FUTO Swipe

Fast, private, open-source swipe typing for Android keyboards

Unclaimed

Added Jul 2026

swipe.futo.tech

Open Source AiFreeopen-source-ai

Visit Futo Swipe

Share: X in LinkedIn

Add your screenshot here

Image or video shown in this spot

Get Verified · $10 lifetime

Follow Futo Swipe

What is Futo Swipe?

FUTO Swipe is an open-source family of AI models and algorithms for fast, accurate swipe typing that runs fully on-device with no privacy concerns. Built for FUTO Keyboard, an offline Android keyboard app, it uses three specialized models—an encoder, context language model, and language-specific decoder—to achieve over 96% top-4 accuracy. Designed for developers and users who want privacy-respecting mobile typing without reliance on cloud services or proprietary libraries.

Explore Futo Swipe

Need help implementing FUTO Swipe?

Find verified specialists who work with FUTO Swipe

Browse specialists

Key Features of Futo Swipe

On-device swipe typing with no cloud dependency
Three-model architecture (encoder, ContextLM, decoder)
Quick, accurate predictions with beam search
Open-source inference library
Permissively licensed models with attribution
1 million swipe dataset available on HuggingFace
Layout-agnostic encoder for multi-language support
Minimal footprint (2.5M total parameters)
QWERTY English decoder with industry-matching accuracy
Integrated with FUTO Keyboard Android app

Who Should Use Futo Swipe?

Mobile typing on privacy-conscious Android devices

Building custom keyboard applications

Training swipe models for new languages or layouts

Offline keyboard functionality without internet dependency

Research in mobile input methods and swipe typing systems

Futo Swipe: Pros & Cons

✓Pros

Fully offline and privacy-preserving
Open-source models and inference library
Very small model footprint suitable for mobile devices
Achieves ~96% accuracy matching major tech keyboards
Publicly available 1 million swipe training dataset
Permissive MIT license for datasets
Layout and language-agnostic architecture options
Community contribution opportunities for validation data
No reliance on proprietary or unlicensed libraries

✕Cons

Currently only has QWERTY English decoder; other languages/layouts require additional training data
Serverside demo has higher latency than production on-device version
Requires swipe typing data for training language-specific decoders
Error rate varies heavily depending on benchmark and real-world conditions

Frequently Asked Questions about Futo Swipe

What is FUTO Swipe?

FUTO Swipe is an open-source family of models and algorithms for fast, accurate swipe typing that runs fully on-device. It was developed primarily for FUTO Keyboard, an offline Android keyboard app, but the models and inference library are available for the broader community to use.

How does FUTO Swipe ensure privacy?

All processing happens on-device with no cloud dependency, meaning your typing data never leaves your phone. Unlike privacy-invasive keyboard apps, FUTO Swipe does not send data to servers.

What accuracy does FUTO Swipe achieve?

With all three models and a beam width of 300, FUTO Swipe achieves a top-4 fail rate of only ~4% on its test set, or below 1% when ignoring out-of-vocabulary cases. This matches the accuracy of major tech keyboards.

How large are the FUTO Swipe models?

The total footprint is only 2,494,767 parameters (1,364,271 active parameters), making them extremely lightweight for mobile devices. The encoder is 635,140 parameters, the decoder is 304,155 parameters, and the ContextLM is 1.5 million.

How do I use FUTO Swipe?

You can use it today in FUTO Keyboard v0.1.29 or later on Android. Alternatively, you can download the models and build your own applications using the open-source inference library.