Understand meetings, calls, videos and system audio as they happen—with private local models, Apple Translate, or optional GPT and Gemini cloud providers.
Built for macOS 15 and later · Apple silicon
Features
Translate your microphone, your Mac’s system audio, or both in real time. Übersetz keeps the original words, translation and context together without taking over your screen.
Use your microphone, your Mac’s system audio, or both at once. Follow meetings, calls, videos, podcasts and live streams directly from the apps you already use—no meeting bot or special integration required. Übersetz is useful beyond translation, too: live transcription can make strong accents, regional dialects and fast or unclear speech much easier to follow.
Place an adaptive subtitle overlay above the app you are using. It grows to keep the current caption visible, returns to a compact height for shorter captions, and can be moved wherever it feels natural.
Add a name, URL or question and ask the Advisor for a concise explanation without leaving the conversation.
They fear time and distance may erase a love they once shared—even if the other person returns.
How it works
No complicated setup. Choose what you want to hear, decide where translation should run, and follow the conversation in the format that suits you.
Use your microphone, your Mac’s system audio, or both at the same time.
Choose GPT Realtime, Gemini Live Preview, Apple Translate, Local models, or a compatible endpoint you control.
Read in the main window, switch to the overlay, and save the finished transcript only when you choose.
Privacy
The honest privacy boundary is simple: built-in local models process on your Mac. Cloud modes send data only to the provider you explicitly select.
In local mode: no processing server. Übersetz stores no raw-audio archive. No screen capture.
The app processes microphone and system audio live and does not write raw audio to disk. A selected cloud provider may apply its own retention terms.
System audio uses a macOS audio tap. Übersetz never receives, saves or transmits screen pixels.
After a built-in model is installed, transcription, translation and Advisor processing run on your Mac.
GPT Realtime, Gemini Live and custom remote endpoints are optional. If you choose one, that provider receives the data needed to translate or advise.
Übersetz retains the current transcript locally so it can restore the session after relaunch. Clear removes it; Save creates the Markdown export you choose.
Übersetz itself is free and does not monetize your attention. It will stay that way.
Models & providers
No single model is right for every conversation. Übersetz lets you use OpenAI or Google realtime translation, keep the workflow on your Mac, use Apple’s native translation framework, or connect an endpoint you control. Each option has a different balance of speed, privacy, memory use and setup.
GPT Realtime Translate provides continuous live transcription and translation through OpenAI. It requires an internet connection and your own OpenAI API key, which Übersetz stores in macOS Keychain.
Audio and session data required for translation are sent to OpenAI. OpenAI’s pricing, retention policies and account terms apply; API usage is not included with the free Übersetz download.
Gemini Live Translate Preview provides direct live transcription and translation through Google's Gemini API. It requires an internet connection, a supported target language, provider disclosure acceptance and your own Google Gemini API key.
The Google key is stored separately from the OpenAI key in macOS Keychain. Audio and relevant Advisor text are sent to Google only while Gemini is selected. Google pricing, quota, retention, regional availability, project access and Preview-model changes apply.
Übersetz bundles Whisper for speech recognition and an embedded llama.cpp runtime for four managed Qwen choices. You download a managed model separately; Ollama is not required for these four choices.
If Ollama is running locally, Übersetz also discovers its installed models. Equivalent managed and Ollama copies appear once per family, with the installed Ollama copy preferred automatically. Add local model opens a curated Ollama catalogue plus an advanced tag field.
Apple Translate combines Übersetz’s bundled Whisper language detection with Apple’s on-device Translation framework and installed language packs.
Low Latency is used by default. Übersetz can prefer Apple High Fidelity when the required Apple Intelligence model is available, with Apple controlling fallback behavior. Translation languages are managed through macOS System Settings.
Connect an Ollama server or an OpenAI-compatible API and choose the model identifier yourself. The endpoint may run on your Mac, elsewhere on your local network, or through a remote provider.
Übersetz supports unauthenticated endpoints, bearer tokens or API keys, and Cloudflare Access service tokens. Credentials are stored in macOS Keychain. Balanced and Low Latency translation cadences are available.
Frequently asked questions
Übersetz helps you understand every word in real time, whether the sound comes from a meeting, a call, a video, your microphone or your Mac’s system audio. Here is what to know about cost, models, privacy, storage and setup.
Übersetz is free to download and use. There are no subscriptions, advertisements or tracking inside the app. Built-in local models can run on your Mac without a paid Übersetz account. If you choose GPT Realtime, Gemini Live or another remote provider, that provider may charge for API usage under its own pricing. Donations are optional and help fund development, testing and future model support.
The disk image is larger than a typical utility because it includes the macOS app plus on-device speech-recognition and speaker-processing assets. Those components let Übersetz hear and prepare live audio locally instead of making every session depend on an Übersetz server. The installer does not contain every Qwen translation model. Local translation models are separate downloads, so you install only the model that suits your Mac and available memory.
Übersetz supports GPT Realtime, Gemini Live Translate Preview, Apple Translate on compatible macOS versions, four managed Qwen downloads, compatible models installed in local Ollama, and custom Ollama or OpenAI-compatible endpoints. Gemini requires its own Google API key and remains subject to Google's Preview availability, quota and regional access. Each provider has different privacy, connectivity, storage and memory requirements.
Compare models and providers ↑Yes. GPT Realtime Translate is optional and requires an OpenAI API key that belongs to you. Übersetz does not include shared OpenAI access or bundle API usage with the free download. You create and manage the key in your own OpenAI developer account, then add it in the app’s settings. OpenAI usage and billing are governed by your OpenAI account.
Create or manage an OpenAI API key ↗Gemini Live Translate is an opt-in Preview provider. It requires your own Google Gemini API key, internet access, a supported target language and acceptance of the in-app provider disclosure. The Google key is stored separately from the OpenAI key in macOS Keychain. Google may require project access or billing, may impose quota, and may change Preview model availability. Übersetz does not silently switch to GPT or another provider if Gemini fails.
Create or manage a Google Gemini API key ↗Yes, when you select a local model or compatible Apple Translate mode. Local-model transcription, translation and Advisor processing run on your Mac after the required model is installed. Übersetz does not operate a processing server for that workflow and does not create a raw-audio archive. If you explicitly select GPT Realtime, Gemini Live or a custom remote endpoint, the chosen provider receives the data required to perform that service and its own terms apply.
No screen pixels are captured. System audio is received through a macOS audio tap, so Übersetz can hear supported audio without reading or transmitting the screen. The app does not save raw microphone or system audio. It does retain the current transcript and Advisor state locally in Application Support so the session can be restored after relaunch. Clear removes that retained session; Save creates a separate Markdown export at the destination you choose.
The current release requires macOS 15 or later and an Apple-silicon Mac. Local model requirements are additional: a small Qwen model can work on a Mac with less memory, while the largest built-in option is intended for machines with substantially more unified memory. Übersetz checks model availability before use, and local model files consume extra disk space after you download them.
The app download includes the core speech assets, but translation models are installed separately. You initiate a managed Qwen download or choose Add local model to pull a catalogue or advanced tag through a running local Ollama service. This avoids bundling several gigabytes of models in the installer. Managed Qwen downloads range from roughly 1 GB to more than 16 GB; catalogue sizes vary and Ollama manages those files.
Übersetz is designed for multilingual speech and lets you enable the languages relevant to you. Coverage and quality depend on the selected transcription and translation model, and not every language performs equally well with every provider. GPT Realtime, Gemini Live, Apple Translate, local models and custom endpoints can have different strengths and target-language lists. For important conversations, test the exact source, target and model combination you plan to use.
Yes. Übersetz can listen to your microphone, your Mac’s system audio, or both, so it can follow supported meetings, calls, videos and live streams without a dedicated integration for each app. You can read the original speech and translation in the main window or keep a compact subtitle overlay close to the conversation. Audio availability can still depend on macOS permissions and the source application.
The initial app download and any model download require internet access. After the required assets are installed, local workflows can process a conversation on your Mac without an Übersetz processing server; a selected Ollama model also requires the local Ollama service. GPT Realtime, Gemini Live and remote custom endpoints require a network connection to their provider. Apple language assets may need to be downloaded by macOS before offline use.
Flutter provides a shared foundation for the interface, translation workflow and product logic while still allowing Übersetz to use native macOS capabilities where they matter, including system-audio and on-device processing integrations. The current release is macOS-only, but web and mobile versions are planned alongside future Windows and Linux editions. The goal is to bring the same focused translation experience to all major platforms without rebuilding the entire product independently for each one. These versions are on the roadmap and are not available yet; their scope and release timing may change as the macOS app evolves.
Free for macOS 15+
Übersetz itself is free. Optional cloud providers may charge separately. Built-in local models download once and then run on your Mac.
Support Übersetz
Your donation supports the work required to improve translation quality, add more languages and models, and polish the macOS experience. It also helps keep Übersetz forever free—with no ads, tracking, or subscriptions.
More reliable speaker turns and audio recovery
