Show HN: Free alternative to Wispr Flow, Superwhisper, and Monologue

Original Article ↗ Hacker News Discussion ↗

Cloud vs local models & “deep context”

  • FreeFlow uses Groq-hosted models for fast transcription and an LLM “deep context” step: capturing a screenshot of the active window and asking an LLM to infer names/terms and fix spelling.
  • Several commenters like the idea but see screenshotting + cloud LLM as overkill, suggesting using accessibility APIs to pull nearby text instead, processed by a small local LLM.
  • Others argue local-only pipelines (e.g., Parakeet, Whisper variants) now feel fast enough and avoid vendor lock‑in and privacy concerns.

Alternative tools mentioned

  • Strong enthusiasm for Handy (cross‑platform, Parakeet support, optional LLM post‑processing, context capture PR in progress), Hex (Mac‑only, CoreML/Neural Engine, very fast), MacWhisper, VoiceInk, Whispering, Whistle, MacWhisper, Murmure, OpenSuperWhisper, Jabber, Auriscribe, hyprwhspr, and others.
  • Multiple Android/iOS options surface: FUTO keyboard/voice, Whisper+, VoiceFlow, Utter, Talkie/ottex-based flows, Whisper Memos.
  • Some say SuperWhisper already offers a free Parakeet‑based local mode comparable to what FreeFlow targets.

Performance, latency, and hardware

  • One camp says local is “too slow” once you add post‑processing; others report sub‑second transcriptions on M1/M2 Macs or any GPU using Parakeet or Whisper large‑v3‑turbo.
  • Parakeet (especially v3) is repeatedly praised for speed, accuracy, and multilingual handling, even on mid‑range phones.
  • Concerns raised about future Groq pricing/free‑tier changes versus the stability of fully local solutions.

UX, workflows, and features

  • People care a lot about hotkeys (Caps Lock, Scroll Lock, right Option, Stream Deck buttons, foot pedals), push‑to‑talk vs toggle, and live/streaming transcription.
  • Some want preserved audio alongside transcripts; FreeFlow currently stores WAVs tied to history entries but auto‑deletes them (noted as easily changeable).
  • Post‑processing to transform raw transcripts into structured, editor‑ready text (for notes, books, coding agents) is seen as the next frontier.

DIY & meta discussion

  • Several describe rolling their own STT scripts around faster‑whisper, sox/arecord, and clipboard tools, valuing instant adoption of new models.
  • One commenter criticizes “free alternative to X” marketing as copycat; others respond that many devs simply reinvented tools they didn’t realize already existed.
  • A minority questions the appeal of speech‑to‑text at all; others point to accessibility, injury recovery, long-form dictation, and hands‑free coding as compelling uses.