Installing riddlg¶

riddlg ships as a self-contained native binary for macOS Apple Silicon and Linux x86_64, with the llama.cpp inference libraries bundled — there is nothing else to install.

Hardware Requirements¶

The AI commands (gen riddl, gen code --fill, and the AI features of serve/mcp) run a large language model locally and require a GPU. Without one, riddlg refuses to run AI commands (pass --allow-cpu to override, but CPU inference is impractically slow). Everything else — validate, gen docs, gen api, plain gen code — runs fine on any machine.

Recommended: Apple Silicon M5 Pro with 64 GB

For the best results, we recommend an Apple Silicon Mac with an M5 Pro (or better) and 64 GB of unified memory — for example, a Mac Mini M5 Pro. The default AI model (~23 GB) is sized for exactly this class of machine and runs fully GPU-accelerated via Metal.

Hardware	Experience
Apple Silicon, 64 GB unified memory (M5 Pro or better)	Recommended — default model at full speed
Apple Silicon, 48 GB	Works with the default model
Apple Silicon, 16–32 GB	Use a smaller model
Linux + NVIDIA GPU	Use the `cuda` build; ~24 GB+ VRAM for the default model, smaller models below that
Linux + AMD/Intel GPU	Use the `vulkan` build with a model sized to your VRAM
No GPU	Non-AI commands only (or `--allow-cpu` if you are very patient)

Homebrew¶

The easiest installation on macOS (Apple Silicon) and Linux (x86_64):

brew install ossuminc/tap/riddlg

To upgrade later:

brew upgrade riddlg

Linux GPU users

The Homebrew Linux build is CPU-only. To use an NVIDIA, AMD, or Intel GPU on Linux, install the cuda or vulkan variant via direct download instead.

Direct Download¶

Release archives are published for each version. Replace 0.3.0 with the version you want:

macOS (Apple Silicon)Linux (NVIDIA GPU)Linux (AMD / Intel / other GPU)Linux (CPU only)

curl -fLO https://storage.googleapis.com/synapify-releases/riddlg/0.3.0/riddlg-0.3.0-Darwin-arm64.tar.gz
tar -xzf riddlg-0.3.0-Darwin-arm64.tar.gz
export PATH="$PATH:$(pwd)/riddlg-0.3.0-Darwin-arm64/bin"

GPU acceleration uses Metal and works out of the box.

The cuda variant is GPU-accelerated on NVIDIA cards (Turing/GTX 16 & RTX 20 series or newer). It bundles the CUDA runtime — you only need the NVIDIA driver installed.

curl -fLO https://storage.googleapis.com/synapify-releases/riddlg/0.3.0/riddlg-0.3.0-Linux-x86_64-cuda.tar.gz
tar -xzf riddlg-0.3.0-Linux-x86_64-cuda.tar.gz
export PATH="$PATH:$(pwd)/riddlg-0.3.0-Linux-x86_64-cuda/bin"

The vulkan variant runs on any GPU with a Vulkan driver (AMD, Intel, and NVIDIA alike).

curl -fLO https://storage.googleapis.com/synapify-releases/riddlg/0.3.0/riddlg-0.3.0-Linux-x86_64-vulkan.tar.gz
tar -xzf riddlg-0.3.0-Linux-x86_64-vulkan.tar.gz
export PATH="$PATH:$(pwd)/riddlg-0.3.0-Linux-x86_64-vulkan/bin"

For servers or machines without a GPU (non-AI commands, or CI use):

curl -fLO https://storage.googleapis.com/synapify-releases/riddlg/0.3.0/riddlg-0.3.0-Linux-x86_64.tar.gz
tar -xzf riddlg-0.3.0-Linux-x86_64.tar.gz
export PATH="$PATH:$(pwd)/riddlg-0.3.0-Linux-x86_64/bin"

For a permanent installation, add the export PATH=... line to your shell profile (.bashrc, .zshrc, etc.), or move the extracted directory somewhere like /opt first.

Linux Packages (deb / rpm)¶

CPU-only builds are also packaged for apt and yum/dnf. They install under /opt/riddlg and put riddlg on your PATH via /usr/bin/riddlg:

Debian / UbuntuRHEL / Fedora

curl -fLO https://storage.googleapis.com/synapify-releases/riddlg/0.3.0/riddlg_0.3.0_amd64.deb
sudo dpkg -i riddlg_0.3.0_amd64.deb

curl -fLO https://storage.googleapis.com/synapify-releases/riddlg/0.3.0/riddlg-0.3.0-1.x86_64.rpm
sudo rpm -i riddlg-0.3.0-1.x86_64.rpm

Verify Installation¶

riddlg version

should print the installed version (e.g. 0.3.0), and

riddlg info

prints build metadata plus the compute devices llama.cpp can see. On an Apple Silicon Mac you should see your Metal GPU listed; on Linux, your CUDA or Vulkan device. If it warns that no GPU was detected, the AI commands will refuse to run — check that you installed the variant matching your hardware.

There is no build-from-source option

Unlike riddlc, riddlg is proprietary (closed-source) software, so installation is via the prebuilt binaries above. The binary download is free; see Free and Pro.

Next Steps¶

Command Reference - Learn the available commands
AI Models - Download the default model ahead of time, or pick one sized for your hardware