Skip to content

Runnable examples ​

Choose the smallest example that answers your question. Commands below run from a repository checkout and use the committed lockfile.

sh
git clone https://github.com/briansunter/bevy-agent.git
cd bevy-agent

A complete environment without a window ​

sh
cargo run -p bevy_agent_runner --example counter --locked
cargo test -p bevy_agent_runner --example counter --locked

You should see: the snapshot/replay success message, then two passing tests. Read Getting started for the complete app and Testing for what its assertions prove.

Counter source · Distributed in the runner crate.

Scripted gameplay ​

sh
cargo run -p sample_platformer --example agent_play --locked

You should see: JSON describing the initial state, and a terminal response if the scripted sequence ends the episode. This example runs movement and jump actions through the Rust API without a window.

Platformer source · Repository only.

A game controlled by another process ​

sh
cargo run -p sample_platformer --example remote_http --locked -- 127.0.0.1:4000 --artifact-dir ./artifacts

You should see: a server that stays running. In another terminal, follow HTTP and the CLI or Python. HTTP uses /rpc, health checks use /health, and WebSocket clients use /ws.

A visible game and screenshots ​

sh
cargo run -p sample_platformer --features visual --example remote_http_visual --locked -- 127.0.0.1:4000 --artifact-dir ./artifacts

You should see: a Bevy window and a running RPC server. This needs a working graphics environment and the platform libraries used by Bevy. It uses the same port as the headless server: stop that process first or choose a different port.

Follow Screenshots and rendering to choose software versus primary-window capture. A displayed frame and a controlled simulation tick are separate events.

A subprocess transport ​

sh
cargo run -p sample_platformer --example remote_stdio --locked

This process waits for newline-delimited JSON-RPC on stdin; it is not an interactive game prompt. Follow WebSocket and stdio for requests and Python subprocess control. Use HTTP/WebSocket when you need retained operation status and retry-key deduplication.

Which files are published? ​

The counter is packaged with bevy_agent_runner. The platformer, Python client, and agent skills are distributed through Git. Installing bevy_agent_cli gives you the agentctl client; start a compatible environment separately.

On the pinned personal Mac mini, use the guarded launchers described in Contributing.

Released under MIT or Apache-2.0.