture fast changes. More images can mean less detail per image. AI can make mistakes.

AI film · duration calculation and input preview
AI film · duration calculation and input preview

From a recording to a video worth sharing

Choose the time range, set the speed and add your own style. Studio prepares the video and editable publishing details.

  1. 1

    Choose the range

    Use the detection, a fixed duration or exact start and end times. The source recording must be available.

  2. 2

    Set the pace

    10–240× speed, your intro, on-screen text and music from Openverse or your own file.

  3. 3

    Review the result

    Edit the title and description derived from AI evidence. Studio also flags skipped optional elements.

  4. 4

    Publish your way

    Download MP4 or upload to your connected YouTube channel. Enable automation when you want it.

YouTube requires a Google OAuth connection. Upload availability depends on your project settings and quota.

YouTube setup guide →
Studio · time range, speed, intro and music
Studio · time range, speed, intro and music

Take a look inside

Current Czech administration, version 5.3.4, with demo data. The AI image sheet above is a real camera example. Click any image to enlarge it.

Camera configuration
Camera configuration
Phenomenon groups and AI rules
Phenomenon groups and AI rules
AI video storage and duration
AI video storage and duration
System · passwords and access
System · passwords and access

Hardware

Raspberry Pi 5, IP cameras and a recording disk. Capacity depends on resolution, bitrate and the other tasks running on your device.

Hardware guide →
BoardRaspberry Pi 5, 4 GB min., 8 GB recommended
SystemRaspberry Pi OS Lite 64-bit on SSD or SD card
RecordingsUSB 3 hard drive. Mbit/s of all cameras × 10.8 ≈ GB per day
PowerOfficial 27 W supply, active cooler
NetworkWired Ethernet. Remote cameras over WireGuard
CamerasAny RTSP camera (H.264/H.265), ONVIF preferred

Install

Prepare Raspberry Pi OS Lite 64-bit and a recording disk. Run the installer and finish setup in the web administration.

  1. Download install.sh from the latest release and run it as root.
  2. Open http://<pi-ip>: cameras → disk → alerts → AI key.
  3. Updates arrive in the admin. One click.

Plain HTTP by design: keep it in your LAN or behind a VPN. Never expose port 80 to the internet.

Installation guide →
Raspberry Pi 5 · Raspberry Pi OS Lite 
curl -fsSL -o install.sh https://github.com/VladimirVecera/atmovio/releases/latest/download/install.sh
sudo bash install.sh
Prefer git? Clone the repository – install.sh is in its root: 
git clone https://github.com/VladimirVecera/atmovio.git
cd atmovio
sudo bash install.sh

♥ Support the project

Atmovio is free – keep it that way

Server, cameras and AI credits cost money. Any amount helps – QR payment or bank transfer, no registration.

⭐ Star on GitHub

API & Home Assistant

Atmovio is not a closed box. Status, cameras, snapshots, detections and videos are one GET away – and Home Assistant needs no custom component.

Home Assistant

Sensors, camera entity and a phone notification with the picture – built-in REST integration, copy the YAML.

rest:
  - resource: http://pi/api/v1/status
    sensor:
      - name: Atmovio disk used
        value_template: "{{ value_json.storage.disk_pct }}"
Home Assistant YAML →

REST API

Status, cameras, snapshots, detections, videos and events as JSON with API keys. PHP scripts included.

GET /api/v1/status
GET /api/v1/cameras/zahrada/snapshot.jpg
GET /api/v1/detections?limit=5
GET /api/v1/videos
Authorization: Bearer at_…
API reference →

Webhook

Atmovio pushes status every minute and every alert with its picture to your website. PHP receiver included.

POST https://your-site/webhook.php
{"kind": "sky", "score": 9,
 "phenomena": ["cervanky"],
 "image": "<base64 jpeg>"}
Webhook docs →
API & Home Assistant guide →

Open source

Made by Vladimír Večeřa for his own roof. MIT license. The admin is in Czech for now, the docs are in English and Czech – contributions welcome.

GitHub Contribute

FAQ

What is an AI film?

Time-ordered images from one camera. In completed-film mode, AI evaluates them after collection ends. It is not a video sent to the provider.

Can it find a good time-lapse without a storm?

Yes. It also assesses cloud-layer motion, growth, dissipation and light changes. Automatic clips can use an independent time-lapse score threshold.

What if an event continues into the next film?

Adjacent observations can form a continuing event. Fixed-length clips follow the configured limit; you can also adjust the exact range manually.

What does AI cost?

Atmovio is free. Providers set model availability, pricing and any free tiers. You use your own key and can set a daily request limit.

What leaves the Raspberry Pi?

Cloud AI receives selected images or a composed image sheet and instructions. Videos go to YouTube when you start an upload or enable automation. Webhooks send the data you enable.

Does it work offline?

Local recording and administration do. Cloud AI, YouTube and internet notifications need a connection. Ollama can run locally or on your own server.

Is detection always reliable?

No. Results depend on the model, image quality and sampling interval. AI can miss or misidentify phenomena; this is not a weather warning system.

How much disk space do I need?

Combined camera bitrate in Mbit/s × 10.8 approximately gives GB per day of continuous recording. Allow extra space for exports and finished videos.

How do I access it remotely?

Use a VPN such as WireGuard. Do not expose the home-network administration directly to the internet.