Skip to main content

Connected Mode

Connected mode delegates selected skill execution from server to client.

Use it for skills that depend on streamed entities or client-native task calls.

Enable on Server​

import { Server } from '@open-core/framework/server'
import { npcIntelligencePlugin } from '@open-core/npc-agents/server'

await Server.init({
mode: 'CORE',
plugins: [
npcIntelligencePlugin({
openRouter: {
model: 'openai/gpt-4o-mini',
},
}),
],
})

Enable on Client​

import { Client } from '@open-core/framework/client'
import { npcClient } from '@open-core/npc-agents/client'

await Client.init({
mode: 'CORE',
plugins: [npcClient()],
})

Transport Behavior​

  • Primary path: net wire transport
  • Fallback path: RPC caller when available

This is automatic when connected runtime pieces are installed.

Debug​

Enable transport debug:

setr OPENCORE_NPC_TRANSPORT_DEBUG 1

Useful log families:

  • DELEGATE_SEND, DELEGATE_OK, DELEGATE_FAIL
  • WIRE_SEND, WIRE_RESULT, WIRE_TIMEOUT, WIRE_FAILOVER

AI debug options on server:

export NPC_AI_DEBUG=1
export NPC_AI_DEBUG_LLM=1

Common Issues​

  • no_executor: no ready client executor available
  • missing_npc_netid: NPC has no network id for transport
  • ...timeout: request sent but no client response before timeout

If delegation works but behavior still stalls, validate observation payloads and destination values.