Using LLM Calls

The op-demo-simple-chat app contains several chat reaktors that showcase how to call an LLM with Operaide aktors. The snippet below is SimpleChat01.reaktor.ts.

Simple chat reaktor

import {
    aktorAISettingProviderModel,
    aktorAICall,
    aktorPatchMessages,
    registerChatReaktorDefinition,
} from '@operaide/ai';
import { aktorSetting } from '@operaide/aktor';
import { z } from 'zod';

registerChatReaktorDefinition({
    reaktorDefinitionId: 'simple-chat-01',
    label: 'Antworte mit allen details und smilies',
    description: 'Mit smileis antworten',
    createReaktor({ messages }) {
        const systemMessage = aktorSetting(z.string(), 'Du antwortest sarkastisch', 'System Message');
        const providerModel = aktorAISettingProviderModel();
        return aktorAICall({
            messages: aktorPatchMessages({
                messages,
                system: systemMessage,
            }),
            providerModel,
        });
    },
});

Key components

• registerChatReaktorDefinition wires the reaktor into Operaide using the chat-specific schema.
• aktorSetting lets operators change the system prompt at runtime without redeploying.
• aktorPatchMessages injects the system message before sending the conversation to the model.
• aktorAICall performs the actual LLM invocation using the provider configured through aktorAISettingProviderModel.

Sample request

After deploying the app, send a request similar to SimpleChat01.http:

POST http://localhost:8811/api/v1/aktor/simple-chat-01/run
Content-Type: application/json

{
  "messages": [
    {
      "role": "user",
      "content": "Kannst du mir etwas über Operaide erzählen?"
    }
  ]
}

The response contains the LLM answer and follows the standard chat output schema.