Jolla Launches AI Assistant for Privacy-Focused GenAI

Jolla's New Privacy-Focused AI Assistant

Jolla, formerly a mobile phone manufacturer, has now evolved into a privacy-focused AI company through its subsidiary, Venho.ai. They have recently unveiled an AI assistant designed as a truly private alternative to the large cloud-based AI systems that often collect user data.

Integration and Functionality

This AI assistant is built to seamlessly integrate with commonly used applications such as email, calendars, and social media platforms. It aims to provide users with a conversational interface capable of both retrieving information and executing tasks.

Users can expect functionalities like email and document summarization, meeting scheduling, social media feed filtering, and web searches. The assistant is designed to enhance productivity and streamline daily activities.

Dynamic AI Agents and Marketplace

The tool’s capabilities extend beyond core functions. It can dynamically create new AI agents as needed, provided the appropriate API keys are accessible.

Furthermore, Jolla plans to launch an AI agent marketplace next month, expanding the assistant’s utility through a wider range of specialized agents.

Personal Assistance and Data Security

Jolla positions the AI assistant as a valuable aid for shopping and personal organization. Users can leverage it for product research or create personal notes for future reference.

A key differentiator is the commitment to data privacy. Personal information remains secure, as it is not transmitted to large data-mining operations in the cloud.

Hardware and Software Synergy

The AI assistant software is not a standalone product. The Jolla and Venho team has concurrently developed dedicated AI hardware – previously described as “AI agents in a box” – to support this vision.

Local Processing and Data Management

Their strategy centers on utilizing smaller AI models that can be hosted locally. This approach handles much of the user’s data processing directly on their device.

Preprocessing techniques and a vector database are employed to unify data from linked accounts, accelerating query response times and enhancing the overall user experience.

Seamless Agent Switching

The AI assistant software is designed to orchestrate the back-end data processes and facilitate smooth transitions between different AI agents, adapting to the user’s evolving needs.

Navigating the Complexities of AI

At the Mobile World Congress (MWC) in Barcelona, Venho.ai co-founder Antti Saarnio and Jolla veteran Sami Pienimäki presented an exclusive preview of their new AI assistant technology. They propose a significant opportunity for a startup to engineer a decentralized AI operating system, poised to challenge established cloud providers as generative AI fundamentally alters software and its applications.

Their background in mobile hardware and operating system development, particularly with Sailfish, and Android app integration, provides a competitive advantage in this emerging field, they contend.

The Need for AI Assistance

“The current influx of information due to AI creates substantial cognitive strain,” states Saarnio. “We require tools to effectively filter and manage this data.” He envisions the AI assistant as a crucial instrument for reducing information overload, emphasizing user ownership: “This is your own tool – you maintain control.”

Introducing Mindy: A Customizable AI Companion

The AI assistant software, branded as Mindy – an avatar with a customizable appearance – will soon be available as a subscription service. It will be hosted on a private cloud infrastructure managed by Venho.ai.

During a demonstration, TechCrunch observed Saarnio interacting with the AI assistant on his laptop. He utilized a chat-style interface to submit queries, receiving responses in both text and spoken formats. Tasks such as checking emails, adding items to a to-do list, and scheduling meetings were successfully executed. Voice commands are also supported.

Performance and Optimization

Initial responses exhibited a slight delay, but subsequent interactions were quicker. Saarnio confirmed ongoing efforts to enhance speed and optimization, aiming for a response time of under one second.

A web search query, intended to demonstrate the AI’s information retrieval capabilities, initially failed. However, Saarnio clarified that this was due to a lack of internet connectivity, preventing access to the Google Search API.

Deployment Options: Cloud and Personal Hardware

Beyond its private cloud offering, the AI assistant will be accessible on the Mind2 Jolla AI device starting next month. This device has been shipping to early adopters since January.

Mind2 owners will have the option to self-host the AI assistant on their personal server-style hardware, eliminating the need for third-party cloud services, even Venho’s “private cloud.” This is particularly relevant for queries processed by the smaller AI models integrated within the device, including DeepSeek’s 1.5-billion parameter model and Meta’s Llama 1-billion parameter model.

Balancing Capability and Privacy

“The device can accurately understand and respond to queries, such as providing email information, without significant errors,” explains Saarnio. “However, it may struggle with large context windows, like processing extensive documents.”

More demanding tasks may necessitate utilizing larger language models (LLMs), which inherently involve privacy trade-offs as user data is subject to external terms and conditions. However, the conversational assistant interface allows users to define parameters, such as preventing the transmission of sensitive health information to LLMs, thereby tailoring the experience to their privacy preferences.

Pricing and Hardware Details

The AI Assistant is priced at $10 per month for the first 1,000 subscribers (following a 14-day free trial), with the standard price increasing to $20 per month. The Mind 2 device, featuring 16GB of RAM and 128GB of storage, currently retails for €699, with discounts available for early adopters.

Hardware Evolution and Manufacturing

Since its initial unveiling, the Mind2 device has increased in size. Pienimäki attributes this to enhanced heat management, incorporating a larger heatsink and additional cooling components alongside the processing hardware.

Final assembly takes place in Finland, at a former Nokia facility in Salo, potentially revitalizing the region’s technology sector. The Sailfish OS itself originated as a fork of an abandoned Nokia software project, developed by former employees.

Community Feedback and B2B Opportunities

Approximately 500 Mind2 devices have been shipped to early adopters, leveraging the established Sailfish enthusiast community. Saarnio highlights the valuable feedback and bug fixes received through a dedicated Discord community, recommending this approach to other startups.

Beyond direct sales, the team identifies a promising B2B opportunity. Telecommunications companies have expressed interest in the hardware as a home hub solution for families, described by Pienimäki as “a private Amazon Alexa.” This prompted the acceleration of Mindy’s development “to demonstrate the power of conversational AI,” adds Saarnio.