A ledger of the ledger.

Tell me what's wrong.

• New “Test connection” button in Settings → Account — checks whether your dictation can reach our servers on each route and explains, in plain language, what's blocking it (network, sign-in, or credits). Free; it doesn't transcribe.

The real fix for slow networks.

• Recordings now upload in a compressed format about 5× smaller, so dictation reaches the server even on congested networks where the larger upload used to stall. This is the true cause behind the "Something went wrong" errors some users saw after speaking — especially in Vietnam.

A lighter touch on the network.

• The route check now pings a 93-byte health endpoint instead of downloading the full model list — checking your network routes uses almost no data.

Routes that follow you around.

• Haynoi re-checks its network routes whenever your connection changes — new Wi-Fi, VPN on or off, waking the laptop somewhere else — not just at launch.

Three roads, raced — your network picks the winner.

• Haynoi probes its API routes at launch and dictates over the fastest one that works from your network. A stalled route triggers a parallel race after 4 seconds, and the winner is remembered — a bad network costs seconds once, not on every sentence.

For everyone who kept seeing “Something went wrong”.

• Dictation now works on networks where the audio upload stalls (reported from Vietnam) — Haynoi automatically retries through an alternate route.

Buttons that do what they say.

• The Settings gear and Sign in buttons in the menu bar panel work again — they had been calling a retired macOS hook that silently did nothing.

Out of the Dock, into the menu bar.

• Haynoi is now a menu-bar app — no Dock icon while you work; it appears only when the main window is open.
• The menu bar panel lists your 5 most recent dictations with a copy button on each row.

The panel that was hiding in 10 pixels.

• Clicking the menu bar icon shows the panel again — it had been opening collapsed to a 10-pixel sliver, which looked like nothing at all.
• Insights, tidied: one headline, one unified row of stat cards (words · pace · streaks), the heatmap, and where your words go. Duplicate app rows merged.

Smooth again.

• Fixed the sluggishness after dictating — an internal signal was re-rendering the UI ~47×/second for a minute after each dictation. Scrolling and the menu bar panel are smooth again.
• The copy button is back on every history row — always visible, rows never shift. Recent is gone; it duplicated History.

History that holds still.

• History rows no longer shift on hover and the language badge is gone — Copy and Delete moved to the right-click menu.

Instant to the touch, cosmic to the eye, quieter to the ear.

• The press registers almost immediately — the start chime now lands about a tenth of a second after your finger does.
• The indicator moved to the top-center of the screen, in a dark cosmic capsule with an aurora halo that swells only with your voice.
• Exactly two tones per dictation: press and release. The “N words” chip appears silently; no more violet dot while transcribing.

A waveform that tells the truth, and a kinder pair of bells.

• New recording indicator: Trail — a scrolling waveform where your last second of speech stays visible. Silence is a flat line; the trail only moves with your voice.
• New default sound theme: Chime — a gentle bell pair on start and stop. Previous themes remain in Settings.
• When your text lands, a small “N words” chip pops up with a subtle chime — both can be turned off in Settings → Sounds.

Dictation now starts the instant you press the key.

• The microphone stays warm between dictations — no more cold-start wait when you press the key, even on Bluetooth headsets.
• A 300ms pre-roll is captured before recording officially begins, so your first word is never clipped again.
• The start tone now plays exactly when the mic is truly live — when you hear it, you can speak.
• The recording waveform follows your real voice: silent means still, speaking means moving.
• Pasting is ~250ms faster when dictating into the app you're already in.
• Hotkey hints corrected everywhere: hold left Option (⌥) alone — there was never a Space involved.

The first public release.

• Push-to-talk dictation: hold ⌥ (or ⌘ / ⌃ / fn), speak, release — polished text lands in whatever app you're using.
• Vietnamese-first transcription that keeps mixed English intact, with a language preference: auto-detect, Tiếng Việt, or English.
• Your words are never lost — failed transcriptions save the recording locally and retry with one click.
• Recording survives real life: AirPods disconnecting mid-sentence, permission hiccups, and flaky Wi-Fi all degrade gracefully.
• Auto-paste with clipboard restore afterwards, and dictated text stays hidden from clipboard managers.
• A living status orb: recording, transcribing, success, and error each have their own state.
• Custom dictionary, spoken snippets, smart modes (Normal, Clean, Email, Auto), and music auto-pause while you dictate.
• Credit balance in Settings with clear out-of-credits guidance — pay as you go through Kyma, no subscription.
• Quality and Fast transcription tiers, searchable local history, sound themes, and silent auto-updates via Sparkle.
• The Mercury interface: a private ledger of everything you've said, on paper and ink, with one aurora hairline.

Download 0.1.0 from GitHub Releases