abhi-arya1/riff

Riff is an early stage voice assistant meant to sit in a car and control Spotify hands free, built to run on a laptop rather than as a packaged product. It listens through a browser using a real time voice connection from OpenAI, can interrupt itself naturally the way a person would in conversation, and remembers explicit preferences a user tells it to keep. The author describes it as a demo project that expects the person running it to already have their own accounts, API keys, and command line tools set up. Simple playback commands like pause, resume, skip, checking what is playing, and adjusting volume go straight to Spotify through a short direct path. More involved requests, such as getting recommendations, working with playlists, looking at listening history, or doing live web research, get handed off to a separate coding agent called Pi, which is given a fixed and limited set of tools rather than open access to the file system or network. Explicit preferences and named listening routines get saved to a local memory system, while secrets, exact locations, routine playback actions, and full conversation transcripts are deliberately left out of that memory. Riff also has optional support for controlling a small set of Tesla features over Bluetooth, limited to things like media volume, climate, cabin temperature, and seat or steering wheel heating. It explicitly does not touch locks, doors, the trunk, driving, charging, navigation, keys, or security settings, and the README is firm that any Tesla pairing or setup should only be done while the car is parked. Setting it up currently requires a Mac, the Bun JavaScript runtime, a Pi compatible coding agent on the command line, the Studio by Spotify Labs application along with its bundled command line tool, an OpenAI API key, and a Vercel AI Gateway key. None of those third party applications or accounts are included in the repository itself. The project states it is not affiliated with Spotify, Tesla, OpenAI, Vercel, Pi, or Supermemory, and it is released under the MIT license.