neuron

Composable async agentic AI framework for Rust

Overview

neuron is the umbrella re-export crate for the neuron workspace. It provides a single dependency entry point for the most common combination of crates, controlled by feature flags.

The framework is built on a 6-layer protocol model where each layer is a Rust trait:

layer0: Operator | StateStore | Environment | Orchestrator | Hook | Observable
         ↓           ↓             ↓               ↓          ↓        ↓
      operators   state        credentials    workflows   middleware  events

Every layer is independently replaceable. You can use the Anthropic provider with the OpenAI operator interface, swap in-memory state for filesystem state, plug in Vault secrets without changing your operator code, and so on.

Quick start

[dependencies]
neuron = { version = "0.4", features = ["op-react", "provider-anthropic", "state-memory", "env-local"] }
tokio = { version = "1", features = ["full"] }

use neuron::prelude::*;
use std::sync::Arc;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let provider = AnthropicProvider::from_env()?;
    let operator = ReactOperator::new(
        Arc::new(provider),
        Arc::new(ToolRegistry::new()),
        Arc::new(HookRegistry::new()),
    );
    let env = LocalEnv::new(Arc::new(EnvResolver));

    let input = OperatorInput::new("What is 2 + 2?");
    let output = operator.invoke(input, &env).await?;
    println!("{}", output.content.as_text().unwrap_or_default());
    Ok(())
}

Feature flags

Workspace crates

The workspace is split into focused crates you can depend on individually:

Core protocol

• layer0 — foundational protocol traits (no implementations)
• neuron-turn — Provider trait + LLM types
• neuron-tool — Tool + ToolRegistry
• neuron-context — context window strategies

Operators

• neuron-op-react — ReAct reasoning loop
• neuron-op-single-shot — single model call

Providers

• neuron-provider-anthropic — Claude
• neuron-provider-openai — GPT
• neuron-provider-ollama — Ollama

Orchestration

• neuron-orch-kit — wiring kit
• neuron-orch-local — in-process orchestrator

Hooks & security

• neuron-hooks — hook registry and middleware
• neuron-hook-security — redaction + DLP hooks

State

• neuron-state-memory — in-memory store
• neuron-state-fs — filesystem store

Environment & credentials

• neuron-env-local — local environment
• neuron-secret — secret resolution traits
• neuron-auth — authentication traits
• neuron-crypto — cryptographic traits

MCP

• neuron-mcp — Model Context Protocol bridge

License

neuron is dual-licensed under MIT and Apache 2.0.

Documentation

• Book (architecture + guides)
• API docs (docs.rs)
• GitHub