Analysis updated 2026-05-18
Run a private, on-device AI companion that comments on your screen activity without sending data anywhere.
Try out a working example of an Electron app that combines local vision models with a desktop character.
Use the optional OpenAI cloud mode to compare cloud versus local vision model responses.
| samuraijack1337/rocky-companion | 000madz000/payload-test-api-route-handler | aliansari22/pdfclear | |
|---|---|---|---|
| Stars | 3 | 3 | 3 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | easy | easy |
| Complexity | 3/5 | 2/5 | 2/5 |
| Audience | vibe coder | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Needs Node.js 22.12+ and either a local Ollama install or your own OpenAI API key.
Rocky Companion is a desktop app for macOS that puts an animated character on your screen who reacts to what you are doing. It is built with Electron and TypeScript, and it is designed around privacy first. The character, Rocky, is based on a five limbed alien engineer from the novel Project Hail Mary, and the software implementation is MIT licensed, though the underlying character and story rights are not. How it works: the app periodically takes a single screenshot and sends it to a vision model, which decides what Rocky should say or do. You choose where that screenshot goes. The default option runs a model locally through Ollama on your own machine at localhost, so nothing is sent anywhere. There is also an optional cloud option using OpenAI, but only if you provide your own API key and explicitly opt in beyond the initial setup. Screenshots are never saved to disk in either case, they exist only in memory for the moment they are analyzed. The app also refuses to describe anything it judges sensitive, such as logins or private messages, and a stricter Classic mode limits Rocky to a small set of fixed categories like coding, meeting, or idle rather than free description. For most people, installation means downloading a ready made file. There is a signed installer for Windows and a disk image for macOS, both available from the project's Releases page. Because the macOS build is not yet officially signed by Apple, the first launch requires running one Terminal command to remove a quarantine flag, and updates may ask for a few one time permission approvals. Running the project from source requires Node.js 22.12 or newer, plus either Ollama installed locally or an OpenAI account if you want the cloud option. A short quick start guide walks through cloning the repository, installing dependencies with npm, optionally setting an API key in a local environment file, and launching the app in development mode with npm run dev.
A privacy-first desktop companion app for macOS that watches your screen and reacts in character, using a local AI model by default so nothing leaves your machine.
Mainly TypeScript. The stack also includes TypeScript, Electron, Node.js.
The code itself is free to use, modify, and share under MIT, but the Rocky character and its origin story are not covered and remain someone else's intellectual property.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.