The full list of questions the beta crowd actually asks — grouped, indexable, no fluff.
A Mac on Apple Silicon (M1 or newer) running macOS 14.4 or later, plus an API key — Mistral's free tier is the quickest start, or bring your own key for OpenAI, Anthropic, Groq, or any self-hosted OpenAI-compatible endpoint. The app is ~30 MB. You're typing translations within ~90 seconds of clicking Download.
The onboarding flow points you to Settings → API. Start with Mistral — French company, EU-hosted, GDPR-friendly, with a generous free tier on mistral-small-latest and roughly €0.20 per million input tokens once you outgrow it. Prefer another provider? Drop in your own key for OpenAI, Anthropic, Groq, or any OpenAI-compatible endpoint.
Yes — InFlow uses the macOS Accessibility API, so it works with the focused field of any Mac app: Mail, Messages, Slack, every browser (ChatGPT included), Notion, Discord, Figma's text fields, Telegram, WhatsApp Web, etc. No per-app integration, no SDK.
Voice transcription, yes — Parakeet runs fully on-device on the Apple Neural Engine. Translation needs a live network connection to your LLM provider (unless you're running a self-hosted model locally, in which case yes, fully offline).
Free during beta. A paid tier with hosted keys is planned for users who don't want to manage their own API key — join the refer queue for early access.
Pennies a day for normal use. Mistral Small runs around €0.20 per million input tokens — a heavy day of dictation and translation comes to a few cents. Bringing your own key? OpenAI's GPT-4o-mini and Anthropic's Claude Haiku are similarly cheap, and Groq's Llama 3.3 is free under their public rate limits.
Yes — 50 requests per day per IP. That's enough to feel out every flow without burning down our Mistral budget. The demo runs on Mistral end-to-end — Voxtral for transcription, Mistral Small for translation, both EU-hosted, audio discarded immediately. The Mac app has no cap because the API key is yours (and it transcribes on-device with Parakeet, no audio upload at all).
Translation: roughly 40 languages handled well by Mistral and the major LLMs — English, French, Spanish, German, Italian, Portuguese, Dutch, Polish, Romanian, Japanese, Korean, Chinese, Arabic, plus everything in between. On-device dictation (Parakeet): 14 languages with strong accuracy on European pairs.
Mistral was trained on a European-skewed corpus, so EU pairs (FR↔ES, IT↔PT, PL↔DE) come back noticeably better than from US-trained generalists. For Asian-language pairs, GPT-4o-mini and Claude Haiku tend to edge out — switch provider per profile in Settings.
Yes — every profile has a fully-editable prompt template. The defaults are good starting points; you can rewrite them entirely (e.g., "Translate to formal Japanese keigo, no preamble"), pin a custom source/target language, and bind your own hotkey.
Yes. Select any passage in any language, tap Summarize (📝) in the selection bubble, and InFlow writes a faithful summary in your own language — never a word-for-word translation. The result lands in the popup as clean Markdown; use the − / + buttons to step the length between Brief, Standard and Detailed on the spot, and set your preferred starting length in Settings.
Those are dictation tools — speech to text, end of story. InFlow's core is translation: selection translation, voice translation, reply composition. Dictation is just one of the modes. The real differentiators: in-place injection straight into the focused field, Reply and Guided Reply for composing responses, Polish for cleaning up text you've already written, Summarize for condensing anything into your language, EU-resident translation by default through Mistral, and fully on-device speech-to-text via Parakeet.
DeepL is a translation service with a web app and a Mac app that opens a window. InFlow stays in your flow — your selection or transcript becomes the translation, in place, in whatever app you were already typing in. No window switch, no copy-paste.
Read the full breakdown at privacy.html. Short version: on-device dictation, EU-resident LLM by default, no telemetry, no account, no logs of your text.
~30 MB, signed, notarized, no account.