In partnership with
Have you tried running AI offline on your phone before?
ChatGPT Plus costs ₹1,999/month ($20), Google AI Pro is ₹1,950/month ($20) and any decent transcription tool like WisprFlow adds another ₹320 ($12) on top of that.
That's ₹4,269/month, or roughly $42 for three tools most of you are juggling separately.
Google just shipped all of it for free on your phone, and it works with your Wi-Fi off.
It's called Gemma 4, and the app is Google AI Edge Gallery. Here's how to set it up and what it can actually do.
But before that, some catchup on AI this week:
NEWS NEWS NEWS
TOOLS of the Week
1. Base44 Superagents: A 24/7 autonomous AI agent you build by describing what you want in plain English. It can connect to 200+ tools, then runs in the background scanning your inbox, following up on leads, monitoring prices, and dropping summaries into WhatsApp or Telegram.
2. Lessie: An AI people-search agent that understands intent rather than just keywords. Describe who you're looking for in plain language and it scans LinkedIn, podcasts, databases, and social platforms to surface ranked, scored contacts ready for outreach.
3. ThinkTask: A project management tool with ChatGPT baked directly into the workflow. Paste in meeting notes or context and it auto-creates tasks, assigns them based on team history and skills, and generates performance reports without anyone having to fill anything in manually.
The Setup
Step 1: Download the app
Search Google AI Edge Gallery on the App Store (iPhone) or Play Store (Android). It's free.
Open it, skip the welcome screen, accept terms, and you'll see four tabs: AI Chat, Ask Image, Audio Scribe, and Agent Skills.
Step 2: Download the model
Before anything works, you download the AI model separately inside the app. Think of the app as the kitchen and the model as the chef.
Go with the 4B version as it is smarter, about 3.6GB, and you only need to download it once.
4 things Gemma 4 can do on your phone
1. Write emails and content with no internet at all
Pull down your control center, turn off Wi-Fi and mobile data, hence disconnecting the phone completely.
Open AI Chat, type anything and hit send.
The response comes back instantly (yup, no loading spinner) because the AI running it is a file on your device, not a server somewhere.
I asked it to write an email to Anthropic and got four complete drafts.
My take: Most of us will never turn our internet off on purpose. But "offline" here is really just shorthand for “your text never leaves your phone.”
For anything sensitive like a personal situation you're thinking through or a message to a difficult client, this is the only free AI that processes it entirely locally.
2. Turn a voice memo into a polished draft
Open Audio Scribe, tap the mic and just talk.
The model transcribes it word for word and it converts the mess into a properly structured draft with a subject line, clean paragraphs, and professional tone.
It directly competes with WisprFlow but there’s a difference.
WisprFlow is system-wide, it follows you into Gmail, Slack, Notion, every app, and formats on the fly as you dictate.
Audio Scribe is a standalone tab inside one app. But for normal transcription use cases, it is a solid free alternative.
My take: I've tested a bunch of transcription tools and the consistent problem is that they all transcribe first and leave the editing to you. The fact that Audio Scribe skips straight to a structured output is a meaningful UX difference.
3. Run your morning routine through Agent Skills
This one is different from the first two.
Agent Skills means the AI picks the right tool on its own and goes and does something.
Some examples I found useful:
Map lookup: I said "my friend wants to meet somewhere near Gateway of India, can you show me where that is" and it loaded an Interactive Map skill on its own, pinned the location, and rendered a working map inside the chat window.
Live lookup: "My friend keeps bringing up the 2026 Oscars and I haven't caught up, can you give me a quick summary of what happened?”
The model calls query-wikipedia/index.html on its own, pulls the 98th Academy Awards summary, returns the key winners and shows details in plain language.
Multimodal quirk: I sent a photo of my breakfast plate and said "what vibe am I giving off, match it with music.”
The model calls mood-music, identifies the vibe from the image as low-energy, and renders a full music player inside the app playing lo-fi.
My take: The agent behaviour where it decides which skill to use without being told is what got to me. This is what people mean when they say "agentic AI" and it's running entirely on a phone.
You can also manage skills from a dedicated screen, toggle them on and off, and build custom ones with a plus button.
My Take
Gemma 4 E2B i.e. the model doing everything above is also the foundation for Gemini Nano 4, which Google is shipping baked into new Android flagships later this year.
Galaxy S26 already has Nano 3. Nano 4 is next, and whatever phone you upgrade to in the second half of 2026 will very likely run it natively without you downloading anything.
Which means the Edge Gallery app is less a standalone product and more a developer preview of your next phone's OS, just packaged so anyone can try it now.
Google has been making on-device AI promises for a few years and most of it was underwhelming. But Gemma 4 is the first version where I've looked at the output and thought this is worth using day-to-day.
Whether Nano 4 holds up on real hardware at scale is still an open question, but the app you can download today is a reasonable preview of what the answer looks like.
Reply to this with what you'd most want an on-device AI to help you with. I'll share the top use cases next week.
Until next time,
Vaibhav 🤝
If you read till here, you might find this interesting
#AD 1
Hiring in 8 countries shouldn't require 8 different processes
This guide from Deel breaks down how to build one global hiring system. You’ll learn about assessment frameworks that scale, how to do headcount planning across regions, and even intake processes that work everywhere. As HR pros know, hiring in one country is hard enough. So let this free global hiring guide give you the tools you need to avoid global hiring headaches.
#AD 2
Are You Ready to Actually Retire?
Knowing when to retire means knowing what it costs, how long your money needs to last, and where the income comes from. When to Retire: A Quick and Easy Planning Guide helps investors with $1,000,000 or more work through all of it.
