xain_ai is no longer actively maintained. Parts of the code have been copied to xayn_discovery_engine and will be further developed there.

Xayn-AI

Models

To download the models use the download_data.sh script.

Tools

In order to generate the code that contains the metadata of the assets, you will need to install gomplate.

Build

To build the library you just need to run cargo build in the root of the project.

To generate the dart ffi you need to run flutter pub get and flutter pub run ffigen in the directory bindings/dart.

To update the non-ffi auto generated dart code after changes to dart code you need to run flutter pub run build_runner build.

The project provides a Makefile.toml that can be run with cargo make. You can install cargo make with:

cargo install --version 0.35.0 cargo-make

All of the above can be automatically done by cargo make build.

To build libraries for mobile targets you can use:

cargo make build-mobile

On Linux this will only build Android libraries, while on Mac it will build for both Android and iOS.

To build the library for web you can use:

cargo make build-web

or

DISABLE_WASM_THREADS=1 cargo make build-web

to build a version which doesn't require the browser to support SharedArrayBuffer and Atomics.

Android

To build for Android the following targets are needed:

rustup target add \
  aarch64-linux-android \
  armv7-linux-androideabi \
  x86_64-linux-android \
  i686-linux-android

Also cargo-ndk is needed:

cargo install --version 2.4.1 cargo-ndk

You also need to install android ndk and might need to set the ANDROID_NDK_HOME variable. If you install the android ndk through AndroidStudio you would need to set ANDROID_NDK_HOME to something like ~/Android/Sdk/ndk/<ndk-version> which for example could be /home/user/Android/Sdk/ndk/22.1.7171670/.

iOS

To build for iOS the following targets are needed:

rustup target add \
  aarch64-apple-ios \
  x86_64-apple-ios

WASM

Prerequisites

• wasm-pack

cargo install --version 0.10.1 wasm-pack

• rustup target add wasm32-unknown-unknown
• nodejs (only if you want to run the tests on nodejs)
• yarn

All wasm-pack commands below are to be run in the directory xayn-ai-ffi-wasm/.

Building the WASM module

wasm-pack build

Running the WASM test

Browser

wasm-pack test --firefox --chrome --safari --headless

Note:

In Safari you first need to enable the WebDriver support via safaridriver --enable, before you can run tests.

nodejs

wasm-pack test --node -- --no-default-features --features=node

Flutter example

Running on a mobile device/emulator

cargo make build-mobile
cd bindings/dart/example
flutter run

Running in Chrome

cargo make build-web
cargo make serve-web
# Then open http://localhost:8000/ in a browser.

flutter run can not be used as it doesn't set the right headers, and even with the right headers it doesn't provide the right files in the right way.

(At least for now flutter run can still be used with DISABLE_WASM_THREADS=1, this is not guaranteed in the future but can be useful as flutter run provides hot reloading and better debug-ability.)

Hint: There had been some cases where for some reason genesis.js wasn't updated when switching from DISABLE_WASM_THREADS=0 to DISABLE_WASM_THREADS=1 it's not clear what causes it as it's not reproducible but it was observed by different dev. If that happens run cargo make clean-non-rust.

Running with a branch of the release repository

bindings/dart/example/pubspec.yaml

  xayn_ai_ffi_dart:
-   path: '../'
+   git:
+       url: git@github.com:xaynetwork/xayn_ai_release.git
+       ref: <branch>

bindings/dart/example/lib/data_provider/web.dart

- const _baseAssetUrl = 'assets';
+ const _baseAssetUrl = 'https://ai-assets.xaynet.dev';

Then use cargo make build-web and cargo make serve-web like above.

If flutter analyze fails there is a good chance that the local version and the used branch in the release repo are not compatible.

License

This repository contains code from other software in the following directories, licensed under their own particular licenses:

• rubert-tokenizer/*: Apache2 (LICENSE)

Xayn-ai and its components, unless otherwise stated, are licensed under

• AGPL-3.0 (LICENSE).