`createLogger`

Problem

Debugging an LLM agent without logs is guesswork. But adding raw console.log statements everywhere creates noise in production and is hard to control. You end up either seeing too much or too little — and toggling it requires editing multiple files.

Solution

createLogger creates a structured logger you configure once. You control what gets logged (steps, tools, errors), what level is visible, and whether logging is on at all. In production, set enabled: false — nothing leaks. In development, turn on steps and tools to see exactly what the agent is doing.

Feature & Use-Case

Use createLogger when:

You want step-by-step visibility into your agent loop during development
You need to log tool calls with their input for debugging
You want production-safe logging — turn it off with one flag, no code changes
You are building a Node.js LangChain agent and need structured, level-gated output

Import


import { createLogger } from "llm-layer-engine";

Function Signature


function createLogger(config?: LoggerConfig): {
  step(step: number, message: string): void;
  tool(name: string, input: unknown): void;
  warn(message: string): void;
  error(err: unknown): void;
  info(message: string): void;
}

`LoggerConfig` Options


type LoggerConfig = {
  enabled?: boolean;   // Master switch. Default: false
  level?: LogLevel;    // "debug" | "info" | "warn" | "error". Default: "info"
  steps?: boolean;     // Log each agent step. Default: false
  tools?: boolean;     // Log tool calls with input. Default: false
  errors?: boolean;    // Log errors. Default: true
};

Log Level Order

debug → info → warn → error

Setting level: "warn" means only warn and error messages are shown. Setting level: "debug" shows everything.

Example — Development Setup


import { createLogger } from "llm-layer-engine";
 
const logger = createLogger({
  enabled: true,
  level: "debug",
  steps: true,
  tools: true,
  errors: true,
});
 
logger.step(1, "Starting agent run");
// → [STEP 1] Starting agent run
 
logger.tool("get_weather", { city: "Karachi" });
// → [TOOL] get_weather { city: "Karachi" }
 
logger.info("Loop completed");
// → [INFO] Loop completed
 
logger.warn("maxSteps reached without final answer");
// → [WARN] maxSteps reached without final answer
 
logger.error(new Error("Provider timeout"));
// → [ERROR] Error: Provider timeout

Example — Production Setup (Silent)


const logger = createLogger({
  enabled: false,
});
 
logger.step(1, "This will not print");
logger.tool("anything", {});
// → (no output)

Pass the same logger object everywhere. When enabled is false, every method is a no-op.

Example — Integrate Into a Custom Loop


import { createLogger, executeStep } from "llm-layer-engine";
 
const logger = createLogger({ enabled: true, level: "debug", steps: true, tools: true });
 
let step = 0;
while (step < maxSteps) {
  logger.step(step + 1, "Executing agent step");
 
  const result = await executeStep({ provider, config, tools, conversation });
 
  if (result.errors > 0) {
    logger.error("Step encountered an error");
  }
 
  step++;
 
  if (result.isFinished) break;
}

Conclusion

createLogger is a zero-dependency, configuration-first logger built for LLM agents. Create one instance per agent or service, pass the config from an environment variable, and you have full control over what gets logged without touching your agent logic. Use enabled: process.env.NODE_ENV !== "production" to toggle automatically.

createLogger