melune turns spoken sessions into a coaching film. After each conversation, it replays the moments that mattered, identifies the patterns behind them, and builds a targeted practice plan around what needs work. Not a streak tracker — a real diagnosis of how your speaking is actually changing.
The product loop is Speak → Replay → Diagnose → Drill → Retest → Update Evidence. You sit down for fifteen minutes, speak with the tutor in your target language, and end the session. But the session is not the product — the diagnosis is.
Afterwards, a written replay opens. Not just a transcript: a coaching surface. Moments are ranked by what they reveal — a grammar pattern that keeps recurring, a vocabulary gap under time pressure, a near-miss you paraphrased your way past, a well-handled structure worth naming. Each moment has the original audio, an improved version, and a brief explanation of why it matters.
Over time, these moments accumulate into a pattern model. "You tend to avoid dative phrases in service contexts." "Sentence-final particles are inconsistent when speaking faster." Each pattern has a confidence level, an evidence count, a freshness timestamp, and a scheduled retest. The next session is built around what needs pressure.
The goal is an honest picture of how your speaking is changing — not a level badge that moves every week. The proficiency estimate updates slowly, with a visible confidence band. The system explains when it doesn't have enough evidence to say more.
These three principles shape almost every design decision. They explain what gets left out as much as what's included.
The product loop is Speak → Replay → Diagnose → Drill → Retest. Conversations are where evidence is collected. Replay is where learning happens.
Corrections explain themselves. Proficiency estimates show their confidence. The curriculum states what it's targeting and why. Low-confidence diagnoses are labeled as uncertain.
No streaks as guilt-trips, no XP, no leaderboards. Sessions cap at fifteen minutes by design — short, spaced practice has better evidence for lasting gains than long, infrequent study.
Honest snapshot of the build. Items marked working are wired end-to-end in internal alpha; in progress is the current active development wave; planned means designed but not yet started.
macOS first. melune is built natively in SwiftUI for the desktop. Sessions happen at a real keyboard with real audio hardware, which is where careful practice tends to happen anyway. An iOS companion will follow.
Subscription or BYO API key. You'll have the option to bring your own Gemini API key or subscribe for us to manage it. BYO is always available; managed pricing is being finalized. Typical BYO usage runs around $10–14 per month at a daily practice cadence. Managed tiers will cap conversation minutes by design — short, spaced sessions have better evidence for lasting gains than unlimited marathons.
Two languages at the start. German and Korean are the initial tracks. The data model and analysis prompts are language-agnostic; the depth comes from language-specific pattern families, scenarios, and placement materials. More languages come after those two feel right — not broad, deep.
A small, careful release. The first public build will go out to a small group, not a wide launch. There will be a way to put your name down before that happens — it's not wired up yet.