Sayful 0.9.7.6

Fixed

• Screenshot text extraction for personal accounts now stays on the vision-capable Gemini route instead of failing through a text-only Groq model.
• Large text, image, speech, activity, and achievement requests no longer fail before authentication when their body exceeds the edge runtime read buffer.

The app is Developer ID signed, Apple-notarized, and distributed through Sparkle.

Sayful 0.9.7.5

Added

• The dictation island's translate / fix / read-aloud busy state now shows the same animated orb as transcribing, instead of the plain spinner.

Changed

• The DMG installer window now matches the app's monochrome icon and dictation-island orbs instead of the previous teal accent.

Fixed

• Cancelling a dictation several times in a row no longer pushes the "Transcript cancelled" toast's lifetime bar outside the pill's left edge.

Sayful 0.9.7.4

Added

• The dictation island's "Transcribing…" indicator now shows an animated dotted orb (ported from the thinking-orbs component) instead of the plain spinner, cycling through five distinct animation styles.

Sayful 0.9.7.3

What’s new

• Nested ToDo items and dividers remain valid Markdown through editing, regeneration, and export.
• Equal nested task labels can be completed independently with stable identities.
• Switching between personal and company accounts remaps API requests automatically without restarting Sayful.
• In-flight requests keep tenant URLs, public keys, and session tokens consistent.

Sayful 0.9.7.2

Reliable transcript polishing

• Corporate dictation polishing now uses Groq Qwen 3.6 27B by default, with an independent OpenRouter fallback.
• Public polishing also falls back to Qwen 3.6 27B after it led the multilingual quality and latency benchmark.
• Provider timeouts and failover prevent long dictations from waiting a minute before returning unformatted text.
• Separate privacy-safe STT and polishing timing logs make infrastructure latency diagnosable.

Sayful 0.9.7.1

Sign in for teams

• Work email sign-in: organizations can run Sayful on their own infrastructure — choose "I have a work email" at sign-in and corporate accounts route to your company's Sayful backend automatically.
• One-click re-login: Sayful remembers your account — signing back in is a single "Continue as …" click, with "Use a different account" to switch.
• Personal sign-in is unchanged: one click of "Continue with Google".
• Polished, branded installer window.

Sayful 0.9.7

Sign in for teams

• Work email sign-in: organizations can now run Sayful on their own infrastructure — choose "I have a work email" at sign-in and corporate accounts are routed to your company's Sayful backend automatically.
• One-click re-login: Sayful remembers your account, so signing back in is a single "Continue as …" click — with "Use a different account" to switch.
• Personal sign-in is unchanged: one click of "Continue with Google".
• More detailed sign-in diagnostics in the app log for faster support.

Sayful 0.9.6.6

Complete dictation and STT diagnostics

• Fast and Turbo dictation no longer sends a language prompt that could omit spoken spans.
• DevTools can preserve and select recordings for repeatable Fast, Quality, and Turbo A/B tests.
• Re-transcription tests have independent language-instruction, Whisper-filtering, and volume-normalization controls.
• Turning off the DevTools language instruction now guarantees the prompt is omitted.

Sayful 0.9.6.5

Progressive letter-by-letter typing lessons with real-word curricula, a full finger-zone keyboard, typing achievements, and layout-independent scoring. Typing languages now follow the keyboard layouts installed in macOS. Dictation History gains Fast, Quality, and Turbo re-transcription actions plus smoothly expandable transcripts and first-click hover actions. Dictionary exceptions now preview their mapped text.

Sayful 0.9.6.4

Dictation reliability

• Quiet recordings are now peak-normalized before transcription — fixes cases where a fully-voiced dictation came back as a couple of words or a stock phrase on low-level microphones.
• Transcripts that are impossibly short for the recording length now silently retry through the Quality model (DevTools-gated).
• Audio levels (peak/RMS/applied gain) are logged per dictation for diagnostics.
• Auto-update feed moved to sayful.app/appcast.xml.