voicer pairs a natural, low-latency voice loop with the full voicer app layer — documents, search, real analysis, navigation, custom skills, messaging and more — so every answer is grounded and every task gets done. It runs in your browser, on your iPhone and Apple Watch, and on a pair of smart glasses.
Access is by approval. Sign up and an administrator lets you in — see how access works.
Real-time, full-duplex audio in any modern browser. No phone number, no dial-in, no install — just allow the mic and talk.
Several AI experts share one call — each on the AI model you choose. A router reads each expert's expertise and hands every turn to whoever fits best.
voicer is model-agnostic. Mix different models across your experts so one conversation gives you answers from many perspectives, never just one.
Build a panel, export it to reuse or hand off — and when it's ready, publish it so others can talk to what you made.
Each seated expert renders as a card. Adjust eagerness, speaking rate, pitch, language and voice — live for this call or saved as default.
Give an authored panel character a personality blurb — it shapes tone and delivery every time that expert speaks.
A single header control starts and ends your call, with a few gentle "still there?" check-ins before it ever hangs up on silence.
Every call is transcribed and saved automatically — and gets an instant spoken and written recap the moment it ends.
Experts and live tuning happen in real time, so guests with no account get the full experience from a link.
Build a panel step by step — topic, sources, experts, then a fully editable review — before you ever go live.
It started in the browser and it still needs nothing but a tab. But the same conversation, the same documents and the same experts also reach your phone, your wrist, and a pair of glasses — and hand off between them. The iPhone app is on the App Store; the browser needs no download at all.

Real-time, full-duplex voice with nothing to install. Allow the mic and talk — on a laptop or a phone browser.

Download it and sign in with the same account. It rings like a real phone call, keeps a home-screen widget and a live call banner, and answers Siri shortcuts.

Ask from your wrist, read the answer there, and use the watch as a remote for your glasses — including walking directions.

A full heads-up interface on Even Realities G2 lenses: menus, answers, email, notes, navigation and a live transcript of the room.
No app, no browser. Text it to start a call, send a message, or ring a contact — and it texts you a working link back.
Ask it to ring your own phone so you can carry on there, schedule a call back for later, or bridge a real person into the conversation.
The Live tab starts as a text chat with your panel — and the moment you start talking, the call picks up the very same thread. Nothing gets lost in between.
Chat with your experts before any call exists. Start a call mid-thread and it absorbs the whole conversation — even text you typed while it was still connecting.
Live chats save themselves. Open a past chat and continue exactly where you left off, or start a fresh one with a tap — the transcript stays yours.
The expert picker surfaces every expert you've ever built, lets you star favorites, and can create a brand-new one inline — no detour to the studio.
Add a file, snap a photo, or record an audio clip right inside the conversation — the panel sees it immediately.
Heard enough? One tap cuts the panel off mid-answer — you're always in charge of the floor.
A live roster shows every expert and every listener on the call, as they join and leave.
Not a persona — a profile modeled on a real person's own words, files or voice.
Teach a Twin from pasted text, an uploaded CV, PDF, deck, recording — or just a link. voicer builds the profile for you.
Give a Twin the owner's cloned voice — recorded or uploaded — or pick from the default roster, and preview it before it speaks.
Propose and pair two Twins automatically for practice conversations, mock interviews, or exploring a match.
Paired with Even Realities G2 glasses, voicer becomes a heads-up display you steer by voice, by the temple touchpad, or from your phone or watch. Hands stay where they were.
A real menu on the lens — a glance view, today's agenda, your notes, your documents, your apps, your email, and the live conversation.
Say what you want to know and the reply is laid out on the lens, paged so you can read at your own speed instead of listening to all of it.
Ask to be taken somewhere. It searches, offers a shortlist, then draws live turn-by-turn guidance where you're already looking.
Listen mode quietly follows the conversation around you and gives you a running transcript, a recap, the unfamiliar terms, suggested replies, and a translation.
Freeze what's on the lens while the room keeps talking, page through it, then let it go and catch up. Nothing scrolls away mid-sentence.
Open your inbox by saying so, read a message, and dictate a note without reaching for a phone.
A playable chess board on the lens with a real engine and opening book behind it — moved by voice or by the touchpad.
Voice, the temple touchpad, or a phone and watch remote — plus a browser simulator, so you can see exactly what the lens shows without wearing them.
Point a camera at something and ask about it, or have it hunt around the room for the thing you put down and can't find.
Not a copy of your calendar — the actual one on your phone. The app is a real phone-call app on the App Store, and it reaches the parts of iOS a web page can't.
Ask it to ring you and your phone rings properly — lock screen, ringtone and all. Say what it's about and the call opens on that.
Read what's actually on your iPhone calendar, add an event, or drop a to-do into Reminders — without emailing anybody an invite.
Play a song, album, artist or playlist through Apple Music and steer it by voice — or ask what's playing in the room and it names the exact recording.
Conditions now, today's high and low, or the shape of the week — from Apple's weather service, for wherever you actually are.
Take a photo and ask about it, make a picture from a description, and come back later to ask a new question about something you already showed it.
A home-screen widget, a live banner while a call is running, and Siri shortcuts so you can start one without opening the app.
An on-device model takes over when you're offline, so basic answers keep coming when the network doesn't.
Ask from the watch and read the reply there, or use it as the remote for your glasses — including stepping through navigation.
Ask what's around you or how to get there — voicer uses your live GPS position, or a location you set yourself.
"Find me a coffee shop nearby" — voicer searches around your live GPS location and reads back real results.
Ask for directions to any place it finds, and get spoken, step-by-step turn-by-turn guidance.
No GPS? Set your location by voice or in Settings — voicer still gets your local time and location-aware answers right.
voicer isn't limited to what it ships with. Describe a small tool mid-conversation and it drafts one on the fly, for you to review before it ever runs.
Ask for a custom capability mid-call. voicer builds a small "skill" for it — you approve or reject it before it's ever used.
Ask the assistant to build, name and check on small apps for you — no code, just a conversation.
See every app you've built, check its status, and run it — from the panel or the app UI.
Have the assistant click buttons and type inside the apps it built for you — you stay hands-free the whole time.
Want a change? Just describe it. voicer revises the app, asks which app you meant if the name is ambiguous, and verifies the edit actually landed before saying so.
Ask "can you do X?" and get a straight answer — voicer checks every built-in tool and every app you've built, and offers to build whatever's missing.
The text chat runs the same tools and apps as the voice — build, run, and check on your apps without ever unmuting.
An app it built for you is a real deployed thing on a link — and whatever a run produces comes back as something you can play, read or download, never a wall of raw data.
Beyond its built-in skills there's a searchable catalog of platform blocks — image generation, document tools, mapping, data stores — that it can run for you on the spot.
Audio, video, images, PDFs, documents, datasets and charts an app produces all come back as real, playable links — and anything still rendering says so plainly instead of showing an empty box.
Every run of every app you built keeps its result, so you can go back to what it made last week without rerunning it.
Hit something broken and just say it. It files the report, sorts and tags it, and you can read your reports back and mark them done later.
Ask what's building or working in the background right now and get a straight answer, rather than wondering whether it stalled.
From what you've actually been doing — the gaps you hit, what you keep coming back to, what you complained about — it proposes an app and texts you the idea.
Deliver a report as a link that answers questions out loud — recipients just open it and ask.
Send any report or document as an interactive voice link. The recipient opens it in a browser and asks questions — no account, no reading assignment.
Recipients can re-share the link. Each new reader gets their own conversation with the material, and the chain carries the context along.
Interactive links say hello in about a second and get straight to your question — no long wind-up, no dead air.
Not a language model guessing at numbers. Real statistics, real filings, real simulations — and it says plainly when it can't support a claim.
Correlations, group comparisons, regressions, trend forecasts, segmentation, data-quality checks and side-by-side comparisons — run on a spreadsheet you uploaded.
Search public datasets by topic and pull one straight into your library, ready to ask about.
Bar, line, scatter, pie and heatmap charts, plus shaded maps of US states, counties or world countries — drawn while you talk.
Live stock prices and history, company financials pulled from SEC filings, and economic numbers like inflation and unemployment. Information only — it never trades.
Run a year-by-year projection of a retirement or budget plan, then a thousand-run simulation over it, and edit the assumptions by voice.
Mine the published literature on a topic for where it's thin or out of date, and rank what would actually be worth studying.
Beyond a quick lookup: a multi-source dive that runs in the background and comes back with a real report.
Send it off to check availability, prices or opening hours on a live site — read-only, and it reports what it saw.
Turn a document or a brief into a real, shareable deck on a link — not a description of one.
Say “branch this” and everything after is a side path. Go back to the main conversation and the detour is lifted out and forgotten.
With several people round one phone it tells the voices apart, and puts a name to each once you say who someone is.
Say “speak Spanish” and the whole conversation follows — what you hear, what the experts say, and what gets written down.
Tell it to go quiet and it stops volunteering — following along, speaking only when spoken to.
Put it right once and no expert, call or chat repeats the mistake afterwards.
Have a drafted answer checked by an independent skeptic before you hear it — with how much it disagreed, and its strongest objection.
Find past conversations by topic, by person, or by roughly when they happened — across calls and chats alike.
It offers to continue your last conversation at the start of a new one, and loads what was said so the two carry on as one.
Switch between you-driven, free discussion, and best-fit routing — and check whether an earlier panel ever finished.
Upload files and voicer retrieves from them with vector search, answering and summarizing with grounded citations.
Pull fresh facts from the open web mid-conversation — pulled in the background, with a spoken heads-up when it's ready.
Duplicate files get flagged automatically, and AI sorts documents and notes into folders for you.
Merge several documents into one new document — on a call or right on screen.
Drop in a YouTube video, an article, or just a link by asking — or drag a file straight into the library.
Open a clean, read-aloud version of any web page, or read books in a Kindle-like layout with highlight-to-ask.
Group related calls, chats and documents under one project — kept tidily out of your main Documents tab.
Capture notes and manage contacts hands-free — phone numbers are even read back digit-by-digit so they're easy to follow.
Send a Google Calendar invite by voice or from the app — no switching tabs to schedule something.
Get a text-only transcript of any episode, or publish your own call as a podcast with a real, public RSS feed.
Turn any conversation into an Instagram post, carousel or reel — voicer drafts the content straight from what was actually said.
SMS voicer to start a panel call, send a message, or ring a contact — fuzzy name matching included, and you always get a working link back.
Ask what's in your inbox and get answers — a private, read-only sync of your own Gmail.
Outbound calls the assistant places connect the other person straight into your live call.
Multiple people can take the CEL quiz together in one voice session, joined from a single shareable link.
Say "show me a chart" or "pull up that website" and it appears on screen, alongside the voice.
Play, pause, replay, seek, zoom and mute video on screen — all by voice, reliable even on mobile networks.
A live, visual map of claims, questions and decisions builds on screen as the call happens.
Ask any expert to talk faster or slower, or set overall speaking speed from a slider — live or as your default.
Expand the call to fill your screen — pause, summarize, listeners, docs and camera controls stay one tap away.
Real, live per-call spend tracking, so cost is never a surprise.
Prefer typing? A full text chat and an admin console ship alongside the voice client in the same app.
voicer watches its own rough edges, drafts small fixes, tests them, and ships only the ones that pass — on a hard budget, with you in the approval loop.
Every night the platform reviews what failed or fell short, builds targeted fixes, and ships only fully tested green builds — with automatic rollback if anything regresses.
Proposed builds arrive on your phone. Reply YES or NO to a numbered list — or just describe the build you mean in plain words.
An automated tester uses the product like a real customer — talks to experts, runs reports, builds and plays apps — and files whatever breaks.
Autonomy runs on a strict daily and monthly budget with alerts along the way. A live spend view shows every dollar — on screen or just ask by voice.
All credentials live in your own secure vault. Nothing sensitive is baked into the app.
Flag a document — or a whole project — as sensitive and it's only ever processed by compliant AI providers, automatically. HIPAA-minded projects stay locked down.
Registration is open; access isn't. An administrator has to accept an account, and that check re-runs on every request — so a suspended account loses access straight away rather than when its session expires. How access works.
It's a cloud service — request access, get approved, and go. No servers to manage, nothing to install or operate. Self-hosting is available on Enterprise.
Retrieval-augmented generation keeps responses tied to your uploaded documents and the live web — so the panel reasons from sources you trust.
Create experts from a simple description or live as you talk. Give each one its expertise, voice, pace and the AI model behind it — and decide how eagerly it interjects.
voicer isn't open sign-up. Anyone can register, but an administrator has to accept the account before it can be used — and that check runs on every request, not just at the front door.
Sign up with an email and password, or with Google. The account is created immediately and left switched off.
Your request lands in an approval queue. Until someone approves it, signing in only ever tells you you're still waiting.
Once approved, sign in normally. New accounts start with the core features, and extras get switched on for you.
Approval can be withdrawn: an account an administrator suspends loses access on its next request, on every surface — the web app, the phone app, and any device it had paired.
Generated straight from the platform's own capability registry — the same list the assistant reads when it answers “can you do X?”. Open a group to see what's in it.
Talk with AI experts one-on-one, or run a panel, debate, or interview.
Bring one or more AI experts onto the call, by role (a lawyer) or by famous name (Plato); a real name is looked up first so the expert is the real person, and if the name matches more than one the best is brought in and can be swapped ('the other one'); an expert you already saved is reused, and 'research him properly first' builds a source-backed profile for this call without saving it; when you haven't said how thorough you want it, you're asked first — a quick look-up or deep web research — and the expert is built in the background and announced when they're ready to join, so the conversation carries on meanwhile.
Create an expert profile and save it to the expert library for reuse, at one of three levels — save as described (instant), AI-drafted from a name or topic (draft), or deep-researched from real web sources (deep); you are always asked which when you haven't said how thorough you want it — nothing is saved until you choose. For an ambiguous real name, first look up candidates and let the user pick which person (or narrow by topic, or hear more) before saving.
Build a new expert whose knowledge comes from one of the caller's uploaded documents or photos, and save it to the expert library.
Run a live interview, debate, or multi-person panel among experts.
Remove one expert, or clear all of them.
Narrow or re-steer an expert's focus mid-conversation.
Save an expert for reuse, teach it durable facts, share it, list or delete saved experts.
Remember a correction after the assistant gets something wrong, so no expert, call, or chat repeats the mistake.
Go into silent listen mode and only speak when spoken to.
Turn an expert's 'check current events' trigger on or off — when on, the expert looks up live web results for time-sensitive questions instead of answering from its training cutoff.
Switch how experts take turns (user-driven, free discussion, or relevance-based), and check whether an earlier panel finished.
Make an expert speak in one of the caller's cloned voices, and speed up or slow down speech.
Upload files and links, then ask about them, take notes, and turn them into experts.
Search everything the caller has talked about before — past conversations by topic, person, or when they happened — and index older ones so they can be found.
List the caller's uploaded documents and search inside them to answer questions.
Read a full document aloud, combine documents into one, and unzip an uploaded archive (.zip, .tar, .gz, and more) into its files.
Organize documents into folders, list those folders, find and remove duplicates, and rename, move, or delete a single document.
Save a quick personal note for later.
Read back the caller's saved notes.
Keep a private daily diary — talk through your day and I'll ask a few follow-up questions, then save it.
Write a full report, summary, or plan in the background and save it to the caller's library.
Build a real slide deck (pitch deck, presentation) from a document or a brief, and put it on screen at a shareable link.
Group calls, chats, documents, and notes into named projects.
Put an Open Audible button on their screen, and keep notes on the audiobook they're listening to — they say "save this", talk, and it works out the idea and asks one question back. It cannot play, pause, skip, or see what's playing: an iPhone app can't reach into another app.
Search the web, run deep research, and get real reports back.
Search the web for live information.
Run a deeper multi-source research dive on a topic.
Do a live read-only browsing errand (check availability, prices, hours).
Find public Kaggle datasets by topic and download one into the caller's documents.
Run real statistics, correlations, group comparisons, data-quality checks, trend forecasts, segmentation, regressions, or a side-by-side comparison on saved CSV datasets (e.g. downloaded from Kaggle) and show a chart of the result.
Forecast a trend, flag unusual values, compare two columns, or break down spending by category from a saved CSV dataset.
Mine the published literature (PubMed) for a topic to find where it's under-studied or out of date, and rank publishable research ideas — what to study next; runs as a background task that texts results and saves them to Tasks, then can design a study for the most promising idea.
Look up live stock prices and history, company financials from SEC filings, and economic numbers like inflation or unemployment — information only, no trading or investment advice.
Stress-test a retirement or budget plan: a year-by-year projection plus a thousand-run monte carlo simulation over the saved budget-and-assets scenario, with optional market crashes, a one-time expense, higher spending, or long-term-care costs — a math model of the caller's own numbers, not licensed financial advice.
Read and edit the owner's saved retirement plans in the retirement planner app: change assumptions by voice with spoken confirmation, undo the last change, run projections and monte carlo simulations, compare and clone scenarios, with charts pushed to the screen — a math model of the owner's own numbers, not licensed financial advice.
Search a topic and save the best pages into the caller's library.
Fetch and read a specific web page, and search for videos or images.
Read, search, and send email without leaving the app.
Search the caller's connected email inboxes and read messages aloud (read-only unless asked to send).
Send an email on the caller's behalf, after confirming.
Send documents, reports, and conversations by email, text, or a live voice link.
Watch an email chain with named people and give you a morning list of what's waiting on you, each with a draft reply you approve before it goes — money, contracts and refusals always wait for your own words.
Send a document, this conversation, or a looked-up article to someone by text or email — as an interactive voice link, PDF, web page, or text file — after confirming channel and format.
Send a plain text message (SMS).
Draft and send a Google Calendar invite with a Meet link, after confirming.
Read what's on the caller's own iPhone calendar, and put new events on it, without emailing anyone.
Add a to-do to the caller's own iPhone reminders list.
Text or email someone a link to join the current live call from their browser.
Save people to the caller's contacts (name plus phone or email), read back the saved contact list, and see or change the default recipient.
Prepare and publish an Instagram post, after confirming.
Hand off longer jobs that keep working and report back when done.
Kick off a durable background research or browsing task that texts results later; list task status; refine the latest task.
Flag a capability the platform can't do yet so it's logged for the team to review, instead of just saying no.
Turn calls and documents into podcast episodes you can subscribe to.
Generate a brand-new fictional podcast (transcript or full audio) in the background.
Publish a podcast episode from a past call OR a written script — clean up a real call, or produce an already-written script faithfully with voices; hear it read aloud or fully voiced, edit the whole script or revise just one line by voice (even mid-playback), and preview on screen before publishing.
On a call, show things on your screen or point your camera and ask.
Look through the caller's camera — take a photo of what's in front of them and answer about it.
Read a picture or recording they already shared again, with a new question — 'how many people are in that photo?', 'what does the label say?', 'this is a painting, interpret it', 'check the grammar in that talk' — instead of answering from the general description written when it arrived.
Look around the room through the caller's camera watch and find something in it — 'where did I leave the clock?', 'is my phone in here?' — by having the watch take several pictures while they turn their wrist; says the watch isn't awake rather than pretending to look.
Put lists, tables, charts, images, videos, websites, and the caller's own saved documents (a report, pitch deck, notes) on their screen, and control them by voice (play, pause, scroll, zoom).
Control the YouTube video player itself when a YouTube video is on screen — play, pause, stop, fast-forward, rewind, jump around, or change playback speed — by sending player controls instead of listening to the video's audio output.
Get a picture of anything onto the caller's screen — either drawn fresh from their description in the look they choose (photo, illustration, watercolour, cartoon, sketch, 3D), or a real one found on the web; asks which of the two they want rather than guessing, keeps every picture in their Images library, and can redraw one they already made with a change ('make the barn red').
Turn the caller's latest picture into a short AI effect video — melt, explode, cakeify, claymation, Ghibli style, action figure, zombie, plushie and about thirty more; the render takes a minute or three and lands on their Live screen by itself.
Find or show anything the caller recently made — a document, a rendered podcast or panel transcript, an app's output (audio/video/image/pdf), or an app they built — even when they don't name the type ('show me the thing I just made', 'is the transcript done?').
Put Voicer back on the caller's heads-up glasses when their display has gone blank or dark — their own control screen takes the lens and leaves it empty until something new is sent.
Draw richer visuals on the caller's screen — standard charts (bar, line, scatter, pie, heatmap…), choropleth maps shading US states, counties, or world countries by a value, and Vega-Lite diagrams like dendrograms, treemaps, and network graphs — built either from data you supply inline or straight from a CSV/Excel dataset already saved in the caller's documents, without re-reading or pasting the rows — and change a chart already on screen in place (a different type, a different order, just the top few) instead of drawing a second one.
Draw a process as a real flowchart from nothing but a description — boxes, decisions and arrows for how something works ('map out our onboarding', 'show me those steps as a diagram'), plus the exchange between named parties, the states of one thing, and a plan over time.
Call real phones, keep contacts, and review past calls.
Look up past calls by topic, read a past call's full transcript, and report the caller's usage spending.
Join a Zoom, Google Meet, or Microsoft Teams meeting as a live participant given its link — it listens quietly and only speaks up once addressed with the wake word, so it never interrupts; leave a meeting on request, and check which meetings it's currently in. Also on the Meetings screen on the web and the phone, in typed chat (pasting the link is enough), and as a "Join with Voicer" button beside a calendar event that carries a meeting link.
Bring experts into a Zoom, Google Meet or Microsoft Teams meeting, each set one of two ways: out loud, where they speak through the meeting bot and everyone in the meeting hears them and sees their name and picture on its video tile, or privately, where they answer only the caller as text on whatever they're signed in to and nobody else in the meeting knows they are there. Several at once, mixed between the two, moved between the two mid-meeting, and grounded in the documents, pages and pictures attached to that saved expert. Also on the Meetings screen on the web and the phone. Each expert also has its own way of taking part: silent until the caller says its name (the default, so nothing ever speaks up uninvited in a client meeting), speaking when the subject is squarely its own, or chiming in freely — set per expert, saved on that expert, and overridable for one meeting. An expert called on by name answers whatever its setting, and is given the whole meeting so far rather than the last minute.
Hand an expert more to read in the middle of a meeting — a file, a picture, a link or some text — and by default it stays with that expert afterwards, so the same expert knows more every time it is used; "just this meeting" is the exception and says so. Everything added can be heard by the meeting unless the caller marks it as theirs alone, and anything given to an expert that answers only them is theirs automatically and is never read out to a room. Say what is still being read, and what couldn't be read at all, rather than leaving an expert that hasn't seen a file looking like one that read it and had nothing to say. Also on the Meetings screen on the web and the phone.
Meet the same people on the same link again and last time's experts come back on their own, each in the mode it left in, with a notice that says so and one tap to take them out; and when a meeting ends, add what was said to an expert so it knows it next time — it adds to what that expert already knew and never replaces it, asking twice does not attach a second copy, and the meeting stays in call history whether or not it is added, so it can be done days later. Also on the Meetings screen on the web and the phone.
Offer at the start of a new session to pick up the caller's previous conversation, and load what was said in it so the two carry on as one.
Ring the caller's own verified phone so they can continue there.
Ring the caller's iPhone later — "call me back in ten minutes" — and open that call with what they said it was about, or call it off if they change their mind.
Call a saved contact or a number you say out loud, and bridge them into the conversation so you talk to them directly — with a spoken check of the digits first when you dictate a number.
Build small working apps by voice and run ready-made platform tools.
Double-check a drafted answer with an independent skeptic pass before delivering it — disagreement score plus the strongest objection.
Analyze your past call history to extract requests that fell outside the platform's scope, find patterns, and report back your top unmet needs.
Tell the caller whether the platform can already do what they want — match their request against this capability list, ask a follow-up if it's unclear, and on a genuine gap offer to build it (admins) or log it for the team.
Give real examples of what to SAY for any part of the platform — "give me examples of creating experts", "what could I ask about email?", "show me some things I can ask for" — the phrasings come from this same list, so they all really work; on a screen they arrive as buttons you can tap to run one, and on a call they are read out and you are asked which to try. The whole set is on the Examples page.
When the caller has nothing specific in mind, invent something interesting for them — survey everything the platform can do and pitch a few (usually multi-step) ideas that chain existing abilities, refine with them, then run the one they pick (building any missing step first, for admins).
Search a live catalog of extra platform abilities (blocks) and run them — the catalog covers things with no dedicated tool.
List the caller's built apps, look up one app's full details (description, publish date, link, inputs), fetch their code, and run them.
Report what's building or running in the background right now, and file a bug report the caller hit (sorted into a folder with tags).
Read back the caller's filed bug reports (by folder or tag, also shown on their screen), mark one open / in progress / done, or delete one they no longer want tracked.
Build and deploy a brand-new app from a spoken description — the plan is summarized and agreed first (propose_app), then built (build_app).
Pick up an app build that stopped — it ran out of tries or hit the money limit for one app — on the same record and the same running total, with any extra spend said out loud first.
Change one of the caller's built apps from a spoken description and redeploy it in place (same link).
Draft a new single-action skill on the fly when nothing else covers the request.
Search and read the platform's own code, propose fixes, deploy or revert updates.
Build the next thing on the platform's own backlog of unmet requests — draft it, test it, and stage it to ship after you approve.
Test the platform against its own power-user persona — run the self-QA probe suite (wake mishears, owner scoping, duplicate titles, done-means-stored), or send a robot caller into real calls to drive whole conversations (deep mode also summons an expert, writes a research report, and builds then plays a real app), and read back the verdict.
Work out what app the owner needs from what he has actually been doing — the gaps he hit, what he keeps coming back to, what he complained about — and pitch one, with the reason; the build itself then goes through the normal app builder, and a no is remembered so the same idea is never pitched twice.
When something keeps failing because the platform is missing a backend tool, draft that block — name, operations, what it unlocks — text the owner, and queue a build spec on a yes.
Check on or drive the self-improvement system by voice — reflection loop status, the full history of every reflection and build it has done, what's waiting for approval, ship or decline one of those tested fixes out loud (the same ones the approval texts are about), start or stop a run.
One unified read-back of what the platform costs and what the autonomous systems are doing — spend by category (LLM, voice, transcription, tools), month-to-date against the cap, the builder and reflection budgets, and the latest autonomous builds, reflections, and robot test runs with their cost.
Report what coding work was done on the platform lately (the daily dev work log, by project) and what the nightly digital-twin review of those sessions found — bugs caught, new recurring failure patterns added to the checklist.
Run the guided business quiz and share its link.
Find places nearby, get directions, and hear the real forecast, right on the call.
Get or set the caller's location, show their position and other places as pins on a real street map, find nearby places, give directions, and hand back a shareable map link to text someone.
Take the caller somewhere: search for the place, offer a shortlist, then show live turn-by-turn directions on their glasses.
Give the real forecast from Apple's weather service — conditions now, today or tomorrow's high and low, or the shape of the week — for where the caller is or any town they name, in the units that country uses.
Play music on your iPhone through Apple Music, name what's playing in the room, and steer it by voice.
Play music on the caller's iPhone through Apple Music — a song, an album, an artist or one of their playlists — and pause, skip or stop what's playing; needs their iPhone signed in to Apple Music, and says so plainly when it isn't.
Listen through the caller's iPhone and name the music playing in the room, then offer to play that exact recording; says plainly when it recognised nothing, and never guesses at a song.
Steer a live call — mute, hand off, invite others, wrap up.
Branch the conversation you're in so you can try a direction and undo it — say 'branch this' and everything after it is a side branch, then 'go back to the main conversation' to lift it out and forget it (it's saved as its own conversation) or 'keep it' to fold it in; saved branches can be read back and merged in again later.
Tell apart the different people talking into one phone in a room, and put a name to each voice when the caller says who someone is ('the second speaker is Dr. Park') so the transcript reads as a real conversation instead of one anonymous voice.
Hold the whole conversation in another language — say “speak Spanish” or “switch back to English” and everything follows: what you hear, the experts, and what's on screen.
Change any of your own settings by just saying what you want — “stop cutting me off”, “you talk too fast”, “wait longer before you answer me”, “turn off the Books tab” — ask what they're set to, and undo the last change.
Switch YOUR speaking voice mid-call — say “use WaveNet”, “switch to Neural2”, or “use Chirp” (male Google tiers; WaveNet/Neural2 cheaper).
Analyse the call so far: summary, key decisions and commitments, risks, next steps.
Record a note into the call's running history.
Wipe the conversation context and start fresh.
End the call for everyone, with a short farewell.
106 capabilities across 13 groups, plus the panels, apps, and devices described above. “On a call” and “in the app” are different engines, so a few things work in one and not yet the other — the labels say which.
Open a tab and meet your panel — every feature, one conversation away.