/**
 * Ollama runtime configuration helpers.
 *
 * Reads env vars and CPU count to produce sensible defaults
 * for num_thread / num_ctx so that Ollama uses the server efficiently.
 */

import { cpus } from "os";

/** CPU cores visible to Node (host cores when no Docker limit is set). */
function detectCpuCount(): number {
  try {
    return cpus().length;
  } catch {
    return 1;
  }
}

/**
 * Number of threads Ollama should use for inference.
 * Priority:
 *   1. OLLAMA_NUM_THREAD env var
 *   2. cpus().length  (leave 1 core for OS / other containers when > 1)
 */
export function getOllamaNumThread(): number {
  if (process.env.OLLAMA_NUM_THREAD) {
    const parsed = parseInt(process.env.OLLAMA_NUM_THREAD, 10);
    if (!Number.isNaN(parsed) && parsed > 0) {
      return parsed;
    }
  }
  const count = detectCpuCount();
  // Leave one core for the OS / DB / app when we have > 1 core
  return Math.max(1, count > 1 ? count - 1 : count);
}

/**
 * Context window size for Ollama models.
 * Priority:
 *   1. OLLAMA_NUM_CTX env var
 *   2. Default 2048 (safe for CPU-only setups)
 */
export function getOllamaNumCtx(): number {
  if (process.env.OLLAMA_NUM_CTX) {
    const parsed = parseInt(process.env.OLLAMA_NUM_CTX, 10);
    if (!Number.isNaN(parsed) && parsed >= 512) {
      return parsed;
    }
  }
  return 2048;
}