OpenAI / ChatGPT

About

OpenAI / ChatGPT is TypeWhisper's bundled cloud provider for speech-to-text, prompt processing, and text-to-speech. Version 1.3.0 added the context-aware gpt-transcribe and gpt-live-transcribe models shown above, plus account-specific ChatGPT/Codex model discovery. Version 1.3.1 keeps incremental previews for file-based transcription and reliably finalizes the complete buffered recording when dictation stops.

The published release is the source of truth for the public version. The plugin release workflow can patch the bundle manifest during release, so a local checkout may show an older manifest version until it is synced.

Sources: OpenAIPlugin v1.3.1 release, OpenAIPlugin v1.3.0 release, TypeWhisper plugin registry, OpenAI plugin source

What the terms mean

This add-on can take three separate roles in TypeWhisper. You can use OpenAI for all three or combine it with other providers. For example, a local transcription engine can create the raw text while OpenAI only handles the optional workflow afterward.

Microphone or audio file → Transcription Engine → Raw text → Optional LLM Workflow → Text inserted into the target app

Text → Text-to-Speech Provider → Spoken output

Term	Meaning	What you can do with OpenAI / ChatGPT
Add-on / integration	The complete OpenAI / ChatGPT package. One add-on can expose several capabilities to TypeWhisper.	Configure the credentials and models once, then use the same integration for transcription, LLM workflows, and text-to-speech.
Transcription Engine	The speech-recognition service that turns microphone or file audio into text. It runs before any optional text-rewriting workflow.	Dictate into any app, transcribe recordings, use GPT Transcribe for complete buffered results, or GPT Live Transcribe for realtime partial text. Select OpenAI globally, in a profile, or as an override in a dictation workflow.
Transcription Model	The specific speech model used by the engine. Models differ in realtime behavior, context support, and translation support.	Choose GPT Transcribe for context-aware file transcription, GPT Live Transcribe for realtime output, or Whisper 1 when the transcription task must translate speech into English.
Transcription Context	Background information for speech recognition, not an instruction to rewrite the result. It helps the model interpret names and ambiguous words.	Describe the topic or setting, such as a medical meeting or software demo. TypeWhisper adds selected language hints and dictionary terms separately and automatically.
LLM Provider / Workflow Provider	The language model that processes text after transcription, or processes selected text/clipboard content without recording.	Clean up grammar, rewrite text as a polite email, create bullet points, summarize notes, translate text, or follow a custom instruction. A workflow can inherit the global LLM fallback list or select OpenAI explicitly.
LLM Model	The OpenAI or ChatGPT/Codex model that follows the workflow instruction. It does not control speech recognition.	Use a larger model for complex restructuring and a smaller model when speed matters. Reasoning Effort changes how much thinking time supported models use.
Text-to-Speech Provider (TTS)	The service that turns existing text back into audio. It is independent of transcription.	Read text aloud, use spoken feedback or manual readback, choose a voice, and optionally tell the voice how to speak.
Connection Method	The credentials used for a capability.	An API Key enables transcription, LLM processing, and TTS. ChatGPT Login enables prompt/workflow processing through a ChatGPT/Codex account, but it does not unlock transcription or TTS.

Typical ways to use it

• Direct dictation: OpenAI is the transcription engine; the recognized text is inserted without an LLM rewrite.
• Polished email: OpenAI transcribes the recording, then an LLM workflow rewrites the raw text as a clear email before insertion.
• Realtime preview: GPT Live Transcribe returns partial text while you are still speaking.
• Meeting or file transcription: GPT Transcribe processes the recording with topic context, language hints, and dictionary terms.
• Text-only workflow: Keep a local transcription engine and use ChatGPT Login only for cleanup, summaries, translations, or selected-text processing.
• Readback: Use OpenAI TTS to hear text with the selected voice and optional speaking instructions.

See the Workflows guide for app-, website-, hotkey-, and fallback-based automation.

Connection Method

Field	What it controls	Example
Connection Method	Chooses how the LLM provider authenticates. API Key uses the OpenAI API directly. ChatGPT Login uses a ChatGPT/Codex login for prompt processing.	Use API Key when you want transcription, TTS, and the full API model catalog. Use ChatGPT Login when you want to process prompts through your existing ChatGPT/Codex account.
API Key	Stores an OpenAI API key in the macOS Keychain and validates it before enabling cloud transcription, TTS, and API-backed prompt processing.	Paste an sk-... key, save it, then wait for Valid API Key.
Show/Hide API Key	Reveals or masks the stored key while editing.	Use the eye button to inspect a pasted key before saving.
Remove	Clears the stored API key or ChatGPT login tokens.	Remove a key before switching OpenAI projects or handing the Mac to another user.
Sign In In Browser	Starts the ChatGPT browser login flow on macOS. TypeWhisper completes the local callback automatically.	Click it, finish OpenAI login in the browser, return to TypeWhisper.
Import Codex Login	Imports an existing Codex login from ~/.codex/auth.json on the same Mac.	Use it when Codex is already signed in and you do not want to repeat browser auth.
Connected plan	Shows the ChatGPT plan type decoded from the login token when OpenAI provides it.	Connected plan: pro or Connected plan: prolite.

Sources: OpenAI API documentation, OpenAI plugin source

Transcription

API Key mode enables OpenAI speech-to-text. ChatGPT Login does not replace the API key for transcription.

Model	ID	Notes
GPT Transcribe	gpt-transcribe	Context-aware file transcription and the default for new configurations. During dictation, TypeWhisper shows incremental previews and then finalizes the complete buffered recording. Whisper Translate is not available.
GPT Live Transcribe	gpt-live-transcribe	Context-aware realtime transcription with partial updates and a configurable live delay. Whisper Translate is not available.
Whisper 1	whisper-1	Classic Whisper transcription. Supports Whisper Translate to English.
GPT-4o Transcribe	gpt-4o-transcribe	Higher-accuracy GPT-4o transcription. Whisper Translate is not available for GPT-4o transcription models.
GPT-4o Mini Transcribe	gpt-4o-mini-transcribe	Smaller GPT-4o transcription model for faster or lower-cost transcription. Whisper Translate is not available.
GPT Realtime Whisper	gpt-realtime-whisper	Streams 24 kHz mono PCM through OpenAI's Realtime API for live transcription. Whisper Translate is not available.

Field	What it controls	Example
Transcription Model	Selects which OpenAI model TypeWhisper sends recorded or live audio to. Existing saved model choices are preserved when upgrading.	Pick GPT Transcribe for context-aware file transcription, GPT Live Transcribe for realtime updates, or Whisper 1 when you need translate-to-English behavior.
Transcription Context	Describes the recording topic, setting, or other relevant background for GPT Transcribe and GPT Live Transcribe.	Weekly product meeting about TypeWhisper's macOS release.
Language hints	TypeWhisper automatically forwards the selected profile/request languages to the two context-aware GPT transcription models.	A German profile supplies de; multilingual hints stay ordered and duplicates are removed.
Dictionary terms	TypeWhisper automatically forwards sanitized dictionary entries to the context-aware GPT transcription models.	Add product names such as TypeWhisper or WhisperKit to the dictionary.
Live Transcription Delay	Appears for GPT Live Transcribe and controls how quickly partial text is returned. Lower levels react sooner; higher levels give the model more audio context.	Start with Low, then increase it when you prefer more context over the earliest possible partial text.

On a fresh configuration, TypeWhisper selects GPT Transcribe. Upgrades keep a previously saved transcription model. Translation requests are blocked locally for every model except Whisper 1.

Sources: GPT Transcribe, OpenAI Realtime transcription, OpenAIPlugin v1.3.1 release, OpenAI plugin source

Text-to-Speech

OpenAI / ChatGPT also provides TypeWhisper's text-to-speech provider. It uses an OpenAI API key, even when ChatGPT Login is selected for prompts.

Field	What it controls	Example
Text-to-Speech Voice	Selects one of OpenAI's built-in voices. TypeWhisper defaults to Marin.	Choose Marin for the default voice or Cedar for another high-quality option.
Voice instructions	Sends extra speaking instructions to the gpt-4o-mini-tts request when the field is not empty.	Speak calmly, with short pauses between sentences.

Supported voices in the plugin are alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer, verse, marin, and cedar. TypeWhisper requests PCM audio so playback can start without an additional file conversion step.

Sources: OpenAI Text to speech, OpenAI plugin source

LLM Models

The LLM section is available when the selected connection method has credentials. API Key mode can fetch the OpenAI API model catalog. ChatGPT Login mode fetches the account-specific ChatGPT/Codex catalog, caches it per account, and preserves the selected model while it remains available.

Field	What it controls	Example
LLM Model	Chooses the model used for prompt processing and workflows.	Use gpt-5.5 for high-quality prompt processing, or a smaller model when latency matters.
Refresh	In API Key mode, calls OpenAI's /v1/models endpoint and keeps current chat-capable models. In ChatGPT Login mode, requests the visible Codex model catalog for the connected account. If that request fails, TypeWhisper keeps the previous cache or its offline fallback.	Click Refresh after your OpenAI project or ChatGPT/Codex account gains access to a new model.
Reasoning Effort	Appears for GPT-5, o-series, and Codex-style models. It controls how much thinking time the model spends before answering.	Use Low for quick cleanup prompts, High or X High for harder rewrite or coding workflows.
Temperature Mode	API Key mode only. Provider Default uses TypeWhisper's default temperature behavior. Custom enables a 0.0-2.0 slider when the selected model supports custom temperature.	Use 0.2 for deterministic cleanup; use a higher value for more varied phrasing.
Custom temperature ignored notice	Shown when a selected GPT-5 reasoning configuration does not accept a custom temperature.	If you choose GPT-5 with reasoning effort, keep Provider Default or ignore the custom slider.

Fallback API Key models include gpt-5.5, gpt-4.1-nano, gpt-4.1-mini, gpt-4.1, gpt-4o, gpt-4o-mini, and o4-mini. The offline ChatGPT/Codex fallback includes current options such as gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5, gpt-5.4, and gpt-5.4-mini; a successful refresh replaces that fallback with the models visible to the connected account.

Sources: OpenAI Models API, OpenAI Responses API, OpenAI plugin source

Setup Examples

API key for transcription, TTS, and prompts

1. Open TypeWhisper Settings > Plugins.
2. Configure OpenAI / ChatGPT.
3. Select API Key.
4. Paste your OpenAI API key and wait for Valid API Key.
5. Choose a Transcription Model, Text-to-Speech Voice, and LLM Model.
6. Click Refresh if you want TypeWhisper to fetch the model list available to your OpenAI project.

ChatGPT or Codex login for prompts

1. Open TypeWhisper Settings > Plugins.
2. Configure OpenAI / ChatGPT.
3. Select ChatGPT Login.
4. Click Sign In In Browser or Import Codex Login.
5. Choose a ChatGPT/Codex LLM Model and Reasoning Effort.
6. Add an OpenAI API key as well if you also want transcription or text-to-speech.

Notes

• ChatGPT Login is documented here for macOS.
• Transcription and text-to-speech require an OpenAI API key.
• Browser login and Codex import are for prompt processing and do not replace API authentication for STT/TTS.
• The public 1.3.1 release requires TypeWhisper host version 1.5.0 or newer.