FUTO Swipe
Open-source AI swipe typing models and library for accurate, on-device word prediction.
| What is it | Open-source AI swipe typing models and library for accurate, on-device word prediction. |
|---|---|
| Pricing | Unknown |
| Platform | API |
| API | Yes |
| Best for | integrating swipe typing into apps, building privacy-focused keyboards |
| Domain registered | 2025 |
Data updated Aug. 1, 2026
What does FUTO Swipe do?
FUTO Swipe is a family of open-source AI models and a C++ library designed for accurate swipe typing on mobile devices. It powers FUTO Keyboard, a fully offline Android keyboard app, but the models are available for anyone to build with. The system takes a user's swipe path — the curve traced by a finger across the keyboard — and predicts the intended word, all running on-device with no internet connection needed. This means your typing data never leaves your phone.
The architecture uses three complementary models. The Encoder is layout- and language-agnostic, handling general predictions. The ContextLM is a tiny language model that filters out nonsensical word choices based on the sentence so far. The Decoder is layout- and language-specific, achieving top-tier accuracy — currently only available for QWERTY English. Combined with a beam search of width 300, the system achieves a top-4 fail rate of about 4% (under 1% ignoring out-of-vocabulary words). All models are small: the largest is 1.5 million parameters, and the entire stack runs in milliseconds even on low-end hardware. The C++ library (swipe-library) handles inference, decoding, and beam search, making it straightforward to integrate into any application.
FUTO Swipe is ideal for developers building custom keyboards, accessibility tools, or any text input system that values privacy and performance. It's already used in FUTO Keyboard, but the open models invite experimentation — from swipe typing in VR to using a laptop trackpad. The models are released under the FUTO Model License (attribution required), and the library is GPL. Researchers and hobbyists can download the dataset of 1 million swipes (MIT licensed) from HuggingFace to train their own models.
Key features
What makes it stand outWho is FUTO Swipe for?
Who benefits most from this toolTrust & presence
Alternatives in Developer Tools
Bito builds AI developer tools for teams working with complex codebases. At the Bito’s core is AI Architect. It’s a codebase intelligence server that creates a dynamic knowledge graph of your codebase. It learns how systems actually fit together, instead of treating them as disconnected files. That intelligence powers AI coding workflows and code reviews across coding agents, pull requests, and IDEs. Use Bito’s AI Architect via MCP or deploy AI Code Reviews directly in Git.
Device-native AI foundation models that run on phones, laptops, and cars — fine-tune and deploy locally.
AI coding assistant for JetBrains IDEs — get intelligent autocomplete and code suggestions directly in your editor
Fine-tune small language models via AI coding agents — describe the task in natural language, get a deployable model back.
Platform for sharing, discovering, and running machine learning models, datasets, and AI apps.
Lightweight open AI models from Google — run text generation, coding, and multimodal tasks on devices from phones to workstations
AI-powered deployment platform that automatically scales your applications with intelligent resource allocation
Download pre-optimized ML models, develop on cloud devices, and deploy apps on Qualcomm hardware
Similar tools
Open-source browser app for interacting with multiple AI language models via chat, notebook, and comparison modes
AI Figma plugin that provides instant design feedback, user persona simulation, and automated usability audits.
Visual AI builder for production mobile apps — describe your idea, edit visually or in code, deploy instantly.
AI-powered spreadsheet tool that automatically generates Python code from your spreadsheet edits and formulas.
AI design feedback tool for Figma — get instant audits, persona testing, and user feedback directly on your canvas.
AI assistant that adds ChatGPT-powered suggestions, search, and coding help to any text field on your computer.