Pradeep Elankumaran has heard the privacy pitch before—mostly from companies who say one thing in their marketing and quite another in their terms of service. Which perhaps explains why he's spending so much time talking about what his new AI assistant doesn't do.
"We never pool user data," Elankumaran says of Jo, the personal AI he and co-founder Kevin Li launched in March. "Your machine is your machine."
It's a positioning that might have felt paranoid a few years ago. Now, with AI assistants reading emails, scanning calendars, and analyzing browsing habits, the question of where that information ends up doesn't sound quite so fringe.
Jo, a Menlo Park startup that emerged from Y Combinator's Spring 2026 batch, is making a straightforward bet: that enough people will pay $39 a month for an AI that stays, mostly, out of the cloud.
The Architecture of Trust
Here's how it works. Jo's Mac app lives on your computer, indexing files and browser history locally. When you ask it something that requires heavy processing—summarizing a long document, say, or researching flight options—it hands the task to a dedicated cloud machine. Not a server shared with hundreds of other users, the company insists, but an isolated instance spun up for you alone.
Raw messages, files, the everyday detritus of digital life? Those stay on the Mac by default. Jo sends only what's needed to answer your query, routing everything through what the company calls zero-data-retention endpoints from the big LLM providers: Claude, ChatGPT, Gemini, Grok, and a Chinese model called Kimi. Contracts with those providers, Jo says, prohibit training on user data.
Whether users will trust those assurances is another question. The company's privacy policy—last updated in late February—notes that conversation data sticks around for the life of your account plus 30 days. That's more transparent than some competitors, though perhaps not as airtight as the truly paranoid might prefer.
The product itself is a study in surface area. There's the Mac app, naturally, which serves as home base. But you can also ping Jo via Telegram, WhatsApp, or just by forwarding an email. It notices what you're browsing, clips articles for later, and delivers a morning briefing stitched together from weather forecasts, calendar appointments, and Gmail reminders.
A browser agent handles research tasks with anti-fingerprinting turned on. Background automations run on schedule. Jo drafts emails, though it won't send them without explicit approval—a wise choice given how spectacularly AI autocomplete can misfire.
The March launch outlined an ambitious roadmap: native mobile apps, "multiplayer assistants" for teams, trusted purchases, even grocery shopping. How much of that materializes remains to be seen. For now, it's a Mac-only product, and not just any Mac. You need Apple Silicon—M1 or newer. Intel chips and Windows machines are out of luck.
The Founder Track Record
Elankumaran and Li aren't first-time founders, which matters in a space where credibility is everything. Their 2011 Y Combinator company, Kicksend, was reportedly pulling in $150,000 monthly by 2013 with 30% month-over-month growth, at least according to a YC blog post from that era. (Revenue numbers from early-stage startups should always be taken with a grain of salt, but the trajectory suggests they knew how to build something people wanted.)
They went on to launch Farmstead, a grocery delivery startup that went through YC's Summer 2016 batch. That pedigree gives Jo a credibility boost, though the current team appears lean—Y Combinator lists two people, though LinkedIn hints at a slightly larger crew. No funding beyond YC's standard check has been disclosed.
A Crowded Room
Jo's timing is tricky. The privacy-focused AI assistant space has gotten noisy fast.
Perplexity introduced its "Personal Computer" agent earlier this year, a Mac mini that runs around the clock, plugged into local apps and Perplexity's servers. Anuma launched in late April with encrypted memory and support for multiple frontier models. Local-first options like Jan and Reyn pitch similar benefits without the cloud dependency at all.
And then there's the elephant in the room—or rather, the apple. At WWDC last week, Apple previewed macOS 27 "Golden Gate," featuring revamped Apple Intelligence and a redesigned Siri slated for this fall. OS-level AI with Apple's privacy reputation and ecosystem lock-in could make third-party assistants feel redundant overnight.
That's the existential risk for any company building on top of someone else's platform. You're always one WWDC keynote away from obsolescence.
The Privacy Premium
Jo is live now at askjo.ai, no waitlist required. The Standard plan runs $39 monthly after a three-day trial—no credit card needed upfront—and covers the account holder plus four family members. You get a cloud machine that "wakes up when needed," 1 GB for conversations and what the company calls "knowledge mapping," plus the ability to switch between AI models.
A pricier Solo plan, cheekily branded "LET ME COOK," costs $99 per month plus usage fees. It offers a beefier always-on machine, 5 GB storage, and access to the latest high-powered reasoning models on a pay-as-you-go basis.
Whether that pricing holds up depends on how much users value privacy over convenience, and whether Jo can stay ahead of competitors who are moving just as fast. It's a narrow window.
For now, Elankumaran and Li are betting that enough people want their AI assistant to know them well—but not too well, and definitely not in the same database as everyone else. It's a bet that's either prescient or a few years too early. Sometimes in this business, it's hard to tell the difference.
