The primary purpose of this API is to provide word predictions and auto-correct suggestions to text entry interfaces. The API supports input via:

Fixed letters - Client provides the exact letters a user has typed thus far.
Tap observations - Client provides a sequence of (x, y) tap(s) over a client-defined virtual keyboard.
Explicit distributions - Client provides a sequence of distributions over possible characters.

The API also supports a variety of other basic natural language processing (NLP) tasks, e.g. finding the most likely words in a sentence.

The documentation is organized into the following sections:

Basic tasks - A good place to start to test out the API, e.g. returning random words or finding a sentence's most likely words.
Fixed predictions - Predict words or characters based on fixed letters.
Future fixed predictions - Predict words or characters based on all possible next inputs. This is for advanced developers working to reduce latency in their interface.
Tap predictions - Word predictions and auto-corrections based on taps over a client-defined virtual keyboard. This includes defining or updating the keyboard topology.
Future tap predictions - Word predictions and auto-corrections based on taps over a client-defined virtual keyboard and possible next inputs based on key centers. This is for advanced developers working to reduce latency in their interface.
Distribution predictions - Word predictions and auto-corrections based on distribution(s) over possible characters.
NLP tasks - Natural language processing tasks, e.g. calculating the perplexity of some text. This is primarily intended for researchers.
Model configurations - For clients that need to specify details of the recognition algorithm or language models used.

If you use this API in your research, please cite this paper:

@inproceedings{vertanen_velocitap,
  author       = {Keith Vertanen and Haythem Memmi and Justin Emge and Shyam Reyal and Per Ola Kristensson},
  title        = {VelociTap: investigating fast mobile text entry using sentence-based decoding of touchscreen keyboard input},
  booktitle    = {CHI '15: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems},
  location     = {Seoul, Korea},
  year         = {2015},
  pages        = {659--668},
}

API version history:

Version	Date	Description
2.7	Jul 2024	Added future caching tap recognition endpoint.
2.6	Aug 2023	Added support for ambiguous keyboards and disambiguation of noisy input.
2.5	Jan 2021	Added distribution predictions.
2.4	Dec 2020	Deletion of keyboards after one hour of inactivity. Reorganization of documentation.
2.3	Sep 2020	Support for different decoder configurations and language models.
2.2	Sep 2020	Added ability to add keyboard, transform keyboard, and make predictions based on a tap sequence.
2.1	Apr 2020	Added future caching endpoints to mitigate latency impacts to user interfaces.
2.0	Mar 2020	Support for prefix, suffix, and substring in random word endpoint. Bug fixes to sort. Addition of character language model. Added log probability results to word prediction results.
1.0	Jun 2019	Initial version supporting basic natural language processing tasks. Designed to support mobile app development in a summer school for high school girls.

This material is based upon work supported by the National Science Foundation under Grant No. 1750193. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

document.write(MAIN_TITLE + ", v" + VERSION);