Logging

Pass a logger implementing the Logger interface to route VernLLM's internal logs through your own stack instead of the default console logger:

interface Logger {
  debug(message: string): void;
  warn(message: string): void;
  error(message: string, meta?: Record<string, unknown>): void;
}

pluggable-logger.ts

import pino from 'pino';
import type { Logger } from 'vern-llm';

const pinoInstance = pino();

const pinoLogger: Logger = {
  debug: (msg) => pinoInstance.debug(msg),
  warn: (msg) => pinoInstance.warn(msg),
  error: (msg, meta) => pinoInstance.error(meta, msg),
};

const llm = new VernLLM({ client: openai, model: 'gpt-4o', logger: pinoLogger });

Note debug and warn only ever receive a single message string, meta is error-only. A wrapper that expects a meta argument on every level (many structured loggers do) needs to default it itself, as pinoLogger.error does above.

Whatever you pass in is wrapped so a throwing (or promise-rejecting) implementation can never break the call it's describing, the failure is caught and dropped, at most that one log line is lost.

Debug-level output can also be scrubbed with redact before it reaches your logger:

redact-debug-output.ts

const llm = new VernLLM({
  client: openai,
  model: 'gpt-4o',
  debug: true,
  redact: (text) => text.replace(/\b\d{3}-\d{2}-\d{4}\b/g, '[REDACTED]'),
});

See Pluggable Logger for the full behavior: exactly what gets logged at each level (with a table of every internal log site), how debug interacts with a custom logger, and where redact does and doesn't apply.