neuron-hooks

Hook registry and lifecycle middleware for neuron operators

Overview

neuron-hooks provides the Hook trait and HookRegistry for attaching pre- and post-execution middleware to operators. Hooks receive lifecycle events (before a turn, after a turn, on error) and can observe, mutate, or short-circuit execution.

Hooks compose cleanly: register multiple hooks in a registry and they execute in order.

Built into the neuron ecosystem: neuron-hook-security provides ready-made security hooks for redaction and exfiltration detection.

Usage

[dependencies]
neuron-hooks = "0.4"

Implementing a custom hook

use neuron_hooks::{Hook, HookContext, HookError};
use async_trait::async_trait;

pub struct LoggingHook;

#[async_trait]
impl Hook for LoggingHook {
    async fn before_turn(&self, ctx: &HookContext) -> Result<(), HookError> {
        println!("Starting turn for operator: {}", ctx.operator_id());
        Ok(())
    }

    async fn after_turn(&self, ctx: &HookContext) -> Result<(), HookError> {
        println!("Turn completed");
        Ok(())
    }
}

Registering hooks

use neuron_hooks::HookRegistry;

let mut registry = HookRegistry::new();
registry.register(LoggingHook);

Part of the neuron workspace

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