neuron-turn

Shared toolkit for building LLM providers and operators

Overview

neuron-turn is the shared toolkit that concrete providers (Anthropic, OpenAI, Ollama) and operators (ReAct, single-shot) build on top of. It provides:

• Provider trait — the async interface that every LLM integration implements
• Request / response types — TurnRequest, TurnResponse, ContentPart, ToolCall, etc.
• Context strategy types — ContextStrategy enum and its resolution logic (used by providers to window conversation history before sending to the model)
• Shared cost / token accounting — TokenUsage, Cost, DurationMs

Usage

[dependencies]
neuron-turn = "0.4"

Implementing a custom provider

use neuron_turn::{Provider, TurnRequest, TurnResponse};
use async_trait::async_trait;

pub struct MyProvider { /* ... */ }

#[async_trait]
impl Provider for MyProvider {
    async fn turn(&self, request: TurnRequest) -> Result<TurnResponse, neuron_turn::TurnError> {
        // call your LLM API here
        todo!()
    }
}

Part of the neuron workspace

neuron is a composable async agentic AI framework for Rust. See the book for architecture and guides.