Test different variations of prompts, models, RAG, and custom code with feature-flag-like config called App Config

​
Adding App Config to your Agent

App config is a key-value object that can you pass into your LLM application for each request. Using this key-value object, you can control the behavior of your LLM application. Here’s an example of an app config that let’s us easily swap different OpenAI models: 
interface AppConfig {
  model: string;
}

const handler: Chat<unknown, AppConfig> = async ({
  userMessage,
  // 1. destructure app config from the request
  appConfig,
}): Promise<ChatHandlerResponse> => {
  const response = await openai.chat.completions.create({
    // 2. use app config in your application logic
    model: appConfig.model,
    messages: [{ role: "user", content: userMessage }],
  });
  return {
    message: response.choices[0].message.content,
  };
};

​
Client-Side Usage

From your external services, such as a frontend application, you can call your agent with the app config. Here’s an example of how you can call an agent with app config: 
await palico.agent.chat({
  name: "math_agent",
  userMessage: "What's 2+2?",
  appConfig: {
    openaiModel: "gpt-4o-mini",
  },
});
Learn more about calling your agent from external services using our client SDKs.

​
Development Use-Cases

LLM Development involves testing lots of different ideas and as such you should look to build a modular application. You can use app config to then drive the behavior of your application. Some common use-cases for app config are:
  • Trying different LLM models
  • Trying different prompts
  • Trying different RAG versions
  • Trying different call-chaining strategies
  • Trying different LLM architectures
Additional, our evaluation framework is built around using app config to test different variations of your LLM application.

​
Production Use-Cases

You can use app config like feature-flags to test different variations of your application in production or easily roll-back changes if needed.