Run AI Models on Your Phone Without the Cloud
Running large AI models usually means sending your data to someone else’s servers. Multiverse Computing, a Spanish startup, wants to change that. The company has launched CompactifAI, a chat app that runs compressed AI models directly on your device. It also released a self-serve API portal for developers and businesses. The technology uses quantum-inspired compression to shrink models from OpenAI, Meta, DeepSeek, and Mistral AI into versions small enough to run locally, without a cloud connection.
What Multiverse Computing Offers
- CompactifAI app runs a compressed model called Gilda locally and offline on mobile devices
- If the device lacks enough RAM, the app automatically routes to cloud-based models via API
- A new self-serve API portal gives developers direct access to compressed models
- Models from OpenAI, Meta, DeepSeek, and Mistral AI have been compressed
- The routing system between local and cloud is handled by a layer called “Ash Nazg”
Why Compressed AI Models Matter Now
Private credit defaults recently hit 9.2%, the highest rate in years. VC firm Lux Capital warned AI-dependent companies to get their compute commitments in writing, because a handshake is not enough when financial instability hits the supply chain.
Running models on your own hardware eliminates that counterparty risk entirely. No cloud provider outage. No surprise pricing changes. No data leaving your device.
Multiverse Computing’s compression technology is quantum-inspired, meaning it applies mathematical techniques from quantum computing to reduce model size while preserving performance.
The Limits of On-Device AI in 2026
There is a catch. The CompactifAI app needs enough RAM and storage on the device to run locally. Many older iPhones and budget Android phones cannot handle it. When the hardware falls short, the app falls back to cloud-based models, which defeats the privacy advantage.
According to Sensor Tower data, the app had fewer than 5,000 downloads last month. Consumer adoption is not the primary goal, though. Multiverse is targeting businesses through the API portal, where enterprises can integrate compressed models into their own products without relying on AWS Marketplace or similar middlemen.
On-Device AI as a Business Strategy
The real play here is for companies that want AI capabilities without external dependencies. A healthcare provider running a compressed model on local servers keeps patient data in-house. A financial firm running inference on its own hardware avoids sending sensitive data to third-party clouds.
As AI models continue to shrink through techniques like quantization and compression, the gap between cloud and on-device performance will narrow. Multiverse Computing is betting that gap closes fast enough for local AI to become the default for privacy-conscious organizations.
