Production Deployment

Checklist

Before deploying:

• Use RedisStateStore for persistent state
• Configure PgVectorMemoryStore for long-term memory
• Set budget limits on all agents
• Configure autonomy policy for sensitive tools
• Enable OpenTelemetry export
• Set up error alerting
• Test with TestHarness and MockLLM

State Persistence

Always use Redis in production:

import { RedisStateStore } from '@ahzan-agentforge/core';

const stateStore = new RedisStateStore({
  url: process.env.REDIS_URL!,
  prefix: 'prod:agentforge:',
  ttl: 86400 * 30,  // 30 days
});

Budget Governance

Set sensible defaults:

budget: {
  maxCostUsd: 2.00,
  maxTokens: 200_000,
  warnThreshold: 0.8,
},

Observability

Export to your monitoring stack:

observability: {
  enabled: true,
  exporter: 'otlp',
  endpoint: process.env.OTEL_ENDPOINT!,
},

Error Handling

hooks: {
  onError: async (error, source) => {
    await alerting.send({
      severity: source === 'framework' ? 'critical' : 'warning',
      message: `AgentForge ${source} error: ${error.message}`,
    });
  },
},

Scaling

• Use queue-backed execution (BullMQ) for high throughput
• Separate agent processes from your web server
• Scale Redis for state and queue workloads
• Use connection pooling for PostgreSQL

Next Steps

• Cost Control — managing costs
• Human-in-the-Loop — approval workflows