Phase 1C: Add centralized configuration with env var support

- Create src/config/defaults.ts with DEFAULT_CONFIG and loadConfig()
- Support env vars: VCG_VLLM_URL, VCG_VLLM_MODEL, VCG_VLLM_API_KEY,
  VCG_DEFAULT_PROVIDER, VCG_MAX_CONCURRENCY, VCG_LOG_LEVEL
- Config priority: constructor args > env vars > defaults
- Add VCGConfig type to types.ts
- Export config utilities from public API

https://claude.ai/code/session_012cMotoivyjuMwbrnDo6YRg

src/config/defaults.ts

@@ -0,0 +1,96 @@
+/**
+ * @fileoverview Centralized configuration with env-var support.
+ *
+ * Config priority: constructor args (overrides) > env vars > defaults.
+ *
+ * Supported environment variables:
+ * - `VCG_VLLM_URL`          — vLLM base URL
+ * - `VCG_VLLM_MODEL`        — vLLM model name
+ * - `VCG_VLLM_API_KEY`      — vLLM auth token
+ * - `VCG_DEFAULT_PROVIDER`   — 'vllm' | 'anthropic' | 'openai'
+ * - `VCG_MAX_CONCURRENCY`    — default concurrency limit
+ * - `VCG_LOG_LEVEL`          — 'debug' | 'info' | 'warn' | 'error' | 'silent'
+ *
+ * @module @vcg/agent-sdk
+ */
+
+import type { VCGConfig } from '../types.js'
+
+/** Valid log levels for the SDK. */
+const VALID_LOG_LEVELS = ['debug', 'info', 'warn', 'error', 'silent'] as const
+
+/** Valid provider names. */
+const VALID_PROVIDERS = ['anthropic', 'openai', 'vllm'] as const
+
+/**
+ * Default configuration values used when no override or env var is present.
+ */
+export const DEFAULT_CONFIG: Readonly<VCGConfig> = {
+  defaultProvider: 'anthropic',
+  maxConcurrency: 5,
+  logLevel: 'info',
+}
+
+/**
+ * Build a fully-resolved configuration by merging (in priority order):
+ *   1. Explicit `overrides` (constructor args)
+ *   2. Environment variables
+ *   3. {@link DEFAULT_CONFIG}
+ *
+ * Only defined override fields take precedence; `undefined` fields fall through
+ * to env vars, then to defaults.
+ */
+export function loadConfig(overrides?: Partial<VCGConfig>): VCGConfig {
+  const env = readEnv()
+
+  return {
+    defaultProvider: overrides?.defaultProvider ?? env.defaultProvider ?? DEFAULT_CONFIG.defaultProvider,
+    maxConcurrency: overrides?.maxConcurrency ?? env.maxConcurrency ?? DEFAULT_CONFIG.maxConcurrency,
+    logLevel: overrides?.logLevel ?? env.logLevel ?? DEFAULT_CONFIG.logLevel,
+    vllm: overrides?.vllm ?? env.vllm,
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Internal: read env vars into a partial config
+// ---------------------------------------------------------------------------
+
+type MutablePartialConfig = { -readonly [K in keyof VCGConfig]?: VCGConfig[K] }
+
+function readEnv(): MutablePartialConfig {
+  const result: MutablePartialConfig = {}
+
+  // Provider
+  const provider = process.env['VCG_DEFAULT_PROVIDER']
+  if (provider && (VALID_PROVIDERS as readonly string[]).includes(provider)) {
+    result.defaultProvider = provider as VCGConfig['defaultProvider']
+  }
+
+  // Concurrency
+  const concurrency = process.env['VCG_MAX_CONCURRENCY']
+  if (concurrency) {
+    const parsed = parseInt(concurrency, 10)
+    if (!isNaN(parsed) && parsed > 0) {
+      result.maxConcurrency = parsed
+    }
+  }
+
+  // Log level
+  const logLevel = process.env['VCG_LOG_LEVEL']
+  if (logLevel && (VALID_LOG_LEVELS as readonly string[]).includes(logLevel)) {
+    result.logLevel = logLevel as VCGConfig['logLevel']
+  }
+
+  // vLLM
+  const vllmURL = process.env['VCG_VLLM_URL']
+  const vllmModel = process.env['VCG_VLLM_MODEL']
+  if (vllmURL && vllmModel) {
+    result.vllm = {
+      baseURL: vllmURL,
+      model: vllmModel,
+      apiKey: process.env['VCG_VLLM_API_KEY'],
+    }
+  }
+
+  return result
+}

src/config/index.ts

@@ -0,0 +1,6 @@
+/**
+ * @fileoverview Re-exports for the config module.
+ * @module @vcg/agent-sdk
+ */
+
+export { DEFAULT_CONFIG, loadConfig } from './defaults.js'

src/index.ts

@@ -110,6 +110,12 @@ export { createAdapter } from './llm/adapter.js'
 export type { SupportedProvider } from './llm/adapter.js'
 export { VLLMAdapter } from './llm/vllm.js'
 
+// ---------------------------------------------------------------------------
+// Configuration
+// ---------------------------------------------------------------------------
+
+export { DEFAULT_CONFIG, loadConfig } from './config/index.js'
+
 // ---------------------------------------------------------------------------
 // Memory
 // ---------------------------------------------------------------------------
@@ -170,4 +176,7 @@ export type {
 
   // vLLM
   VLLMConfig,
+
+  // Config
+  VCGConfig,
 } from './types.js'

src/types.ts

@@ -293,6 +293,23 @@ export interface OrchestratorConfig {
 // vLLM configuration
 // ---------------------------------------------------------------------------
 
+/**
+ * Centralized SDK configuration.
+ *
+ * Resolved by {@link loadConfig} using the priority:
+ * constructor args > env vars > defaults.
+ */
+export interface VCGConfig {
+  /** Default LLM provider. */
+  readonly defaultProvider: 'anthropic' | 'openai' | 'vllm'
+  /** Maximum concurrency for agent pools. */
+  readonly maxConcurrency: number
+  /** Logging verbosity. */
+  readonly logLevel: 'debug' | 'info' | 'warn' | 'error' | 'silent'
+  /** vLLM connection settings (populated from env vars or constructor). */
+  readonly vllm?: VLLMConfig
+}
+
 /** Configuration for connecting to a vLLM inference server. */
 export interface VLLMConfig {
   /** Base URL of the vLLM server (e.g. `'http://localhost:8000/v1'`). */