Audio engine
Audio path before AI
Why Tojito started with a reliable local audio path—and Linux-first PipeWire work—before adding models.
Early in Tojito’s development we made a deliberate choice: prove a trustworthy local audio path before introducing AI.
Decision recorded: July 2, 2026. The first product work focused on capture → process → output on the desktop—moving sound reliably, with local-first processing as the baseline—rather than bolting a model onto an unproven pipeline.
That meant investing in the audio engine itself. Development proceeded Linux-first, including PipeWire-oriented live audio, so we could iterate quickly on a real desktop path. The goal was simple: audio should flow cleanly through the system with no transformation yet, so later AI stages would sit on a foundation we already trusted.
Why this order? Local-first products fail quietly when the audio path is fragile. Buffering, devices, shutdown, and long-running sessions are not glamorous, but they decide whether offline processing feels calm or broken. AI can wait until that path is honest.
Profanity filtering and richer speech understanding came later, as consumers of a stable local pipeline—not as the first line of code.