Flunkey

Using Flunkey

Dictionary & Polish

Raw transcription gets you words; dictionary + polish gets you text you can send. Two layers, applied before every paste.

Custom dictionary

Add people, products, acronyms, and domain terms once — “Priya Natarajan”, “Supabase”, “HbA1c”, matter numbers — and Flunkey prefers your spelling over the model’s guess. Ships with Groq, Flunkey, and VS Code. Entries apply longest-first with fuzzy typo repair (Levenshtein, 0.82 confidence), and each word can be searched, edited, deleted, or toggled on/off individually.

Entry format
spoken "supabase"      → Supabase
spoken "doctor patel"   → Dr. Patel
spoken "h b a one c"    → HbA1c

Grammar polish

After dictionary fixes, Flunkey normalizes capitalization, sentence boundaries, and punctuation, drops filler (“um”, “uh”), and resolves spoken corrections (“Tuesday… actually Wednesday”). The goal: output that reads like you wrote it, not like you spoke it. Two Settings toggles — Dictionary F8 and Dictionary F10, both default ON — gate each flow separately while sharing the same engine.

Tips that work

  • Add the spoken form you actually say, not just the correct spelling
  • Prefer full names (“Priya Natarajan”) over first names for accuracy
  • Keep entries short — one term per line beats long phrases
  • Re-test after adding: dictate the tricky sentence once and confirm

Team glossaries (manual for now)

There is no shared cloud dictionary — by design, there is no cloud. Export your list (flunkey-dictionary-YYYY-MM-DD.json) and share the file; import merges on the other side. A one-click import is on the roadmap.