bevy_llm 🤖

bevy llm plugin (native + wasm). minimal wrapper over the llm crate that:

• re-exports llm chat/types, so you don’t duplicate models

• streams assistant deltas and tool-calls as Bevy events

• keeps history inside the provider (optional sliding-window memory)

• never blocks the main thread (tiny Tokio RT on native; async pool on wasm)

install crate

cargo add bevy_llm
# or in Cargo.toml: bevy_llm = "0.2"

capabilities

• Bevy plugin with non-blocking async chat

• Structured streaming with coalesced deltas (~60hz or >=64 chars)

• Fallback to one-shot chat when streaming unsupported

• Tool-calls surfaced via ChatToolCallsEvt

• Provider-managed memory with sliding_window_memory

• Multiple providers via Providers + optional ChatSession.key

• Native + wasm (wasm uses gloo-net)

• Helper send_user_text() API

• Built-in UI widgets

• Persisted conversation storage

• Additional backends convenience builders

usage

use bevy::prelude::*;
use bevy_llm::{
    BevyLlmPlugin, Providers, ChatSession, ChatDeltaEvt, ChatCompletedEvt,
    LLMBackend, LLMBuilder, send_user_text,
};

fn main() {
    App::new()
        .add_plugins(DefaultPlugins)
        .add_plugins(BevyLlmPlugin)
        .add_systems(Startup, setup)
        // read chat events after the plugin drains its inbox
        .add_systems(Update, on_events.after(bevy_llm::LlmSet::Drain))
        .run();
}

fn setup(mut commands: Commands) {
    // OpenAI-compatible backend (point base_url at your server)
    let provider = LLMBuilder::new()
        .backend(LLMBackend::OpenAI)
        .base_url("https://api.openai.com/v1/responses")
        .model("gpt-5")
        .sliding_window_memory(16)
        .build().expect("provider");

    commands.insert_resource(Providers::new(provider.into()));

    // start a streaming chat session and send a message
    let session = commands.spawn(ChatSession { key: None, stream: true }).id();
    send_user_text(&mut commands, session, "hello from bevy_llm!");
}

fn on_events(
    mut deltas: EventReader<ChatDeltaEvt>,
    mut done: EventReader<ChatCompletedEvt>,
){
    for e in deltas.read() { println!("delta: {}", e.text); }
    for e in done.read()   { println!("final: {:?}", e.final_text); }
}

examples

• chat: simple text streaming UI with base url / key / model fields
• tool: demonstrates parsing JSON-as-text and handling ChatToolCallsEvt

run (native):

# optional env for examples
export OPENAI_API_KEY=sk-...
export LLM_BASE_URL=https://api.openai.com
export LLM_MODEL=gpt-5

cargo run --example chat
cargo run --example tool

wasm is supported; integrate with your preferred bundler and target wasm32-unknown-unknown.

backends

Configured via the upstream llm crate. This plugin works great with OpenAI‑compatible servers (set base_url to your /v1/responses endpoint). Additional convenience builders may land later.

compatible bevy versions

license

licensed under either of

• Apache License, Version 2.0 (http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (http://opensource.org/licenses/MIT)

at your option.

contribution

unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.