Jolla, a former mobile manufacturer, has become a privacy-centric AI business through sister startup Venho.ai.
The AI Assistant is designed to integrate with apps such as email, calendars, and social media accounts, providing a conversational power tool that allows users to view information but allows them to take actions on their behalf. This means summarizing emails and documents, booking meetings, filtering social media feeds, running web searches, and more.
However, they say the tool can spin up new AI agents on the spot to further expand the utility as long as the required API keys are available. AI Agent Marketplace is also part of the plan and is scheduled to be released next month.
Elsewhere, they lean towards the idea of AI assistants as shopping aids and personal aide notes. For example, users can investigate potential purchases and send AI notes about things they want to remember on their behalf.
AI Assistant software is not a standalone. The Jolla with Venho team has been developing their own AI hardware over the past year. This is a technology that we previously called “AI Agent in A Box.”
Their approach relies on leveraging small AI models that can be hosted locally to manage many data tasks for users, along with vector databases that integrate data from linked accounts to speed up user experience at the query level.
IoComing AI Assistant Software is intended to automatically automate switching between different AI agents, sitting on top of all this backend data orchestration as required by the user.
Cut off noise
Co-founders Antti Saarnio and Jolla veteran Sami Pienimäki, who will provide TechCrunch with an exclusive preview of the new AI assistant technology at Barcelona’s Mobile World Congress (MWC) trade show.
They point to a long history of mobile hardware and OS development (using Sailfish) and integration of Android apps.
“We’re experiencing a massive cognitive load at the moment because of AI, and I think we need this help to filter things,” says Saarnio. It suggests that AI assistants can become an essential tool for cutting noise. “It’s your own tool — you own that tool.”
The AI Assistant Software is an avatar with a female look and feel, on display under the brand name Mindy that allows users to customize to their own preferences, but is imminent as a subscription service hosted in a private cloud operated by Venho.ai.
In that demo, TechCrunch saw Saarnio interacting with an AI assistant running on a laptop. I entered the query through a chat style interface and retrieved the reply as text and speech. Ask him what emails he has. Email related tasks have been added to your to-do list. Let AI book a meeting slot. (Note: You can also talk to your assistant about your questions.)
There was a bit of a pause before the initial response came back, but the response below was faster. Saarnio also noted that it is working on further acceleration and optimization with the aim of lowering the response rate to one second.
The web search query he tried to demo – asks AI to get information about the company – could not return what is being asked for for the data, but Saarnio later said this was because the laptop is not connected to the internet and the Google Search API didn’t work.
In addition to providing AI assistants via their own private cloud, the tool will be available on Mind2 Jolla AI devices next month. The team has been shipping to early recruits starting in January.
This means that Mind2 owners can self-host AI assistants on this personal server-style hardware. This abandons the need to use third-party cloud services to retain data (even the “private cloud” proposed by Venho). Certainly, when it comes to queries that can be run on small AI models within the device, including Deepseek’s 1.5 billion parameter model and Meta’s Llama 1B parameter model per Saarnio.
“You can understand the right answer, the email you have, what content is, and it won’t make any major mistakes. But giving a huge context window (such as uploading a large document to a query) can get confusing,” he says.
More processing-enhancing queries require the technology to leverage large-scale language models (LLM). Here, privacy promises must break down as the user’s data is exposed to other people’s T&Cs. But having a conversation assistant sitting on top of the system means that users can tell the AI how it works. For example, telling your assistant not to send health information to LLM to further customize your experience to your comfort zone.
The price for the AI Assistant (the first 1,000 users) Early Bird is set at $10 per month (after a 14-day free trial), but the final price is about $20m per Saarnio. (Those who buy Jolla hardware also have 16GB RAM, and the full price tag on the 128GB Memory Mind 2 device costs 699 euros, but the team is still offering discounts to early adopters.)
Since we last approached with Yorla hardware, this year has grown the now-shipped gadgets. According to Pienimäki, that increase in footprint (now the size of a small, thick paperback) depends primarily on thermal management requirements.
He points out that the final assembly of the kit is taking place in Finland at Saro’s former Nokia facility. At the very least, it retains the fascinating possibility that the efforts of reborn startups may have an opportunity to revive the glory of former Finnish technology. (Reminder: The Sailfish OS was developed as a fork of abandoned Nokia software project by former staff.)
So far, around 500 Mind 2 devices have been shipped to early adopters. Saarnio says they have got useful feedback and help fix bugs from these early adopters through the discrepancies community. “I recommend it to any startup,” he says of the approach.
And while they are in the business of selling kits and related services themselves, they also believe there are interesting B2B opportunities that are just starting out.
Certainly, Saanio says they are those who thought the hardware could offer a home hub-style solution that could work well for multiple families living under one room.
