Native server / reproducible source

Build Ai00 from source

Build the pinned upstream server for your own machine, then stage its Resources overlay for the 4D component. This is a native build, not a cross-compilation recipe. For released components, start with installation.

Prepare your machine

Install Git with Git LFS, Python 3.9+, Rust/rustup, and native build tools: Xcode Command Line Tools on macOS, or Visual Studio C++ Build Tools and the Windows SDK on Windows. Network access is required for source, LFS assets, crates, and the default ONNX Runtime build dependency.

RWKV inference needs a WebGPU-compatible GPU/backend and enough model memory. The 4D launcher does not support Linux or Windows ARM.

Build the pinned server

Run these commands for Apple Silicon; substitute your target from the table on Intel Mac or Windows, and use python instead of python3 on Windows.

git clone https://github.com/miyako/ai00.git
cd ai00
git lfs install
rustup toolchain install 1.91.0 --profile minimal
rustup target add --toolchain 1.91.0 aarch64-apple-darwin
python3 scripts/build-server.py --target aarch64-apple-darwin

The script builds upstream commit 54ddbd6818499b11bfd0469d2ea59577d8076c92 (release v0.7.1) with the locked dependencies and default features, then checks --help and --version. The upstream Cargo package reports 0.6.2; that is not the release-tag version. See the version and provenance reference.

Stage and run

The output in dist/<target>/ is a Resources overlay, not a standalone component. It contains the server, tokenizer, WebUI, licenses, provenance, checksums, and the Windows ONNX Runtime DLL where applicable.

For local 4D development on Apple Silicon:

cp -R dist/aarch64-apple-darwin/. Ai00/Resources/

Open the 4D project after staging the overlay, restore its declared component dependencies, and follow Use from 4D. For standalone use, edit .build/ai00_server/assets/configs/Config.toml to select a compatible .st model, then run from the upstream checkout:

cd .build/ai00_server
./target/aarch64-apple-darwin/release/ai00-server \
  --config assets/configs/Config.toml --ip 127.0.0.1 --port 8080

The upstream config enables authentication and defaults to port 65530; the command above explicitly overrides the listener. Set your own authentication settings rather than using upstream example secrets. Windows embedding users must retain assets/ort/onnxruntime.dll. Follow the CLI and model reference for model conversion and compatibility limits, and the first-request example once your model is ready.

Component publishing

Publishing requires successful native builds for all three targets, verified overlays, the existing tool4d component build, and Apple signing/notarization. The legacy converters are not rebuilt by the server workflow and have no recorded source revision. See the complete build reference and release pipeline before preparing a component release.