Two on-device models. Your words, your Mac, your call.
Echo Flow AI runs entirely on your Mac after a one-time download. No cloud, no account, no data leaves your machine. Choose the tier that matches your hardware.
Echo Flow AI Pro
Higher-quality polishing with better reasoning, punctuation, and entity preservation. Best for technical writing, long dictation sessions, and nuanced tone adjustments.
- check_circle32K token context — handles long dictation without losing the thread
- check_circleStrongest entity preservation — code identifiers, URLs, names, numbers stay intact
- check_circleBest for complex rewriting — the rewrite palette's nine actions all benefit from Pro's reasoning
Echo Flow AI Lite
Lighter model that still beats the rule-based standard on most inputs. Best for everyday dictation on 8–15 GB Apple Silicon Macs where Pro would exceed available memory.
- check_circle32K token context — same wide context as Pro
- check_circleFast first-token latency — snappier feel for short utterances
- check_circleFits 8 GB unified memory — the recommended tier for base M1/M2 Macs
Echo Flow picks the right tier for your Mac at install time. You can switch in Settings → Echo Flow AI at any time.
How Echo Flow AI compares
Echo Flow can use four polishing strategies. It picks the best one available on your Mac. You can override the choice in Settings.
| Strategy | RAM | Where it runs | Best for |
|---|---|---|---|
| Apple Intelligence | Apple Silicon, macOS 26+ | On-device (Apple framework) | macOS 26+ users who want zero setup |
| Echo Flow AI Pro | 16 GB+ | On-device (MLX) | Best polishing quality, technical writing |
| Echo Flow AI Lite | 8–15 GB | On-device (MLX) | Everyday dictation on base M1/M2 Macs |
| Standard (rule-based) | Any | On-device (no model) | Intel Macs, <8 GB, or when AI is unavailable |
Frequently asked questions
Where does the Echo Flow AI model download from?
Echo Flow AI models are hosted on Cloudflare R2 and download once during onboarding. The download is a one-time ~1.7–2.3 GB transfer. After that the model is loaded entirely from local disk and never re-downloaded.
Do my transcripts leave my Mac?
No. Transcripts and audio are processed on-device. Echo Flow AI (both Pro and Lite) runs locally via MLX. Apple Intelligence, when used, also runs on-device. Nothing is uploaded to Echo Flow servers, and there are no Echo Flow servers to upload to.
Can I switch tiers later?
Yes. Open Settings → Echo Flow AI and pick Pro or Lite. Switching downloads the new model (~1.7–2.3 GB). The previously downloaded model is removed automatically.
What if my Mac has less than 8 GB RAM?
Echo Flow AI is disabled and polishing falls back to the Standard rule-based pipeline. The Standard pipeline is shipped with the app — no download needed, works on any Mac that runs Echo Flow.
How much disk space do the models need?
Echo Flow AI Pro needs about 2.3 GB of disk space; Echo Flow AI Lite needs about 1.7 GB. The download is one-time and the file stays on your Mac — it does not re-download each time you dictate.
Do you ship the model in the app?
No. Models are downloaded on demand to keep the DMG small (~9 MB vs. multi-GB) and to let you pick the tier that matches your hardware. The DMG is notarized and the download is over HTTPS.
Try Echo Flow on your Mac
14-day free trial. Rewrite palette stays free after trial. No account, no subscription for the AI model itself.
macOS 15+ · Apple Silicon recommended · 9 MB download