Jolla, the erstwhile mobile maker turned privacy-centric AI business — via sister startup, Venho.ai — has taken the wraps off an AI assistant that’s touted as a “fully private” alternative to letting data-mining cloud giants crawl all over your personal information.
The AI assistant is designed to integrate with apps like email, calendar and social media accounts to provide the user with a conversational power tool that can surface information but also perform actions on the user’s behalf. This means stuff like summarizing emails and documents, booking meetings, filtering social media feeds and performing web look-ups.
But they say the tool will also be able to spin up new AI agents on the fly to further expand utility so long as the necessary API keys are available to it. An AI agent marketplace is also part of the plan, slated for launch next month.
Elsewhere, they’re leaning into the idea of the AI assistant as a shopping aid and personal aide memoire — meaning the user could, for example, get help to research potential purchases or send the AI notes on things it wants it to remember on their behalf — storing the info for future recall while the user can be safe in the knowledge that these personal snippets are not being fed back to some data mining giant’s business empire in the cloud.
The AI assistant software does not stand alone; the “Jolla with Venho” team has been developing proprietary AI hardware over the past year — tech we’ve previously referred to as ‘AI agents in a box’ — which is also aimed at bringing to life this vision of highly personalized AI convenience without privacy trade-offs.
Their approach relies upon leveraging smaller AI models that can be locally hosted to manage many of the data tasks for the user, along with pre-processing and a vector data-base that unifies data from their linked accounts to speed up the user experience at the query level.
The iocoming AI assistant software is intended to sit atop all this backend data orchestration and amoothly automate switching between different AI agents, as user demands require.
Cutting through the noise
Giving TechCrunch an exclusive preview of the incoming AI assistant tech here at the Mobile World Congress (MWC) trade show in Barcelona, co-founder Antti Saarnio and Jolla veteran Sami Pienimäki argue there’s an unfolding startup opportunity to develop — essentially — a decentralized AI operating system to disrupt cloud giants as generative AI continues to rewrite the rules around software and how we use it.
They point to their long history of mobile hardware and OS development (with Sailfish) and Android app integrations — arguing this gives them a leg up in such a race.
“There’s massive cognitive load coming at the moment because of AI. And I think we need this assistance to filter things,” says Saarnio, suggesting the AI assistant will become an essential tool to cut through the noise. “It’s your own tool — you own that tool.”
The AI assistant software — which is being showcased under the brand name Mindy, an avatar with a female look and feel, but which the user will be able to customize to their own tastes — is launching imminently as a subscription service hosted on a private cloud operated by Venho.ai.
In a demo of it, TechCrunch saw Saarnio interacting with the AI assistant running on his laptop — typing queries through a chat-style interface and getting replies back as text and speech. Such as asking what emails he has; having email-related tasks added to a to-do list; and getting the AI to book a meeting slot. (Note: It’s also possible to speak queries to the assistant.)
There was initially a bit of a pause before the first response came back but the following responses were faster. Saarnio also noted that they’re working on further acceleration and optimization — with the aim of getting the response rate down to one second.
A web search query he tried to demo — asking the AI to get info about a company — failed to return the sought for data but Saarnio subsequently said this was because the laptop was not connected to the Internet, meaning the Google Search API did not function.
In addition to offering the AI assistant via its own private cloud the tool will next month be made available on the Mind2 Jolla AI device — which the team has been shipping to early adopters since January.
This means Mind2 owners will be able to self host the AI assistant on this personal server-style hardware — jettisoning the need to use any third party cloud service to hold their data (even the “private cloud” that’s pitched by Venho). Certainly when it comes to queries that can be run on the small AI models housed inside the device — which include DeepSeek’s 1.5BN parameter model and Meta’s Llama 1B parameter model, per Saarnio.
“It’s capable of understanding and giving you proper answers — what emails you have got, what is the content — then it doesn’t make massive mistakes. But if you give it a huge context window [such as uploading a large document to query] it gets confused,” he says, explaining when the device might need to head to someone else’s cloud for answers.
More processing-intense queries can require the tech to tap into large language models (LLMs) — where privacy promises must fall away since the user’s data will be exposed to someone else’s T&Cs. But by having a conversational assistant sitting atop the system, the suggestion is that users will be able to instruct the AI how they want it to work for them — such as, say, telling the assistant never to send any health information to LLMs — further customizing the experience to their comfort zone.
Early bird pricing for the AI Assistant (first 1,000 users) is set at $10 per month (after a 14-day free trial) — but the final price will be around $20pm, per Saarnio. (For those buying the Jolla hardware too the 16GB RAM, 128GB memory Mind 2 device has a full price-tag of €699, though the team is still offering a discount price for early adopters.)
In the year since we last got up close and personal with Jolla’s hardware the now-shipping gadget has grown in size. Per Pienimäki, the increase in its footprint (it’s now about the size of a small, chunky paperback) is mostly down to heat management requirements — with the box now packing a larger heatsink and other heat management components alongside processing hardware.
Final assembly of the kit is being done in Finland, at a former Nokia facility in Salo, he also notes, which at least holds up an enticing possibility that the reborn startup’s efforts might have the chance to revive former Finnish tech glories. (Reminder: The Sailfish OS was developed as a fork of an abandoned Nokia software project by ex staffers.)
So far, around 500 Mind 2 device have been shipped to early adopters — with the team tapping into interest from the Sailfish enthusiast community. Saarnio says they’re also getting useful feedback and help with bug fixes from these early adopters through a Discord community. “I would recommend that to any startup,” he notes of the approach.
And while they’re in the business of selling the kit and associated services themselves, they also believe there’s an interesting B2B opportunity that’s just getting started.
Indeed, Saarnio says it was some telcos who saw potential for the hardware to offer a home hub-style solution that could work well for multiple family members living under one room — in other words “It’s like a private Amazon Alexa,” as Pienimäki puts it — which in turn encouraged the team to accelerate development of Mindy “to show the power of conversation of AI”, Saarnio adds.
