REAL-TIME MEETING COPILOT FOR MAC AND WINDOWS
Say the right thing, live.
Repeeter listens to the other side of your call, notices questions and doubts, and shows you what to say in a floating window that screen sharing never sees. Sign in and it just works, or run it with your own AI keys or fully local models.
Every answer comes from your own notes, in your voice, with a Check line when something is missing. Not a chatbot guessing on a second screen.
macOS 13 or later on Apple Silicon · Windows 10 or later. Free account with a trial of our AI, or free forever with local models.
Let me make sure I bring the right people in.
Usually that is whoever signs, whoever uses it every day, and whoever owns the budget line.
Security review tends to gate procurement, so I would loop them in early rather than at the end.
Who would you want on that call from your side?
~s
typical time from the end of their question to the first line of your answer, managed or with cloud keys
audio channels, so it knows which words were theirs and only answers them
meeting audio or transcripts stored on our servers, on every plan
%
offline in Local mode: Whisper and Ollama on your machine, no network during a call
What it does
Built for the second after they ask.
Hears the meeting
Listens to the other side of your call and catches questions and doubts as they happen.
Answers from your notes
Every answer is grounded in the files you drop in. Anything it cannot find gets a Check: line, never a guess.
Invisible when you share
The overlay is excluded from screen capture on macOS and Windows. They see your slides, you see the answer.
Written to be said
Spoken English that sounds like you, not bullet points to read off. Code arrives in a copyable block.
Anything on screen
A question in the chat, a slide, an error. One shortcut captures the screen and answers it.
A teammate on your side
Wingman sends each question to a colleague’s phone; their answer lands in your overlay as From Priya. Their own ChatGPT or Claude can answer too.
How it works
Four steps. Then it stays out of your way.
Click through it. The whole setup is shorter than this section.
Your AI, your choice
Managed, your keys, or local.
Sign in and we run the AI with frontier models from Anthropic and OpenAI, streamed so the first line lands about two seconds after they stop talking. Or bring your own keys, or keep everything on your machine. Transcription and answers can each be set separately in Settings.
Managed, sign in
Sign in and go
- Transcription
- OpenAI realtime, streamed straight from your computer with a credential we issue
- Answers
- Frontier models from Anthropic and OpenAI, chosen and tuned by us
- Where it runs
- Provider APIs; transcript text passes through our server and is never stored
- Cost
- Free trial: 60 minutes and 40 answers. Pro: $19 a month
- Best for
- An instant feel with nothing to configure
Your keys, on Pro
Your accounts, direct
- Transcription
- OpenAI realtime, your key
- Answers
- Claude, GPT or any OpenRouter model, your key
- Where it runs
- Provider APIs, under your own accounts
- Cost
- What the providers charge, cents per meeting
- Best for
- Full control over models and spend (Pro plan)
Local, on this machine
Nothing leaves the machine
- Transcription
- Whisper on your GPU
- Answers
- A model through Ollama
- Where it runs
- This machine, no network during a call
- Cost
- Nothing, no keys, no accounts
- Best for
- Privacy and offline work
Technical calls
Code when the question needs code.
Switch the meeting type to Technical and answers come engineer to engineer: the trade-off first, then a minimal, runnable snippet in a block with its own copy button, then the edge case worth mentioning. A screenshot of code or an error in the chat gets the same treatment.
Make the consumer idempotent: key each event by its delivery id and reject repeats at the database.
CREATE TABLE processed_events ( event_id text PRIMARY KEY, seen_at timestamptz NOT NULL DEFAULT now() );INSERT INTO processed_events (event_id) VALUES ($1)ON CONFLICT (event_id) DO NOTHING RETURNING event_id;
If the insert returns a row, process it; if nothing comes back, it is a duplicate.
Wingman
Bring a teammate into the call.
Invite a colleague with one link. From their phone they see each question as it is asked and send back an answer, which lands in your overlay under that question within a second. Their own ChatGPT or Claude can answer directly too, through a private connector. Nothing but the questions you share ever reaches them.
They asked · 10:54
Can you integrate with our Salesforce instance?
You on their screen
Yes, natively. One OAuth connection by your admin, about ten minutes, no managed package.
They asked · 10:56
What does onboarding look like for ten reps?
Private by design
Your conversations stay yours.
You are letting an app listen to your meetings, so here is exactly where the audio and the words go in each mode. The full privacy policy lists everything we store.
Local mode
Audio and transcript stay on your machine. The app makes no network requests during a call.
Cloud with your keys
Audio goes to the transcription provider you chose and transcript text to the answer model you chose, under your own account. Nothing passes through us.
Managed plan
Audio goes straight to the speech provider with a short-lived credential we issue; it never reaches us. Transcript text passes through our server to the model and is not stored. We keep minutes and token counts for billing, nothing else.
- Sessions are saved as plain files on your computer; delete them whenever you like.
- API keys you enter are encrypted by the operating system in your own settings file.
- Our servers hold your account and plan, and never a transcript or a recording.
- We remind you to tell participants when you transcribe a call.
Your next call is a good time to try it.
Free to start: sign in for a trial of our AI, or run local models. Setup takes a few minutes, and the first answer tells you whether it is for you.