PHORMINXPHORMINX FOR WINDOWS · v0.2.5
Dictate into any app, follow a meeting as it is transcribed and translated, and ask what was decided while the call is still going.
DICTATION
Press Ctrl+Alt+Space, speak, and Phorminx types into the app you were using. Recognition runs on your PC: Instant mode uses Vosk for speed, Accurate mode uses Whisper with optional Vulkan GPU acceleration. Live text shows what it heard while you talk.
REVIEW PANEL
Open the last dictation from the status pill to edit it, reformat it with another strength or a one-off instruction, copy it, or replace the text you inserted. Replace only touches text that is still there and unchanged; otherwise the result goes to the clipboard.
FORMATTING
A local Ollama model turns rambling speech into clean text, from light punctuation to structured lists, and can run more than once to refine the result. Teach names and jargon in the lexicon, and give each app its own style with profiles. Formatting is optional.
PROMPTS
Keep named, tagged instructions in one library. Type @ in any instruction field to use one: formatting, app profiles, the review panel, the meeting assistant or Actions. Every save is a version you can restore, and references always use the latest one.
MEETINGS
Phorminx captures your microphone and your computer's audio together and writes the meeting down as it happens. Turn on translation and a local model adds each line in your language right under the original, which is always kept. Plan upcoming meetings with prep material and important items the assistant keeps in mind during the call.
ASSISTANT
Ask a question and get an answer drawn from the conversation so far, without stopping capture. The compact companion stays beside your call. Use a local Ollama model, or OpenAI or Anthropic with your own key.
INVESTIGATIONS
Collect pasted text, screenshots, links and voice notes as cards. Screenshot text is read by local OCR. Press play on one investigation and add to it from anywhere through the launcher: a voice note, the clipboard, a note, a link or a screenshot. See the cards as a chronology or a diagram, and ask the assistant, which answers with citations to your cards.
MESSAGES
Add people with a one-time invite and send text straight to them over the Tor network. There is no Phorminx server, and neither side learns the other's IP address. It is off until you turn it on, and history stays on your PC.
BENCHMARKS
Phorminx publishes quick benchmarks of its recognition engines and local models for each feature, with a plain recommendation for each. From the Benchmarks page, use a benchmarked model for a feature or download it in one click. Every place that needs a model opens the same picker, showing what is installed and what fits.
AND ALSO
Search everything you kept by exact words or meaning, send a dictation to any HTTP endpoint with Actions, set up verified models in one guided run, keep Windows awake with one switch, reinstall Phorminx in one click when something breaks, and get signed updates that never interrupt a recording.
Teach names, products, and jargon once. Profiles adapt formatting to each app you dictate into.
Find any dictation or meeting by its words, then reopen the session it came from.
Signed updates arrive automatically and never interrupt a dictation or a meeting.
PRIVACY
Speech recognition, translation, the assistant, and your history work with local models. Your audio and text stay on your PC unless you send them yourself: to a remote assistant or action you connect, or as feedback.
See exactly what connects whereMicrophone and computer sound are held in memory while they are transcribed.
Speech models, translation, and the assistant run here with local models.
Saved on this PC and kept for as long as your retention settings allow.
OPEN SOURCE
Read the code, follow the roadmap, and report what would make Phorminx better. Developed within Impossible G.
Free for Windows 10 and 11.
Download for Windows