The workflow is Ableton Live’s: a grid of clips you launch against a global tempo grid, quantised to musical boundaries. The architecture is not Live’s, and that is the point.

Your set is text

A project is one line-oriented plain-text file with a byte-identical round-trip and stable per-entity uids. A melody is eight note <beat> <len> <pitch> <vel> lines. Diff a mix, branch an arrangement, review a track in a pull request, generate a set from a script.

The audio thread shares nothing

No allocation, no locks, no pointers into GUI memory: the interface reaches the engine through a lock-free command ring and a block of atomics. Launches are split inside the buffer, so a clip starts on the exact frame of its quantum rather than at the top of the next block. nxtaktd already runs that engine as its own process, playing clips out of a shared-memory sample pool.

CI plays what it builds

Every push builds the whole application, runs the headless suite — 2,500+ assertions across the engine, the IPC layer, the daemon and the devices — and then renders five passes of the demo set to FLAC, printing the peak and RMS of each into the run summary. The pipeline does not just compile this DAW. It hands you the audio.

github.com/nerdrx/nxtakt