Observer AI logo

Observer AI

An open-source app that turns your screen, webcam or an old phone's camera into an AI watcher: you describe what to look for, and a small model running on your own device calls, texts or messages you when it happens.

Last verified: November 11, 2026
Alternative to: Ring smart alerts
Observer AI app screenshot

Short description

Observer AI builds small "micro-agents" that watch a screen, a camera, the microphone or the clipboard and do something when an event happens. You type a sentence such as "when my Steam game downloads, Telegram me" and Observer builds the agent for you. Each agent runs in a loop. It captures a frame or a transcript, asks a model about it, and then runs a short piece of JavaScript that decides what happens next: log the event, save an image, record a clip or send you an alert.

The models can run on your own device. The web app runs them inside a browser tab, the desktop app ships with llama.cpp, and the phone apps use models of about 0.8 billion parameters, which are small enough for an old phone. You can also connect your own Ollama, llama.cpp or vLLM server, or pay for Observer's hosted cloud monitoring.

Alerts can go to Discord, Telegram, email, Pushover, WhatsApp, SMS or a phone call. Observer is open source under the AGPL-3.0 license and is built by Roy Medina through Observer AI LLC, with the developer listed in Mexico. The project started in February 2025, the iOS app followed in February 2026, and version 3.2.4 came out on November 10, 2026. It's a good fit if you want to hear about one specific event, such as a finished render, a free machine at the gym or someone walking up to your car, without paying for a cloud camera subscription.

Features

  • Agents from one sentence: Describe what to watch and how you want to hear about it ("when an appointment is available, send me a WhatsApp") and the Agent Builder creates the micro-agent. You can also build one by hand in the Micro Agents tab.

  • Screen, camera and audio inputs: Agents can read a screenshot, a camera frame, clipboard text, text pulled from the screen, a transcript of your microphone, a shared tab's audio, or both audio sources mixed together for meetings. Audio is transcribed with Whisper.

  • Local models on any device: The browser app runs models with WebGPU, the desktop app uses llama.cpp and can load any GGUF vision model, and the phone apps run small models on the phone itself. On mobile the defaults are OneJev-0.8b for yes/no checks and Qwen 0.8B for agents that describe what they see.

  • Yes/no answers with a confidence score: A decision model answers a question such as "is there a person in the room?" with a probability instead of text. It can also answer multiple-choice or rating questions, and it can tell when a value has dropped, which suits "tell me when this finishes".

  • Calls, texts and chat alerts: Agents can phone you, send an SMS or WhatsApp message, post to Telegram or Discord, send an email or Pushover alert, show a browser notification or play a sound. SMS doesn't work for US or Canadian numbers.

  • Old phone as a watcher: The iOS and Android apps run agents on the phone's camera, so a spare phone can keep an eye on a washing machine, a bird's nest or a parked car.

  • Memory and video clips: Agents keep text and image notes, can read each other's notes, and can start, stop and label video clips around an event.

  • Phone lock: On mobile, an agent can lock every other app on the phone for up to 720 minutes until it sees the condition you set (iOS 16 or later and Android).

  • Desktop actions and a terminal command: The desktop app can show overlays and pop-up questions and can click for you. It also includes an observe command that calls you when a terminal command finishes.

  • Community agents: Browse ready-made agents other users have shared and run them yourself.

  • Your own model server: Point the app at any server that accepts OpenAI-style requests, such as Ollama, llama.cpp or vLLM.

Pricing

Agents that run on local models are free. Paid plans add hours of monitoring on Observer's own servers, more Agent Builder messages and more SMS, WhatsApp and call alerts. Observer has no public pricing page. These prices come from the pricing table in the app, and the limits come from Observer's Terms of Service.

  • Free: $0. Local models, logging and recording, plus Discord, Telegram, email and Pushover alerts. Cloud monitoring of 60 credits a day and 1,200 a month, which is about 30 minutes a day and 10 hours a month if the agent checks every 30 seconds. The Agent Builder allows 45 messages a day and 150 a month, enough for about 3 agents a day. SMS, WhatsApp and calls are capped at 5 each a day within a $0.50 monthly budget.

  • Plus: $8/month, web only (not sold on iOS). The same as Free, but cloud monitoring rises to 480 credits a day and 4,800 a month, about 4 hours a day and 40 hours a month.

  • Pro: $20/month, or $22.99 in the iOS app, with a free trial whose length isn't published. Cloud monitoring of 1,440 credits a day and 12,000 a month (about 100 hours). The Agent Builder allows 1,000 messages a day and 1,500 a month. SMS, WhatsApp and calls go up to 300 each a day within a shared $5.00 monthly budget.

  • Max: $80/month, or $99.99 in the iOS app. Cloud monitoring of 2,880 credits a day with no monthly cap. The Agent Builder allows 1,000 messages a day and 3,000 a month. SMS, WhatsApp and call limits match Pro.

  • Enterprise: A custom number of seats, a usage dashboard, direct support from the team and the option to run Observer on your own servers. Pricing isn't published; you book a meeting.

  • How credits work: One credit is one check by an agent on Observer's servers, so a slower loop makes credits last longer. The free tier's 60 daily credits cover 30 minutes at one check every 30 seconds, or 3 hours at one check every 3 minutes. Local models don't use credits. Daily limits reset at midnight UTC and monthly limits on the 1st, and unused credits don't carry over.

  • Annual billing: Not published.

Integrations

  • Local model servers: Ollama, llama.cpp, vLLM, LM Studio or any other server that accepts OpenAI-style requests.

  • Cloud model providers: When you use Observer's hosted monitoring, its Terms list Google AI Studio, OpenRouter and Fireworks.ai as the services that run the models.

  • Messaging: Discord webhooks, Telegram through Observer's notification bot, email and Pushover. WhatsApp, SMS and voice calls go through Twilio, and you first link your own phone by sending Observer a 4-word code.

  • Terminal: The bundled observe command wraps any terminal command and calls you when it ends.

Platforms

  • Web
  • macOS
  • Windows
  • Linux
  • iOS
  • Android
😕

What users complain about Observer AI

  • Sign-in needed even for local use: A Windows user expected an offline app and found they had to sign in before creating any agents. The developer said the sign-in was there for the hosted Agent Builder, and version 3.0.3 added a way to skip it by entering your own model server.

  • Too many upgrade prompts: One person who tried the phone app said it pushes paid plans so often that it gets annoying to use.

  • Home servers don't connect on Android: An Android user couldn't link the Play Store app to an Ollama server on their home network over plain HTTP, even though the same address worked in the phone's browser. The issue was still open on November 11, 2026.

  • Early versions needed lots of setup: A user who tried Observer about a year earlier found it flaky and slow to set up by hand. The developer says the one-sentence builder was made to fix that.

💡

Frequently asked questions about Observer AI

Can I use an old phone as a security camera with AI alerts?â–¼

Yes. Install Observer on the phone (iOS 15 or later, or Android), point the camera at what you want watched and describe the event, for example "when someone who isn't me approaches my car, send me a picture". A small model on the phone checks each frame and alerts you by Telegram, Discord, email, WhatsApp, SMS or a phone call. The default phone models have about 0.8 billion parameters, so they run on older phones too.

Is Observer AI free?â–¼

The app is free and open source under the AGPL-3.0 license, and agents that run on local models cost nothing. Paid plans add monitoring on Observer's own servers and more SMS, WhatsApp and call alerts: Plus is $8/month, Pro $20/month and Max $80/month on the web, with higher prices in the iOS app. The free tier includes 60 cloud credits a day along with Discord, Telegram, email and Pushover alerts.

Is there a Ring smart alerts alternative without a subscription?â–¼

Observer can do a similar job with no subscription if the model runs on your own phone or computer. You describe the event in a sentence, such as "someone from UPS shows up" or "my dog is at the back door", and the model checks the camera for that instead of a fixed list of alert types. It isn't a doorbell or a video storage service. It watches a camera or screen you point it at and sends you an alert, a picture or a short clip.

Does Observer AI keep my camera and screen private?â–¼

It can. If you use local models in the browser, desktop or phone app and send alerts to your own Discord webhook or server, nothing passes through Observer's servers. If you use Observer's cloud monitoring, frames go to the model providers it uses (Google AI Studio, OpenRouter and Fireworks.ai), and SMS, WhatsApp and calls go through Twilio. The source code is public, so you can read it or build the app yourself.

Is there an AI that can watch my screen and tell me when something happens?â–¼

Observer does this. Open the web or desktop app, share a screen or window and describe the trigger, for example "when my render fails, call me" or "when Claude Code asks for my permission, Telegram me". On each loop the agent takes a screenshot, asks the model whether the event has happened and sends the alert once it has.