roy3838/Observer — reverse-engineered prompt

Build me a privacy first Observer AI app where I can create little agents that watch what is happening and react automatically. I want a clean dashboard where I can make a new agent, give it a name, description, system prompt, model, loop timing, and choose inputs like screen, screen text, camera, camera text, clipboard, microphone, screen audio, and shared memory. The agents should work with local AI models when possible, keep data on my device, and be able to log text and images, remember things between runs, record clips, and label moments.

I also want a simple code area for each agent so the model response can trigger actions like sending email, Discord, Telegram, WhatsApp, SMS, Pushover, browser notifications, phone calls, popups, overlays, and basic clicks. Include handy built in functions for memory, time, sleep, starting and stopping other agents, and recording controls.

Please make it usable as a real app with a web experience too, and add one demo agent like spotting a logo on screen and sending an alert. If you need details, check the current docs online.