ChatterKey v4.5.0 — One model. One request.

ChatterKey v4.5.0 — One model. One request.

Released September 8, 2026 · macOS 14+ · Apple Silicon download · Build 8

ChatterKey now uses one configurable audio-capable model for every dictation and voice-edit attempt, defaulting to Gemini 3.5 Flash-Lite through OpenRouter. Speak, release, and receive the final text without a separate transcription or polishing stage.

What's changed

• A single audio request in all eight writing modes: same-language cleanup, English translation, professional, casual, concise, bullets, technical, and verbatim.
• Magic Voice Edit in one request: selected text and the spoken instruction audio are sent together; the model returns the replacement text directly.
• No hidden follow-up requests: removed separate Whisper/GPT transcription, second-stage polishing, English-compliance repair, automatic network retries, and automatic provider fallback routing. Retry is an explicit user action.
• One model field in Settings: choose another audio-input/text-output model in the future without adding another processing stage. Update its cost rates when switching.
• OpenRouter-only active connection: one fixed official endpoint and one provider key. Legacy provider identities remain readable in historical usage records.
• Updated onboarding and documentation: the README, website, architecture overview, privacy/security disclosures, and downloads now describe the single-model workflow.

Fixes and safeguards

• Translation and other writing modes still apply when optional smart cleanup is disabled.
• Added a clipboard-copy fallback for selection detection when standard Accessibility attributes are unavailable. This improves compatibility with browsers, Electron apps, PDFs, Notes, and editors; support still depends on the target app.
• Synthetic selection-copy events are ignored by the push-to-talk event handler.
• Empty, blocked, and truncated model responses are rejected instead of inserting incomplete output.
• Verbatim output no longer expands local snippet cues or spoken formatting phrases.
• Voice-edit output bypasses local snippet and formatting transformations so the selected document is not inadvertently expanded.
• Processing requests are pinned to OpenRouter, HTTP redirects are rejected, and legacy-provider configurations cannot send credentials through the new client.

Costs and migration

• Cost estimates now include audio tokens + instruction/selected-text tokens + output tokens, not a separate per-minute transcription charge.
• The rough on-device preview is not counted as an additional text input; voice-edit selections are counted.
• Legacy settings and usage records migrate to the single-model format, preserving writing preferences, vocabulary, snippets, and local history preferences. Existing OpenRouter Gemini model selections are retained.
• Previous OpenAI/custom or text-only configurations move to the Gemini default and require an OpenRouter API key. Keys stay in their original Keychain accounts and are never copied between providers.
• Estimates are approximate, not invoices. Additional reasoning, failed requests, explicit retries, taxes, and platform fees are excluded. Review rates when changing models or when provider pricing changes.

Requirements and installation

1. macOS 14 or later on an Apple Silicon Mac for the downloadable binary.
2. Your own OpenRouter API key; Gemini 3.5 Flash-Lite is selected by default.
3. Microphone and Accessibility permissions. Speech Recognition remains optional for the on-device live preview, which does not add a cloud model request.
4. Quit the old app, extract ChatterKey-v4.5.0.zip, replace the app in Applications, and launch it. Review the migrated model and rates in Settings.

Distribution warning: This is an ad-hoc signed community-test build, not Developer ID signed or Apple-notarized. It has not been validated on a clean Mac account. Gatekeeper may block it. Do not disable system-wide security protections to install it. The ZIP is arm64, not a universal Intel/Apple Silicon binary.

The accompanying ChatterKey-v4.5.0.zip.sha256 verifies download integrity; a checksum is not Apple approval or a publisher identity guarantee. After downloading both files into the same directory:

shasum -a 256 -c ChatterKey-v4.5.0.zip.sha256

Validation and known limits

• Swift 6 model/endpoint regression harness covers all writing modes with cleanup on/off, request contents/counts, voice edits, migration, cost estimates, invalid responses, cancellation, and failure paths.
• Debug and release builds, public-source checks, signature verification, and release-archive checksum checks are part of this release's validation.
• Single-request behaviour is verified with mock transport tests; this release does not claim measured real-provider latency, transcription accuracy, or universal cross-app compatibility.
• Switching to another model requires audio-input and text-output support. A text-only model cannot replace Gemini in this pipeline.
• No automatic translation repair means an imperfect model result is returned without a hidden second call. Review important output before using it.

Short announcement

ChatterKey v4.5.0 is here: one model, one request. Gemini 3.5 Flash-Lite now handles dictation, translation, verbatim, and voice edits directly from audio. This release removes separate transcription/polishing stages and hidden retries, improves text-selection handling, simplifies Settings, and updates cost estimates to audio-plus-text token pricing. Your local preferences migrate, and provider keys remain isolated. Available as an Apple Silicon community-test build; not Apple-notarized.