in this demo, i’m showcasing a new AI agent in beta that automates tedious, repetitive workflows on your desktop without needing additional integrations. all you need to do is install screenpipe and define the workflow.
in this example, ai agent is logging into my gmail account, finding invoices from x, downloading them, renaming the files according to the proper structure (invoice-supplier-date), and then uploading the invoice to the correct folder in my dropbox.
this is just an example of how i can save a few minutes on something i’d otherwise do again and again, hundreds of times.
the workflow is defined in two ways:
- either you press a record button and show screenpipe what you do on the desktop, then the screenpipe agent tries to repeat it until you’re satisfied with the outcome;
- or a gpt-vision model defines a step-by-step plan based on your goal and tries to execute it. each time it successfully performs a task, it records the coordinates and key presses to pass them to a faster downstream workflow, which can quickly perform the recorded actions without needing to process each frame separately.
available for beta users now!
download here: screenpipe
build from source: GitHub - mediar-ai/screenpipe: 24/7 local AI screen & mic recording. Works with Ollama. Llama3.2 control your computer. Alternative to Rewind.ai & Zapier. Open. Secure. You own your data. Rust.
what is screenpipe?
#1 trending fully open source repo on github last week
thousands of people are already using and building on top of it