Warnf logs a formatted message at WARN level.

Init initializes the configuration system with the specified file path. If the file doesn't exist, a default configuration is created.

Reload rebuilds the weighted account list from config. Weight <= 1 → 1 entry; weight >= 2 → weight entries. Over-quota accounts are dropped unless eit

Save persists the current configuration to the JSON file. Uses indented formatting for human readability.

AddApiKey appends a new API key entry. Generates ID and CreatedAt if missing, rejects empty Key values, and refuses duplicates of an existing Key.

==================== OpenAI -> Kiro 转换 ====================

authenticate validates an incoming request against the configured API keys. Master switch: config.RequireApiKey. When false, requests pass without ch

httpClient 返回当前全局 auth HTTP 客户端

GetNextForModelExcluding 获取下一个支持指定模型的可用账号，并跳过指定账号。

GetApiKeyEntry returns a copy of the entry with the given ID, or nil if not found.

GetThinkingConfig 获取 thinking 配置

Infof logs a formatted message at INFO level.

InitKiroHttpClient initializes (or reinitializes) the HTTP clients used for Kiro API requests.

ensureValidToken 确保 token 有效

recordFailureWithDetails records a failure and stores it in the request logs.

ResolveProfileArn returns the account profile ARN, fetching and caching it when it is missing. First tries ListAvailableProfiles; if that returns empt

recordSuccessForApiKey is recordSuccess + per-API-key usage attribution. When apiKeyID is empty (legacy single-key path or unauthenticated path), only

requireAuth flips the master gate on. Auth is now opt-in via RequireApiKey so most tests need this after seeding their keys.

saveLocked persists cfg to disk. Caller MUST already hold cfgLock. This is identical to Save() (which does not take the lock either) but is named dist

CallKiroAPI calls the Kiro streaming API, trying each configured endpoint with automatic fallback.

Debugf logs a formatted message at DEBUG level.

ParseModelAndThinking resolves a client-supplied model name to a Kiro model ID and reports whether thinking mode was requested via the configured suff

ResolveAccountProxyURL returns the effective proxy URL for an account. Falls back to global config.GetProxyURL() if the account has no per-account pro

getContextWindowSize returns the context window size (in tokens) for a model. Per Kiro's ListAvailableModels, the 1M-token context window applies to

regionalizeURL points a hardcoded us-east-1 Kiro endpoint at the profile's data-plane region. Amazon Q is regional (q.{region}.amazonaws.com), but the

GenerateAccountID 生成账号 ID

GenerateMachineId generates a UUID v4 format machine identifier. This ID is used to uniquely identify the proxy instance in Kiro API requests, helping

GetNext 获取下一个可用账号（加权轮询）

GetRestClientForProxy returns a rest http.Client (30s timeout) for the given proxy URL. If proxyURL is empty, returns the global kiro REST HTTP client

RecordApiKeyUsage atomically adds tokens and credits to the entry's counters, updates LastUsedAt, increments RequestsCount, and persists.

RecordSuccess 记录请求成功，清除冷却

RefreshToken 刷新 access token Returns: accessToken, refreshToken, expiresAt, profileArn, error

SetOIDCTokenURLForTest replaces the OIDC token URL builder. Test-only.

UpdateStats 更新账号统计

buildKiroTransport constructs an HTTP Transport with optional outbound proxy support.

extractThinkingFromContent 从内容中提取 <thinking> 标签内的内容

--------------------------------------------------------------------------- GetNextForModelExcluding -------------------------------------------------

recordSuccessLog records a successful request in the request logs.

ListApiKeys returns a snapshot of all configured API key entries.

RefreshAccountInfo 刷新账户信息（使用量、订阅等）

SetModelList 缓存账号支持的模型集合（由 handler 在刷新后调用）