middleware

Middleware module for agent lifecycle hooks and cross-cutting concerns.

This module provides an extensible middleware system for intercepting and augmenting agent execution. Middleware can be used for:

Logging - Track agent activity and debug execution
Metrics - Record performance data and usage statistics
Validation - Validate inputs or outputs
Error handling - Implement custom error recovery strategies
Tracing - Add distributed tracing context

Examples

Using the built-in logging middleware:

import { createAgent } from '@providerprotocol/agents';
import { logging } from '@providerprotocol/agents/middleware';
import { anthropic } from '@providerprotocol/ai/anthropic';

const agent = createAgent({
  model: anthropic('claude-sonnet-4-20250514'),
  middleware: [logging({ level: 'debug' })],
});

Creating a custom middleware:

import type { Middleware } from '@providerprotocol/agents/middleware';

const metricsMiddleware: Middleware = {
  name: 'metrics',
  async before(context) {
    context.metadata.set('startTime', performance.now());
    return context;
  },
  async after(context, result) {
    const duration = performance.now() - (context.metadata.get('startTime') as number);
    metrics.recordLatency('agent.execution', duration);
    metrics.recordCount('agent.tokens', result.turn.usage?.totalTokens ?? 0);
    return result;
  },
};

Functions

logging

References

LoggingOptions

Re-exports LoggingOptions

Middleware

Re-exports Middleware

MiddlewareContext

Re-exports MiddlewareContext