SpeechcapDownload
ChangelogRelease notes

What we shipped.

Every release of Speechcap, newest first. Auto-updates ship within hours of a publish; if you're on a recent build, the in-app banner will point you here.

  1. July 27, 2026

    v1.2.0

    Free is now cloud-powered (2,500 words/week, zero setup); Pro gains offline/on-device privacy, transforms, 89-language translation, and unlimited dictation.

    Added
    • Free plan now includes cloud Whisper transcription and cloud AI cleanup (filler removal, smart punctuation, context-aware formatting) — no model downloads, works immediately on any Mac, up to 2,500 words per week.
    • Pro plan now includes the full on-device / offline pipeline (Whisper + local AI): audio never leaves your Mac, so it's the right choice for sensitive content or air-gapped environments.
    • Pro unlocks the transforms menu (Improve / Formal / Email / Fix grammar / Refine prompt / Custom) and 89-language translation.
    • Pro removes the weekly word cap — dictation is unlimited.
    Improved
    • Free-tier onboarding is now instant: no offline pack download required, dictation works the moment you press the hotkey.
    • Billing screen and upgrade flow updated to reflect the new tier benefits.
    Removed
    • On-device / offline mode is no longer available on the Free plan — it moves to Pro as the privacy-first power tier.
    • Transforms and translation are no longer available on the Free plan.
  2. July 5, 2026

    v1.1.0

    Offline goes free: unlimited on-device dictation, AI cleanup, transforms, and translation on the Free plan. Pro is now the cloud engine — instant setup, fastest latency, any Mac.

    Added
    • The Free plan now includes the full on-device pipeline: unlimited Whisper transcription, AI cleanup (Granite 4 Micro), the transforms menu, and translation in 89 languages — all running locally, nothing leaves your Mac.
    • Transforms and translation now work fully offline on every plan: the selection menu, in-flight transform letters, and always-on translation route through the local model when the on-device engine is active.
    • A trial-ending banner offers the offline pack download while cloud dictation still works, so the move to the Free plan is seamless when the trial ends.
    • A gentle hint appears when on-device dictations run noticeably slow on your machine, pointing at Pro's cloud engine as the faster option.
    Improved
    • llama-server cold starts: a liveness watchdog now respawns the local AI runtime in the background if macOS reaped it, instead of your next dictation paying the 2-10s startup cost.
    • Prewarming now also recovers from a crashed local AI server at hotkey-press, not just a missing one.
    • Long recordings transcribe faster: trailing silence is trimmed before it reaches Whisper.
    Fixed
    • Long dictations no longer end in a hallucinated "Thank you. Thank you. Thank you." loop — silent tails are trimmed from the audio, and any surviving repetition loop is collapsed from the transcript.
    • Microphone setting can be switched back to system default (carried from 1.0.2).
    Removed
    • The 2,500 words/week Free-plan cap is gone, along with the sidebar usage meter and the word-count bookkeeping behind it. Free is unlimited — it runs on your Mac, so there is nothing to meter.
    • Note for existing Free users: dictation now runs on-device, which needs a one-time offline pack download (~2.4 GB, Settings → Offline). Until it's downloaded, the app shows a banner pointing you there.
  3. July 3, 2026

    v1.0.2

    Dismissible error panel with auto-close, and the microphone picker can return to system default.

    Fixed
    • The overlay's error panel is now dismissible and auto-closes with a countdown, instead of sticking around until the next dictation.
    • Switching the microphone back to "System default" works again — the empty selection was previously ignored on save.
  4. June 27, 2026

    v1.0.1

    One-click offline setup and a paste hint that cleans up after itself.

    Added
    • The Offline page now has a single Download button covering all three on-device pieces (Whisper model, AI cleanup model, llama.cpp runtime) with one combined progress bar.
    Improved
    • The "copied — press ⌘V to paste" hint auto-dismisses with a countdown ring; hovering pauses the countdown.
  5. June 23, 2026

    v1.0.0

    Speechcap 1.0 — the first stable release, wrapping up the 0.1.x run of onboarding, teams, offline mode, and overlay polish.

    Added
    • Clicking the Dock icon now opens the main window, matching standard macOS behaviour.
    Improved
    • Stability milestone: everything shipped across 0.1.8-0.1.42 — team billing with per-seat management, in-app password reset, autostart reliability, fullscreen overlay handling, and dictation-drop fixes — is considered stable as of this release.
  6. June 10, 2026Critical

    v0.1.7

    Now runs on Intel Macs (universal binary), recognises spoken file extensions like "auth dot ts", and stops false-flagging accidental quick taps as in-progress transcriptions.

    Added
    • Universal binary — the same DMG now runs on both Apple Silicon (M1/M2/M3/M4) and Intel Macs. Previous releases were Apple-Silicon-only and refused to launch on Intel.
    • Spoken filename joining — "open auth dot ts", "App dot tsx", "Open auth. Ts" (Whisper sometimes capitalises after a period) all become the proper extension form: auth.ts, App.tsx. Driven by a curated whitelist (ts, tsx, py, json, md, css, html, json, png, mp4, …) so prose like "connect the dot to the line" is left untouched.
    • Snippets are now editable in place — click any snippet's row (or the pencil icon) to edit trigger and expansion. ⌘-Enter saves, Esc cancels.
    Fixed
    • A short accidental tap no longer trip-wires the "still transcribing" toast on the next press. Sub-second pipelines are now silently cancelled when you press again; only deliberate long recordings (≥1s of audio) preserve themselves and show the toast.
    • The sticky-nav Download button on the landing page now actually starts a download instead of scrolling to the pricing section.
  7. June 10, 2026

    v0.1.6

    Voice-triggered text snippets, dictation into Speechcap's own windows, live-updating history, and a busy-state safety net when you press the mic mid-transcription.

    Added
    • Snippets — define short voice triggers (e.g. "my email") that get replaced with longer text (your email address, signature, address) automatically on every dictation. Case-insensitive substring match, longest-trigger-first. Manage them under Settings → Snippets; edit, enable/disable, or delete inline.
    • Press the mic (or PTT) while a previous transcription is still finishing and you'll see a toast: "Still transcribing — press Esc to cancel and start over." Previously the second press silently started a new recording that raced the first one — both transcripts trampled each other.
    Improved
    • History page now updates in real time. New dictations appear at the top of the list immediately — no need to switch tabs and back.
    • Dictation now works when targeting Speechcap's own windows (e.g. testing a snippet by dictating into the Snippets page's expansion textarea, or replying in the in-app support form). Previously the inject silently dropped to the manual-paste fallback.
    • Toast appears when the pipeline is busy, and clears the instant the transcription completes — the busy state no longer lingers after a no-op dictation.
  8. June 10, 2026Critical

    v0.1.5

    Click-to-mic dictation now lands in your text field, the overlay no longer steals your caret, and Escape cancels in-flight dictations from anywhere.

    Fixed
    • Clicking the overlay's mic button now reliably injects the transcription into your source app — previously the paste landed on the empty overlay window and dropped silently to a manual-copy fallback.
    • The source app's text caret stays put when you hover or click the overlay. Notes, Mail, code editors — wherever your cursor was, it's still there after dictation.
    • Pressing Escape during recording or transcription cancels the dictation regardless of which app has focus. The old behaviour only worked while the overlay was the key window, which it almost never was.
    • The mic button no longer wedges after a manual-paste fallback — you can start a fresh dictation right away without dismissing the prior result first.
    Improved
    • Hovering the overlay shows the pointer cursor and native button tooltips immediately, without needing a prior click to "wake" the panel.
    • The app icon is recentred — the orange waveform sits on the vertical midline of the rounded square in every platform variant.
    • Removed the redundant page headings (the green pip + h2 above each settings page); the sidebar already names the current page.
    • Removed the third "collapse" button from the expanded pill row to reduce noise — hovering the pill expands it, hover-out collapses it.