Walking in Bevy

A 3D multiplayer game built with Bevy. Players run around as donuts (or cubes), jump, and swap models on the fly — with physics-based movement and networked multiplayer.

• Networked multiplayer — client/server architecture powered by lightyear with client-side prediction, server reconciliation, and visual interpolation
• Local split-screen — two players share a keyboard with automatic viewport layout
• Physics-based movement & jumping — powered by Avian3d + bevy-tnua character controller
• Swappable player models — toggle between a Blender-exported donut (glTF) and a procedural cube at runtime
• Input mapping — clean action-based input via leafwing-input-manager

• Rust
• Git LFS (for binary assets)

A server is deployed at imperius.janxyz.de:5000 — running cargo run --bin client connects there by default. Override with the SERVER_ADDR environment variable to point at a local or different server instead:

git lfs pull   # fetch binary assets if not already present

# connect to the deployed server
cargo run --bin client

# or run everything locally
# Terminal 1: start the server
cargo run --bin server

# Terminal 2: start a client pointed at localhost
SERVER_ADDR=127.0.0.1:5000 cargo run --bin client

Each client controls one player. The server spawns players at different positions with different colors. Your own player uses client-side prediction; other players are smoothly interpolated from the server state.

cargo run

Two players share a keyboard with split-screen viewports.

cargo test

cargo fmt --check
cargo clippy -- -D warnings

The server can be built as a container image and deployed to Kubernetes. The image is packaged from a prebuilt binary (no compilation happens inside Docker), so make docker-build only produces a runnable image on a Linux x86_64 host — do this in CI or on a Linux x86_64 machine:

make docker-build   # builds walking-in-bevy-server:latest
make docker-run      # runs it locally with UDP 5000 published
make k8s-render      # renders k8s/ manifests via kustomize (no cluster required)
make k8s-deploy       # kubectl apply -k k8s/ against your current kube context

.github/workflows/server-build-and-deploy.yml builds the server binary natively on the GitHub Actions runner (ubuntu-latest, linux/x86_64 — matching the cluster architecture, so no cross-compilation is needed) using Swatinem/rust-cache for fast incremental builds, packages it into the container image, pushes it to GitHub Container Registry, and deploys to Kubernetes automatically on every push to main that touches server-related files. It requires a KUBECONFIG repository secret (base64-encoded kubeconfig) to be able to deploy.

• Blender 4 Donut Tutorial
• Blender Character Design Playlist
• Chris Biscardi's Bevy Channel