Crate bevy_llm 

bevy_llm (minimal): a thin bevy wrapper over the llm crate.

• re-exports llm chat/types so you don’t duplicate data models.
• streams deltas and tool-calls as bevy events.
• lets the llm provider manage history (via builder memory).
• never blocks the main thread: on native we spawn onto a tiny tokio runtime (no bevy pool blocking); on wasm we use bevy’s async pool, which yields to the browser/event loop.

api docs (types & traits): https://docs.rs/llm

• chat provider: llm::chat::ChatProvider
• message builder/roles: llm::chat::{ChatMessage, ChatRole, MessageType}
• streaming: llm::chat::{StreamResponse, StreamChoice, StreamDelta}
• tools / tool calls: llm::builder::FunctionBuilder, llm::chat::ToolChoice, llm::ToolCall

Structs

BevyLlmPlugin

bevy plugin: wires systems, events, resources. requires you to insert a Providers resource before/after adding the plugin. on native, also inserts a tiny tokio runtime resource by default.

ChatCompletedEvt

ChatDeltaEvt

ChatErrorEvt

ChatMessage

re-export the llm types so downstream code can use the same structs/enums. A single message in a chat conversation.

ChatRequest

insert this component to trigger a chat request for the session entity. the provider manages the history; you only provide the new messages.

ChatSession

attach this to an entity you want to chat with a provider.

ChatStarted

events emitted by the wrapper during/after chat.

ChatToolCallsEvt

FunctionBuilder

re-export the llm types so downstream code can use the same structs/enums. Builder for function tools

LLMBuilder

re-export the llm types so downstream code can use the same structs/enums. Builder for configuring and instantiating LLM providers.

Providers

a map of ready-to-use llm providers.

StreamChoice

re-export the llm types so downstream code can use the same structs/enums. Individual choice in a streaming response

StreamDelta

re-export the llm types so downstream code can use the same structs/enums. Delta content in a streaming response

StreamResponse

re-export the llm types so downstream code can use the same structs/enums. Stream response chunk that mimics OpenAI’s streaming response format

TokioRt

on native we keep a tiny tokio runtime to drive llm futures. we spawn onto this rt from compute tasks so neither the main thread nor bevy’s compute pools block.

ToolCall

re-export the llm types so downstream code can use the same structs/enums. Tool call represents a function call that an LLM wants to make. This is a standardized structure used across all providers.

Enums

ChatRole

re-export the llm types so downstream code can use the same structs/enums. Role of a participant in a chat conversation.

LLMBackend

re-export the llm types so downstream code can use the same structs/enums. Supported LLM backend providers.

LLMError

re-export the llm types so downstream code can use the same structs/enums. Error types that can occur when interacting with LLM providers.

LlmSet

system ordering so uis can run after we emit events

MessageType

re-export the llm types so downstream code can use the same structs/enums. The type of a message in a chat conversation.

StreamMsg

ToolChoice

re-export the llm types so downstream code can use the same structs/enums. Tool choice determines how the LLM uses available tools. The behavior is standardized across different LLM providers.

Traits

ChatProvider

re-export the llm types so downstream code can use the same structs/enums. Trait for providers that support chat-style interactions.

LLMProvider

re-export the llm types so downstream code can use the same structs/enums. Core trait that all LLM providers must implement, combining chat, completion and embedding capabilities into a unified interface

Functions

send_user_text

helper to enqueue a text user message on a session entity.