Install Arc Flow
Download the installer, run it, then open Arc Flow from Start.
Setup Guide
Set up Arc Flow in a few clear steps.
Start privately on your device, or choose a cloud provider when performance matters more than local processing. This guide helps you choose, connect, and start dictating.
Step 1 of 4
01 First launch
Follow the onboarding in order.
It only asks for the choices Arc Flow needs. You can revisit every one later in Settings.
Choose where processing happens
Start with on-device. Choose cloud only when your device needs the extra performance.
Set your microphone and shortcut
Confirm your mic, then keep the default shortcut or choose one you will remember.
Try it anywhere
Click into a text field, press the shortcut, speak, then press it again to finish.
When setup is complete, place your cursor in any app and press Ctrl + Space to start and stop dictation.
02 Choose a local model
Start local when privacy is the priority.
On-device processing keeps your voice on your PC and works offline after the model download. Arc Flow starts with Whisper by default and checks your hardware to recommend the best size.
Parakeet
Faster, with automatic language detection.
Parakeet TDT 0.6B v3 recognizes English and 25 European languages automatically. Choose it for fast, low-latency dictation when those are the languages you use.
Whisper
The default for broad language coverage.
Whisper supports 99 languages and lets you choose your dictation language. It is the better fit for Hindi, Hinglish, non-European languages, or a language you set manually.
Simple rule
Keep the default Whisper model unless English or the 25 supported European languages are your primary use, then try Parakeet for lower latency. Within Whisper, accept Arc Flow’s recommended size first.
Whisper models in Arc Flow
Pick the smallest modelthat gives you the accuracy you need.
Tiny74 MB
Fast notes on CPU-only PCs
Base141 MB
More accuracy on CPU-only PCs
Small465 MB
Default. Balanced speed and accuracy
Medium1.4 GB
Non-Latin scripts and longer audio
Large v32.9 GB
Best accuracy on a capable discrete GPU
Large v3 Turbo1.5 GB
Near Large-v3 accuracy, faster on NVIDIA RTX
Local AI models in Arc Flow
Choose the model that fitsyour PC and the writing help you need.
These models power text cleanup and Command Mode on your device. Larger models usually need more RAM but give stronger, more natural results.
Qwen3 4B2.3 GB download8 GB+ RAM to run
Best all-round choice for multilingual cleanup, grammar, and natural formatting
Gemma 4 E4B5.0 GB download8 GB+ RAM to run
Natural writing and strong multilingual grammar support
Qwen3 1.7B1.7 GB download6 GB+ RAM to run
Fast multilingual cleanup for everyday use
Phi-4 Mini2.3 GB download6 GB+ RAM to run
Fast grammar and punctuation help for English and simple edits
Gemma 3 1B0.8 GB download4 GB+ RAM to run
Reliable formatting and grammar on lower-end devices
Llama 3.2 1B0.8 GB download4 GB+ RAM to run
Lightweight basic text cleanup with limited multilingual support
Qwen3 0.6B0.6 GB download2 GB+ RAM to run
Ultra-light multilingual cleanup and basic edits
Listening locallyVoice → text
Ready to send
“Send this to the team tomorrow.”
Cleaned up on your PC. Nothing leaves your device.Want to use Ollama? Arc Flow can also use a model you already have installed in Ollama.
03 Match speech and cleanup
Your cleanup model should understand the language you dictate.
Use this quick check before pairing speech recognition with AI Mode or Command Mode.
Filter by role
Model & docsRoleLanguage capabilityBest use
04 Use cloud when it fits
Cloud is for performance-first setups.
Cloud processing sends the audio or text you choose to a provider. It is useful when a local model is not a good fit for your hardware. It is not the private, offline option.
05 Available cloud providers
Use the provider and model that already works for you.
These are the exact cloud choices currently shown by Arc Flow. API keys stay on your device and are sent only to the provider you choose.
Speech recognition providers
Voice to text
Deepgram
Nova-3, Nova-2
ElevenLabs
Scribe v2, Scribe v1
Mistral
Voxtral Mini
OpenAI
GPT-4o mini Transcribe, GPT-4o Transcribe, Whisper
Groq
Whisper Large v3 Turbo, Whisper Large v3
Custom
Your OpenAI-compatible model ID
AI cleanup providers
Punctuation, formatting and Command Mode
OpenAI
GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5.4 nano, GPT-5.2, GPT-5.1, GPT-5, GPT-5 mini
Anthropic
Claude Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 5, Sonnet 4.6, Haiku 4.5
Google
Gemini 3.5 Flash, Gemini 3.1 Flash Lite, Gemini 3 Flash, Gemini Flash (latest)
Groq
Llama 3.3 70B Versatile, Llama 3.1 8B Instant, GPT-OSS 120B, GPT-OSS 20B
DeepSeek
DeepSeek V4 Flash, DeepSeek V4 Pro
Custom
Your OpenAI-compatible model ID
06 A no-cost cloud starting point
Try Groq if you want to get started without a paid provider.
Groq is one optional provider in Arc Flow, not a partner. It is a straightforward choice for cloud speech recognition and AI cleanup. Free-plan limits depend on the selected model and can change.
How to get a Groq API key
- 1
Open Groq Console, then log in or create an account.
- 2
Open API Keys. You can also reach it from the account controls in the console.
- 3
Select Create API Key, give it a clear name such as “Arc Flow,” and choose no expiration if you want this installation to keep using it.
- 4
Copy the key when Groq shows it. Save it somewhere secure, treat it like a password, and never share it.
Groq API Keys
Key name
Arc Flow
Expiration
No expiration
gsk_••••••••••••••••••••Limits change and apply to your Groq organization. Check Groq’s live rate limits before relying on a quota.
07 Connect your chosen provider
Choose it during setup, or change it later.
You do not need to get every choice right the first time. Arc Flow lets you return to Speech to Text and AI Mode whenever your needs change.
Phase 01
Start speaking in minutes.
Three quick choices. You can change every one later.
- Allow microphone access
Accept the microphone permission so Arc Flow can turn your voice into text.
- Pick how you want to dictate
- On device: private and offline. Whisper is ready by default; choose Parakeet for fast English or supported European languages.
- Cloud: choose a provider and model, then paste its API key.
- Choose AI help—or skip it
- On device keeps cleanup on your PC.
- Cloud AI uses your chosen provider and API key.
- Skip for now only if you need plain dictation. Without an AI model, Arc Flow can type your speech but cannot clean up text or run Command Mode.
Phase 02
Fine-tune it when you’re ready.
Open Settings any time. Your current setup stays in place while you explore.
AI Mode and Command Mode
Settings → AI Mode → Processing mode
- Cloud: select a provider and model, paste the API key, then test it.
- On device: use Direct to download Arc Flow models, or Ollama for models you already have. Pick one active model.
Speech to Text
Settings → Speech to Text
- Local: download Whisper or Parakeet, then switch models whenever you want. Whisper can use auto-detect or a language you choose.
- Cloud: select a provider and model, paste and test its API key. The endpoint is filled in for you.
You are ready.
Download Arc FlowOpen any app, put your cursor where you want text, press Ctrl + Space, speak naturally, then press it again.