Multiverse unveils CompactifAI for offline mobile AI

Multiverse Computing has launched a mobile application, CompactifAI App, that runs advanced AI models directly on phones and tablets without an internet connection.

The software is aimed at professionals and organisations dealing with patchy connectivity or strict rules on where data can be processed. When more computing power is needed, users can switch from on-device models to cloud-based models via an API.

On-device focus

CompactifAI App lets users download AI models to a mobile device and run them locally, avoiding the need to send prompts or files to a remote service. According to Multiverse, keeping processing on the device keeps data local and gives users control over how information is handled.

The launch comes as more organisations test AI tools for document drafting, search, summarisation, and decision support while weighing privacy and security constraints. In healthcare and legal services, sensitive data requirements can limit the use of general-purpose cloud AI services. Defence, manufacturing, and field operations teams also often work in environments with limited or intermittent connectivity.

Compression method

The app relies on Multiverse's CompactifAI compression technology, which the company says uses quantum-inspired mathematics and can compress models by up to 95%.

Multiverse says the method keeps precision within a 2% to 3% margin, compared with what it describes as an industry standard of 20% to 30% accuracy loss at similar compression rates. It says this enables models that would otherwise require cloud infrastructure to run on standard mobile devices.

The app also includes a feature that routes queries between models, using on-device models for lighter tasks and API-based models for more complex reasoning. Multiverse frames this as a way to balance speed with access to larger models when required.

Use cases

Multiverse is pitching the app to mobile professionals and teams working away from reliable networks, as well as organisations with data sovereignty requirements that require data to remain within specific jurisdictions or internal systems.

Offline execution may also appeal to companies that want predictable operating conditions for certain workflows. Local processing avoids interruptions tied to network outages and reduces reliance on external providers for day-to-day use, although it can shift responsibility for device management, model distribution, and policy enforcement to the customer.

Hybrid approaches are becoming more common as AI products spread across organisations. Some tasks suit small, fast models that can run locally, while others require larger models typically hosted in data centres. Routing between the two can reduce cost and latency for routine work while still providing access to more advanced models when needed.

Enrique Lizaso, co-founder and CEO of Multiverse Computing, said the launch addresses long-standing trade-offs in mobile AI deployments.

"Running advanced AI locally has historically required compromising on model size or performance," said Enrique Lizaso, co-founder and CEO of Multiverse Computing. "What we're demonstrating with the CompactifAI App is that efficiency and intelligence are not mutually exclusive. Sophisticated reasoning models can now be deployed without the overhead of cloud-scale infrastructure, enabling powerful systems to operate directly on-device."

Broader context

Multiverse has been promoting compressed models as a way to reduce computing and energy demands for AI workloads. It recently released a compressed open-source model called HyperNova, which it says demonstrated "frontier-level reasoning" with lower compute, memory, and energy requirements.

CompactifAI App extends that strategy into a mobile software product, rather than a model release alone. It also reflects a broader shift in the AI market, with vendors increasingly offering tools that run at the edge-on personal devices and on-site systems-alongside cloud services.

Multiverse is headquartered in Donostia, Spain, with offices in the United States, Canada, and across Europe. The company says it serves more than 100 customers, including Iberdrola, Bosch, and the Bank of Canada, and is positioning CompactifAI App as a field-ready product for teams that need AI access when networks are limited and when information cannot leave a device.