Each card names its shape first and its industry second. The industries have nothing in common; the shapes keep repeating.
Several things in one breath. Each lands on screen in the user’s own words, then resolves against the app’s real data on its own clock.
One intent becomes many edits. They appear in place, in the document the user is already reading, for review before anything is kept.
Nothing transacts. The agent takes the user to the place, highlights the control, and explains it there — instead of reciting a menu path aloud.
The agent does the assembly — pulling the record, reconciling, drafting — and stops at the commit. The human presses the button.
The exception. It shows what the speech layer under all of the above does on its own — switching language mid-sentence, in native script, without losing the thread.
Plan a trip in one sentence.
Talk through dates, cities and budget — the itinerary builds itself on screen as you speak, flights and hotels searched in parallel in the background.
Ask your bank, out loud.
A voice assistant on a bank site: it opens the how-to, signs you in securely, reads your balance, and reads the journey behind the question to offer the right next product.
Shop by talking.
Ask for a phone by budget and need; the store filters, compares specs and drops it in the cart — on the screen, in real time.
Return an order by voice.
The agent troubleshoots to deflect an unneeded return, pulls up your order, reads a product photo you show it, and fills the return form for you.
A colleague at the console.
A mortgage-servicing copilot that works up a case in the background, reconciles the numbers, catches the blocker that gates a regulated step, and drafts for a human to approve.
It starts the conversation.
A daily diabetes check-in that starts from a notification, turns "curd rice and two rotis" into a logged meal with calories, and knows exactly where the clinical line is.
A copilot for the clause you’re on.
An ambient copilot for in-house counsel: talk through a vendor MSA and it marks the clauses on screen as you read, staying in the document with you instead of taking it over.
A first-round interviewer.
The job, the candidate and the interview plan arrive with the session; the bot works the plan, asks its way through, and tracks the conversation on an interview panel as it goes.
Build a workflow by talking.
An ITSM/HR admin assembles a Service Request Workflow by talking to a copilot: it wires the steps, surfaces the untested edge cases, and runs the tests — live on screen.
Say the order, watch it resolve.
A pharmacist takes a 9 AM call and reels off a bulk order in Hindi. Every line lands on screen in their own words, then resolves against a real 20,148-SKU catalog — where a name is ambiguous you get pills to pick from, not a question to sit through. The order is confirmed by hand.
A talking head that explains itself.
Ask the open-source avatar how it works and it puts the architecture on screen, demonstrates its own protocol on its face, and changes which avatar it is — voice and all — mid-call.
Qualified in a conversation.
A gold-loan agent asks what it needs, checks eligibility on the spot, and switches to your language mid-sentence — native script and all. A conversation, not a form.