What Is FUTO Swipe and Should You Switch Your Android Keyboard?

What Is FUTO Swipe?

FUTO Swipe is a family of open model weights and algorithms for swipe typing on mobile devices, developed by FUTO Holdings. It is the engine behind the swipe-to-type feature in FUTO Keyboard, a privacy-first Android keyboard that does not request internet access.

FUTO released the swipe models publicly in June 2026 alongside a dataset of 1 million swipes (MIT-licensed, hosted on HuggingFace) and a C++ inference library on GitLab. The project is distinct from FUTO Keyboard itself — developers can download the model weights and build their own swipe typing applications under the FUTO Model Weights License 1.0.

On the official demo page at swipe.futo.tech, you can try the swipe engine in a browser. In production, the models run entirely on-device with much lower latency than the server-side demo suggests.

Who Should Use FUTO Keyboard?

• Privacy-conscious Android users who want a keyboard that never phones home. FUTO Keyboard does not request internet permission and processes all input locally.
• People who type in English (QWERTY) — this is the primary supported language and layout right now. Other languages and layouts are limited or in development.
• Users willing to accept tradeoffs in exchange for privacy. Swipe accuracy is good but not best-in-class, and some features you expect from Gboard or SwiftKey are still missing.
• Developers who want to build on open swipe models — the weights, dataset, and C++ library are all available for integration into other projects with attribution.

FUTO Keyboard is not ideal if you rely on multilingual typing with seamless language switching, GIF search, stickers, or cloud-based suggestions. It is also Android-only with no iOS version planned or announced.

How FUTO Swipe Works Under the Hood

FUTO Swipe uses a three-model architecture that separates concerns across different stages of swipe prediction:

• Encoder model — A universal model that is layout-agnostic and language-agnostic. It makes the initial swipe-to-word predictions based on the touch path alone. FUTO describes its accuracy as sufficient for general use but says it "does not offer leading accuracy."
• ContextLM — A very small language model trained for a single language. It filters out nonsensical word predictions by considering the preceding words in the sentence. This model requires only text data for training, not swipe data.
• Decoder — A language-specific and layout-specific model that learns the peculiarities of a given keyboard layout. This is where the system achieves what FUTO calls "leading accuracy," but it requires swipe typing data for the target language and layout.

The C++ inference library handles dictionary-based search and word prediction on top of the model outputs. According to FUTO, the entire training pipeline never required more than one workstation GPU, keeping development costs low.

FUTO Keyboard vs Gboard vs SwiftKey

FUTO Keyboard competes directly with Google's Gboard and Microsoft's SwiftKey, the two dominant swipe keyboards on Android. The core differentiator is privacy: FUTO Keyboard is the only option that is fully offline by design.

All three keyboards support swipe typing, autocorrect, and word predictions. Gboard and SwiftKey have years of refinement behind their swipe engines and generally offer higher raw accuracy, especially in less common languages. They also support multiple languages simultaneously, cloud backup of your personal dictionary, GIF search, stickers, and voice typing — features FUTO Keyboard either lacks or handles differently (voice input is offline but less polished).

The tradeoff is that Gboard and SwiftKey process keystrokes through cloud services and collect usage data under Google's and Microsoft's respective privacy policies. If your keystrokes never leave your device, FUTO argues, they cannot be logged, analyzed, or leaked.

Privacy, License, and What 'Open' Means Here

FUTO describes the swipe models as "open," but it is important to understand the license terms. The model weights are released under the FUTO Model Weights License 1.0, a custom license — not MIT, Apache, or any OSI-approved open-source license.

Key terms of the license:

• You may use the weights for any purpose, including commercial products and services.
• You must display a visible notice to end users stating the product is powered by "FUTO Swipe" technology — in settings, an about screen, or equivalent.
• Derivative models must include a copy of the license and a notice stating they are derived from FUTO Swipe.
• The license includes a patent retaliation clause: if you claim the weights infringe a patent, your license terminates.
• Inference code, source code, and the swipe dataset are governed by separate licenses. The dataset of 1 million swipes is MIT-licensed.

This is a permissive but attribution-required license, not a standard open-source license. The C++ inference library is hosted on GitLab under a separate license not detailed on the swipe project page.

Limitations and What's Still Missing

FUTO Keyboard is actively developed and still in an early stage. FUTO's own site warns that users "may experience missing features or languages." Concrete limitations as of version 0.1.29 include:

• Android only — no iOS version exists, and FUTO has not announced plans for one.
• Limited language support — English QWERTY is the primary supported configuration. The architecture supports other languages and layouts, but trained decoder models for them are not yet available.
• Accuracy gap — FUTO acknowledges the base encoder does not offer top-tier accuracy. The decoder improves this significantly, but swipe recognition overall is less mature than Gboard or SwiftKey, which have trained on orders of magnitude more swipe data.
• No cloud features — no dictionary sync across devices, no cloud backup, no GIF or sticker search. This is by design but worth knowing before you switch.
• Paid license model — the $11.99 one-time license is optional but the purchase prompt appears in the app. Gboard and SwiftKey are entirely free (funded by data collection and ecosystem lock-in).

At a glance

FUTO Keyboard vs Gboard vs Microsoft SwiftKey at a glance.

FAQ

Is FUTO Keyboard free?

FUTO Keyboard is free to download from the Google Play Store, F-Droid, or GitHub. The app prompts users to purchase an optional one-time $11.99 license to support development. There are no ads and no recurring subscription fees.

Does FUTO Keyboard work on iPhone?

No. FUTO Keyboard is Android-only as of June 2026. FUTO has not announced any plans for an iOS version, and Apple's keyboard extension APIs may make a fully offline implementation more difficult on iOS.

How accurate is FUTO Swipe compared to Gboard?

FUTO acknowledges its base Encoder model does not offer top-tier accuracy. The Decoder model improves results significantly for supported languages and layouts, but overall swipe recognition is less mature than Gboard, which has trained on years of real-world swipe data from billions of users. For common English words on QWERTY, accuracy is solid. For rare words, multiple languages, or non-QWERTY layouts, Gboard and SwiftKey currently perform better.

Related reading

ToolBistro Radar

Sources

• FUTO Swipe official project page
• FUTO Keyboard official site
• FUTO Model Weights License 1.0
• FUTO Keyboard on Google Play