AI Model & Product
Google AI Edge
Also known as: AI Edge, Google on-device AI
Google AI Edge is Google's collection of tools and runtimes for running AI models directly on devices such as phones, browsers and embedded hardware rather than in the cloud. It includes components like LiteRT for on-device inference, MediaPipe tasks for common vision, audio and text jobs, conversion tooling for models built in other frameworks, and APIs for on-device generative models. Developers use it to build features that work offline, respond quickly and keep data local.
What it is
Google AI Edge is an umbrella for Google's on-device machine learning stack, spanning model conversion, optimisation, runtime execution and debugging tools. It covers classic tasks such as image classification, object detection, speech and text handling, as well as running compact generative models on device. It targets Android, iOS, web and embedded targets, with hardware acceleration where the device supports it.
Why it matters
As more AI moves to the device, a growing share of user questions can be answered without a server request, a search query or a page visit. That shifts measurement, because on-device answers leave little trace in analytics, and it raises the value of being the source that on-device systems retrieve from when they do go online. It also affects product teams, since low latency and offline capability change what experiences are possible in apps and stores.
How it works
Developers take a trained model, convert and quantise it for the target runtime, then call it from an app through an inference API, often combining it with local retrieval over app data. Teams profile memory, latency and accuracy trade offs across device tiers, and fall back to cloud models for heavier tasks. Marketing and SEO teams engage indirectly by keeping factual brand information structured, current and retrievable, and by testing how on-device assistants describe their products.
When it applies
Applies when building mobile or embedded AI features, when privacy or connectivity constraints rule out cloud inference, or when assessing how much of your audience journey is being handled on device.
Examples
- A retail app runs on-device image recognition so shoppers can point a camera at a product and get a match without a network connection.
- A field service app uses a compact on-device model to summarise engineer notes while working in buildings with no signal.
- A browser based tool runs a small model locally to redact personal data before anything is uploaded.
How it is measured
- Proportion of inference requests served on device versus cloud
- Median and p95 inference latency per device tier
- Model size and peak memory use after quantisation
- Offline task completion rate and associated support ticket volume
Insights on Google AI Edge
Related terms in AI Model & Product
- Agents APIAn Agents API is a programming interface for building applications where a model plans, calls tools and completes multi-step tasks rather than returning a single reply. The term is most often used for OpenAI's agent building interfaces, though other vendors offer equivalents. Teams use it to connect models to search, internal data and actions such as booking, updating records or generating reports.
- AI assistantAn AI assistant is a software product that uses a language model to hold a conversation, answer questions and carry out tasks on a user's behalf. It sits on top of one or more underlying models and adds an interface, memory, tools and safety controls. Examples include ChatGPT, Google Gemini, Microsoft Copilot, Claude and voice assistants such as Siri and Alexa.
- AI transcriptionAI transcription is the automatic conversion of spoken audio into written text using speech recognition models. Modern systems often add speaker labels, punctuation, timestamps and language detection, and can feed the output into summarisation or search. It is also called automatic speech recognition or ASR.
- BingBing is Microsooft's web search engine, available at bing.com and integrated into Microsoft Copilot and other Microsoft products. It maintains its own crawler, Bingbot, and its own index, and provides Bing Webmaster Tools for site owners. Its index and APIs have also supplied results to third-party search products and AI assistants.
- ChatGPT VoiceChatGPT Voice is the spoken conversation mode in OpenAI's ChatGPT apps, which lets you talk to the assistant and hear its replies read aloud. Advanced Voice Mode processes speech directly, so exchanges feel closer to a phone call and you can interrupt mid-answer. People use it hands free on mobile, and it is also available on desktop.
- Claude CodeClaude Code is Anthropic's agentic coding tool that works directly with a codebase, reading files, proposing and making edits, and running commands with permission. It runs in the terminal and in supported development environments, driven by natural language instructions rather than manual file by file editing. Marketing and growth teams use it for technical SEO, structured data and data tasks that would otherwise wait for engineering.
Primary research · August 2026
How ChatGPT Shortlists Software Brands
An audit across 10 categories and 60 buying questions. I recorded what ChatGPT reads, throws away and links to when a buyer asks it which software to buy, and what that decides.
- 60
- Questions asked
- 10
- Software markets
- 2,680
- Results read
- 367
- Links shown
Free download
Get the full report
35 pages · PDF · 536 KB. Enter your details and it downloads straight away.