Beta 1.0 · macOSMeet Veyra.
Your AI assistant for live meetings. It listens, works out what is being asked, and puts an answer on screen while you are still talking.
macOS 14 or later, Apple Silicon. Intel Macs are not supported.
Live captions on
0:41How it works
Four steps. Three of them never leave your Mac.
Veyra is built to a two-second budget: end of speech to first readable word.
Hear
System audio and your microphone, echo-cancelled and turned into text by a local model.
Your voice never leaves the machine.
Understand
A local model decides whether the turn is even worth answering.
Most of what it hears, it deliberately ignores.
Retrieve
Your briefing, your profile and your documents — searched locally.
18 ms at the 95th percentile, measured.
Answer
The question and the context you chose go to the model you picked.
The one step that leaves your Mac.
Veyra is designed to a two-second budget from the end of speech to the first readable word. That is the target the system is built against.
Any conversation
It is not one kind of meeting.
The same assistant, drawing on whatever context you gave it for this room.
Client call
Interview
Standup
One to one
Dana Ruiz: What would you say you have owned end to end this year?
- Three things: the payments migration, the retrieval rewrite, and onboarding for the two engineers who joined in May. All three are in the profile you built from your résumé and notes.
0:34Your context
It answers with what it knows about you.
Give Veyra your résumé and your notes. It reads them once into a profile, and every answer after that is written from your side of the table.
What you add
- Resume 2026.pdfPDF · 212 KB
- Project notes.mdMarkdown · 18 KB
- Short bioTyped · 2 KB
Read once, on your Mac. Nothing here leaves until you ask a question.
Your profile
- Role
- Staff engineer, platform reliability
- Focus
- Distributed systems · migrations · incident response
- Owned
- Payments migration · retrieval rewrite · onboarding
- Tools
- TypeScript · Postgres · Kubernetes
So the answer sounds like you:
- You owned the payments migration end to end — the rollback call, the pool-sizing fix, and the write-up afterwards.
1:11Briefings
Walk in already prepared.
Put the agenda, the documents and the relevant threads in one place before the call. During the meeting, that is what Veyra draws on.
Northwind — delivery review
Thursday, 14:00 · 4 participants
- AgendaWhat the meeting is for, and who is in it.
- DocumentsContracts, specs, notes. Indexed locally.
- SourcesCalendar, mail and Slack, for this meeting only.
- BuildVeyra reads it all and writes a digest.
Connected sources
One Google sign-in
- Google Calendar
- Gmail
Separate sign-in
- Slack
Read-only. Calendar and Gmail share one Google sign-in. What a connector brings in is used for the meeting and disposed when it ends — it is never added to your permanent knowledge base.
Workspaces
One meeting never leaks into another.
Keep a workspace per client, per project, per part of your life. Veyra can only retrieve from the one you are in.
- NorthwindClient
- HiringInterviews
- PlatformInternal
- PersonalPrivate
Retrievable in Personal
3 items- Career notes
- Salary research
- Side project spec
Everything in your other workspaces is invisible here. Not filtered — unreachable.
Local first
Most of the work happens on your Mac.
As much as possible stays on your Mac. What leaves is the question and the context you chose to give it — sent to the provider you picked.
Your Mac
No account needed to runAudio capture
System audio and microphone, echo-cancelled on device.
Speech to text
A two-pass local model. Your voice never leaves the Mac.
Understanding
A local model decides whether a turn is even worth answering.
Embeddings
Your documents are turned into vectors locally.
Retrieval
A local vector database. 18 ms at the 95th percentile, measured.
Everything else
Workspaces, briefings, profile, reports, settings.
Your API key lives in the macOS Keychain and is read only by the part of Veyra that talks to your provider. No part of the interface can request it.
The provider you chose
- OpenAI
- OpenRouter
- Google Gemini
Or one running on your own machine
- Ollama
- LM Studio
- vLLM
Anything that speaks the OpenAI-compatible protocol works. Adding one is a base URL and a model name, never new code.
Your cost is whatever your provider and model charge. Point Veyra at a model running on your own machine and there is no bill at all.
The overlay
It floats above whatever you are in.
Put it wherever it suits you. Veyra remembers, per screen. It stays on top of a full-screen app, and it never takes your keyboard focus.
Knowledge
Embedded locally. Never sent to a provider.
- Master services agreement.pdfIndexed · 42 chunks
- Rollout notes.mdIndexed · 11 chunks
- Incident review — payments.pdfIndexed · 28 chunks
- Q3 delivery plan.docxIndexing…
Listening…
0:51Control
You decide when it speaks.
Two settings, and they answer different questions. One decides what is worth answering. The other decides when an answer is paid for.
What Veyra hears
- “So what did we land on for the timeline?”detected
- “Mm-hmm.”ignored
- “…and the regions, remind me?”ignored
- “Right, go on.”ignored
1 detected
- Two weeks for the schema change, then a staged cutover across three regions.
1:03Automatic: Veyra answers what it judged worth answering, without you touching anything.
Incognito
Veyra stays on your screen and leaves the shared one.
Your screen
What you see, the whole time.
Dana Ruiz: Can you pull up the timeline we agreed?
- Two weeks for the schema change, then a staged cutover.
0:58The shared screen
Veyra is left out of the captured video.
Dana Ruiz: Can you pull up the timeline we agreed?
- Two weeks for the schema change, then a staged cutover.
0:58Incognito asks macOS to leave the overlay out of captured video. It is off by default — with it on, Veyra is also absent from your own screenshots and recordings.
After the meeting
The meeting ends. Veyra does not.
What was decided, what you agreed to do, and what nobody answered — written up while it is still accurate.
Meeting report
Northwind — delivery review
Thursday · 48 minutes · 4 participants
Summary
Decisions
Action items
Open questions
Risks
References
What comes next
And this is only the beginning.
Three things Veyra is being built toward. None of them are in the beta — they are here so you know where this is heading.
- reads the shared screen01Planned
Visual understanding
Veyra will read what is on the shared screen and use it when it answers.
- offers the next question02Planned
Proactive intelligence
Veyra will offer the question worth asking before you have to think of it.
- knows the project itself03Planned
Codebase awareness
Veyra will know the project you are describing, not just your notes about it.
Everything above is a direction, not a date. What is on this page above this section is what the beta actually does today.
Meetings move fast. Veyra helps you keep up.
Beta 1.0 · macOS 14 or later, Apple Silicon