WebAI StudioWebAI Demos
Explore what's possible with On-Device AI directly in the browser. No API keys, no server costs, full privacy.
Embeddings
Moderation Cascade
The production pattern for cheap moderation: embeddings clear the obvious 90% in microseconds, the LLM judges only the borderline.
Chat With Your Document
Fully local RAG: retrieve the relevant passages with embeddings, then answer with the Prompt API — citations included.
Semantic vs. Keyword Search
Search a help center by meaning, side by side with keyword search — and watch keywords miss what embeddings find.
Smart Triage
A zero-shot classifier: route incoming messages to categories defined only by a few example phrases.
Duplicate Detector
Catch near-duplicate bug reports before they are filed, even when they share no words with the original.
Cluster & Label
Group raw user feedback into themes with k-means over embeddings, then let the Prompt API name each cluster.
Semantic Cache
Serve instant answers for paraphrased questions by caching LLM responses under their embeddings.
Semantic Command Palette
A ⌘K palette that understands intent: describe what you want in your own words and the right action lights up.
Hot or Cold
Guess the secret word: every guess is embedded on-device and scored by how semantically close it lands.
Speech
Tongue Twister Trial
Say "she sells seashells" as fast as you can — the on-device recognizer scores how much of it survived.
Live Translated Captions
Speak into the mic: on-device captions appear instantly, and a second track renders them live in another language.
Speak to Fill
Say "table for four next Friday at seven, outside" and watch the booking form fill itself — speech in, structured data out.
ASR Quality Tier Lab
Read the same passage against the command, dictation, and conversation models — then compare accuracy and latency.
Jargon Dictation
Dictate product names the recognizer has never heard — then boost them with contextual biasing and watch the diff.
Dictate & Polish
Speak your messy first draft, then let the Proofreader and Rewriter turn it into publishable text.
Story Time
The Writer invents a bedtime story from three ingredients, then the browser reads it aloud — highlighting each word as it speaks.
Text Input
Mystery Language
Guess which language the snippet is written in — then see if you beat the Language Detector's confidence-ranked answer.
Proofreader, Inline
The dedicated Proofreader API: every error underlined in place, with a label, an explanation, and one-click accept.
Tone Pad
A 3×3 pad of the Rewriter's option space: pick a tone and a length, and the same text reshapes live.
Summarizer Options Matrix
One article, every Summarizer option: tldr, key-points, teaser, and headline at each length, side by side.
Reply Composer
Pick an intent — accept, decline, stall — and the Writer drafts the reply to a real email, streaming, with tone and length knobs.
Translation
Translate text from one language to another with native-like fluency.
Summarization
Condense long articles or text into concise, digestible bullet points.
Proofreading & Grammar
Fix grammatical errors and improve sentence structure.
Tone Changer
Rewrite casual text into a formal, professional tone.
Brainstorming Ideas
Generate creative ideas for a given topic.
Write JavaScript
Generate functional JavaScript code from a description.
Write HTML/CSS
Generate responsive UI components with HTML and Tailwind CSS.
Explain Like I'm 5
Simplify complex technical or scientific concepts.
SQL Query Generator
Convert natural language questions into executable SQL queries.
Draft an Email
Quickly draft a polite email from a short description.
Regex Lab
Describe a pattern in English, get a regex with an explanation — and test it live with match highlighting.
Branching Conversations
Fork one conversation into two futures with session.clone() — both branches inherit the context, then diverge side by side.
Image Input
Photo Semantic Search
Drop in your photos, and search them by meaning — "food on a table", "someone smiling" — without one pixel leaving your device.
Live Camera Q&A
Point your camera at anything, freeze a frame, and ask the on-device model about what it sees.
Draw & Guess
You sketch, Gemini Nano guesses: the model interprets your doodle after every stroke.
Image OCR
Extract raw text from images, receipts, or documents.
Image Description
Generate detailed alt-text or descriptions for images.
Explain a Meme
Understand the context, joke, or cultural reference in a meme.
Recipe from Fridge
Take a photo of your fridge contents and get recipe ideas.
Image Categorization
Categorize photos for auto-organization.
Screenshot → UI Code
Drop a UI screenshot and the multimodal model rebuilds it as HTML — rendered live next to the original.
Audio Input
Lecture Study Kit
Drop a lecture recording: transcript, key-point notes, a Q&A chat over the content, and auto-generated flashcards.
Audio Transcription
Transcribe spoken audio into accurate text.
Meeting Notes Extractor
Listen to meeting audio and generate action items.
Audio Summarization
Get a quick summary of a long voice note.
Tools Calling
Tool Calling: Smart Home
Real function calling: tell the model what you want and watch it invoke JavaScript tools that drive a live dashboard.
CSV Q&A
Paste a CSV and ask questions in plain English — answers come back structured, with the reasoning shown.
Structured JSON Output
Force the model to output a strictly formatted JSON object.
Extract Entities (NER)
Extract people, locations, and organizations into JSON.
Mix-and-Match
The Omnibox
One input, every API: whatever you paste is intent-routed to the right tool — translate, summarize, proofread, rewrite, or answer.
Polyglot Chat
Two people, two languages, one conversation — every message is detected and translated both ways as it is sent.
Localization QA
Scan a locale file for strings left in the wrong language, then fix the stragglers with one click.
Universal Inbox
A five-API pipeline: detect each message's language, translate it, triage it into folders, digest the inbox, and draft replies in the sender's language.
Image + Audio Query
Ask a question about an image using your voice.
Receipt to JSON
Extract line items from a receipt image into structured JSON data.