Hey PH 👋 I'm a musician, not a mastering engineer. For years I had two bad options: pay a studio for every mix, or upload my song to a subscription black box and get a slightly louder stereo file back with no explanation of what changed. What actually pushed me to build this was realizing my masters were never the problem — my mixes were. No amount of AI mastering saves a vocal that's buried in the second verse. That's a mixing decision, and every tool I could find starts after the mixing is done. So StemMaster starts a step earlier. You load the stems — 4 or 40 — and it mixes and masters the whole thing: rides the vocal fader word by word, ducks the bass under the kick (only the colliding frequencies, only while they collide), automates verses against choruses, then masters to streaming or club loudness. The rule I built everything around: it's not allowed to be a black box. The console faders animate in real time as its automation rides your mix — that's the actual UI, not a visualization — and an Engine Notes panel writes every decision down in plain English: "Unmasking: Vocals → Guitar @ 2.5k", "shaved 5.9 dB of peaks before the limiter". Disagree with any of it? Grab a fader and it re-renders around your move in seconds, loudness-matched so louder can't cheat the A/B. How the approach evolved: that transparency rule turned out to be an architecture constraint, not a UI feature. Every DSP stage had to become evidence-gated — the app measures first and only applies what the measurements justify, so if your low end is already mono it won't touch it. The optional LLM engine (bring your own key) proposes parameters, but a sanitizer clamps everything to safe ranges — it can ask for +40 dB of gain and it'll get +12 — and a closed loop re-analyses the render against the target before you ever hear it. The genuinely interesting engineering ended up in unglamorous places: repairing truncated LLM replies so a token cutoff doesn't discard 20 good stem decisions, and caching the expensive per-stem half so a fader tweak re-renders in seconds instead of minutes. Local-first: your audio never leaves your machine, no account, no subscription. The free demo is the full app with exports disabled. Windows 10/11 only for now — Mac is the most common request and I'm listening. Throw your messiest session at it and tell me where it falls over. AMA about the DSP, the AI, or shipping a desktop app solo.

StemMaster

AI powered Mixing and Mastering suite

AI powered Mixing and Mastering suite