Loading SaaS profile...
FUTO Swipe
Fast, accurate offline swipe typing system and open models.
What it does
Decodes continuous touch swipe trajectories into accurate word predictions using lightweight neural network models. Runs dictionary-constrained beam searches via a high-performance C++ inference library. Operates completely offline on mobile devices to protect user typing privacy. Provides layout-agnostic and language-agnostic encoder features to support custom keyboards.
Why it works
Utilizes an extremely small parameter footprint allowing models to execute on low-end mobile devices in milliseconds. Maintains a local-first offline architecture that prevents sensitive typing data from transmitting to the cloud. Combines a universal layout encoder with a layout-specific decoder to achieve high prediction accuracy. Releases models under permissive licenses with an open-source inference library and dataset.
Category: AI & Automation
Team Size: 11-25
Visit Website
As of 2026
AI & Automation
Revenue: 0-$100k ARR
Fast, accurate offline swipe typing system and open models.
Overview
FUTO Swipe is an open-source family of neural models and algorithms designed to enable fast, accurate mobile swipe typing completely offline. Developed primarily for FUTO Keyboard, it solves the privacy and licensing limitations traditionally locked behind big tech or proprietary keyboard applications. The framework features an ultra-lightweight architecture using an encoder, context language model, and a C++ inference library to run efficiently on low-end devices.
Team size: 11-25
Funding status: Funded
Revenue source: Open Source / FUTO Grant Backing
Customer type: B2B2C
Pricing: Free (Open Source)
Tech stack: C++, Python, PyTorch, TCN, Android NDK
Platform: Android • Web
Integrations: FUTO Keyboard, Android Input Method Framework, C++ Inference Library, HuggingFace
Founder story
FUTO Swipe was developed by the FUTO organization, founded by Louis Rossmann and based in Austin, Texas. FUTO focuses on building independent, user-respecting technology and open-source tools to give control back to individuals, creating FUTO Swipe to break the industry monopoly on high-performance mobile swipe typing.
What it does
- Decodes continuous touch swipe trajectories into accurate word predictions using lightweight neural network models
- Runs dictionary-constrained beam searches via a high-performance C++ inference library
- Operates completely offline on mobile devices to protect user typing privacy
- Provides layout-agnostic and language-agnostic encoder features to support custom keyboards.
Who it's for
- Android Developers
- Privacy-Conscious Users
- Mobile Engineers
- Open Source Contributors
Why it works
- Utilizes an extremely small parameter footprint allowing models to execute on low-end mobile devices in milliseconds
- Maintains a local-first offline architecture that prevents sensitive typing data from transmitting to the cloud
- Combines a universal layout encoder with a layout-specific decoder to achieve high prediction accuracy
- Releases models under permissive licenses with an open-source inference library and dataset.
Growth strategies
- Open Source Hugging Face Releases
- Integration with FUTO Keyboard App
- Community Data Collection Campaigns
- Word- of-Mouth in Privacy and Android Communities
Alternatives
Comparison overview
- FUTO Swipe provides a completely open-source, offline neural swipe decoding toolkit and lightweight C++ inference library.
- Proprietary alternatives like Gboard or SwiftKey rely on cloud telemetry and closed-source commercial prediction algorithms.