@wengine-ai/llms 2.3.231 → 2.3.234

package/dist/utils/fallback-promotion.d.ts

@@ -0,0 +1,65 @@
+/**
+ * Fallback Promotion Cache
+ *
+ * When a primary model fails and fallback succeeds, we "promote" the fallback model
+ * so subsequent requests skip the failing primary entirely until TTL expires or
+ * the promoted model becomes unhealthy.
+ */
+export interface FallbackPromotionConfig {
+    enabled?: boolean;
+    ttlMinutes?: number;
+}
+/**
+ * Fallback Promotion Store
+ * Tracks successful fallback promotions to avoid retrying failing primaries
+ */
+declare class FallbackPromotionStore {
+    private promotions;
+    private enabled;
+    private ttlMs;
+    configure(config?: FallbackPromotionConfig): void;
+    /**
+     * Generate a cache key for a primary model
+     * Format: ${provider},${model}:${scenarioType}
+     */
+    private getKey;
+    /**
+     * Promote a fallback model for a failing primary
+     * After this, all clients will use the promoted model until TTL expires
+     */
+    promote(provider: string, model: string, scenarioType: string, fallbackProvider: string, fallbackModel: string, ttlMs?: number): void;
+    /**
+     * Get the promoted model for a primary, if one exists and is valid
+     * Returns null if: disabled, no promotion, expired, or promoted model unhealthy
+     */
+    getPromotion(provider: string, model: string, scenarioType: string, providers: any[]): string | null;
+    /**
+     * Clear a specific promotion
+     */
+    clear(provider: string, model: string, scenarioType: string): void;
+    /**
+     * Clear all promotions for a specific primary model
+     * Called when a model recovers to closed (healthy) state
+     * @returns number of promotions cleared
+     */
+    clearByPrimary(provider: string, model: string): number;
+    /**
+     * Clear all promotions
+     */
+    clearAll(): void;
+    /**
+     * Clear expired promotions (cleanup utility)
+     */
+    clearExpired(): void;
+    /**
+     * Get all active promotions (for debugging/monitoring)
+     */
+    getAllPromotions(): Array<{
+        key: string;
+        promotedModel: string;
+        remainingMs: number;
+    }>;
+}
+export declare function getFallbackPromotionStore(config?: FallbackPromotionConfig): FallbackPromotionStore;
+export declare function resetFallbackPromotionStore(): void;
+export {};

package/dist/utils/gemini.util.d.ts

@@ -0,0 +1,11 @@
+import { UnifiedChatRequest } from "../types/llm";
+export declare function cleanupParameters(obj: any, keyName?: string): void;
+/**
+ * Transform a tool object
+ * @param {Object} tool - The tool object to transform
+ * @returns {Object} - The transformed tool object
+ */
+export declare function tTool(tool: any): any;
+export declare function buildRequestBody(request: UnifiedChatRequest): Record<string, any>;
+export declare function transformRequestOut(request: Record<string, any>): UnifiedChatRequest;
+export declare function transformResponseOut(response: Response, providerName: string, logger?: any): Promise<Response>;

package/dist/utils/image.d.ts

@@ -0,0 +1 @@
+export declare const formatBase64: (data: string, media_type: string) => string;

package/dist/utils/request.d.ts

@@ -0,0 +1,2 @@
+import { UnifiedChatRequest } from "../types/llm";
+export declare function sendUnifiedRequest(url: URL | string, request: UnifiedChatRequest, config: any, context: any, logger?: any): Promise<Response>;

package/dist/utils/router.d.ts

@@ -0,0 +1,84 @@
+import { Usage } from "./cache";
+import { ConfigService } from "../services/config";
+import { TokenizerService } from "../services/tokenizer";
+import { type ClientContext } from "../clients/adapters";
+export declare class ProjectRoutingError extends Error {
+    statusCode: number;
+    code: string;
+    type: string;
+    sessionId?: string;
+    configuredTarget?: string;
+    constructor(message: string, opts?: {
+        sessionId?: string;
+        configuredTarget?: string;
+        code?: string;
+        statusCode?: number;
+    });
+}
+interface Tool {
+    name: string;
+    description?: string;
+    input_schema: object;
+}
+interface ContentBlockParam {
+    type: string;
+    [key: string]: any;
+}
+interface MessageParam {
+    role: string;
+    content: string | ContentBlockParam[];
+}
+export declare const calculateTokenCount: (messages: MessageParam[], system: any, tools: Tool[]) => number;
+/**
+ * Look up a provider and model by name in the current provider list.
+ * Returns canonical (case-matched) values or null when the provider is
+ * missing, disabled, or the model is not listed under it.
+ */
+export declare function findProviderModel(providers: any[], providerName: string, modelName: string): {
+    provider: any;
+    model: string;
+} | null;
+/**
+ * Diagnose why a "provider,model" route string fails to resolve.
+ * Returns a human-readable reason and a stable error code so the
+ * ProjectRoutingError can carry actionable diagnostic info.
+ */
+export declare function diagnoseResolutionFailure(configuredModel: string, providers: any[]): {
+    reason: string;
+    code: string;
+};
+interface RouterFamilyConfig {
+    default: string;
+    background?: string;
+    think?: string;
+    longContext?: string;
+    longContextThreshold?: number;
+    extendedContext?: string;
+    extendedContextThreshold?: number;
+    enableExtendedContext?: boolean;
+    webSearch?: string;
+    image?: string;
+    fallback?: Record<string, string[]>;
+}
+export declare function getEffectiveTokenCount(tokenCount: number, lastUsage: Usage | undefined, clientContext?: ClientContext): number;
+export declare function getExtendedContextThreshold(clientContext: ClientContext | undefined, familyConfig?: RouterFamilyConfig, globalExtendedContextThreshold?: number): number;
+export interface RouterContext {
+    configService: ConfigService;
+    tokenizerService?: TokenizerService;
+    event?: any;
+}
+export type RouterScenarioType = 'default' | 'background' | 'think' | 'longContext' | 'extendedContext' | 'webSearch' | 'modelMapping' | 'image';
+export interface RouterFallbackConfig {
+    [key: string]: string[] | undefined;
+    default?: string[];
+    background?: string[];
+    think?: string[];
+    longContext?: string[];
+    extendedContext?: string[];
+    webSearch?: string[];
+    modelMapping?: string[];
+    image?: string[];
+}
+export declare const router: (req: any, _res: any, context: RouterContext) => Promise<void>;
+export declare const searchProjectBySession: (sessionId: string) => Promise<string | null>;
+export {};

package/dist/utils/session-id.d.ts

@@ -0,0 +1,2 @@
+export declare function normalizeSessionId(sessionId: unknown): string | undefined;
+export declare function extractSessionIdFromUserId(userId: unknown): string | undefined;

package/dist/utils/sse/SSEParser.transform.d.ts

@@ -0,0 +1,6 @@
+export declare class SSEParserTransform extends TransformStream<string, any> {
+    private buffer;
+    private currentEvent;
+    constructor();
+    private processLine;
+}

package/dist/utils/sse/SSESerializer.transform.d.ts

@@ -0,0 +1,3 @@
+export declare class SSESerializerTransform extends TransformStream<any, string> {
+    constructor();
+}

package/dist/utils/sse/index.d.ts

@@ -0,0 +1,3 @@
+export { SSEParserTransform } from './SSEParser.transform';
+export { SSESerializerTransform } from './SSESerializer.transform';
+export { rewriteStream } from './rewriteStream';

package/dist/utils/sse/rewriteStream.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Rewrite stream utility
+ * Reads source readablestream and returns a new readablestream,
+ * processor processes source data and pushes returned new value to new stream,
+ * no push if no return value
+ * @param stream
+ * @param processor
+ */
+export declare const rewriteStream: (stream: ReadableStream, processor: (data: any, controller: ReadableStreamController<any>) => Promise<any>) => ReadableStream;

package/dist/utils/thinking.d.ts

@@ -0,0 +1,2 @@
+import { ThinkLevel } from "../types/llm";
+export declare const getThinkLevel: (thinking_budget: number) => ThinkLevel;

package/dist/utils/toolArgumentsParser.d.ts

@@ -0,0 +1,8 @@
+/**
+ * Parse tool call arguments function
+ * First try standard JSON parsing, then JSON5 parsing, finally use jsonrepair for safe repair
+ *
+ * @param argsString - Parameter string to parse
+ * @returns Parsed parameter object or safe empty object
+ */
+export declare function parseToolArguments(argsString: string, logger?: any): string;

package/dist/utils/vertex-claude.util.d.ts

@@ -0,0 +1,42 @@
+import { UnifiedChatRequest } from "../types/llm";
+interface ClaudeMessage {
+    role: "user" | "assistant";
+    content: Array<{
+        type: "text" | "image";
+        text?: string;
+        source?: {
+            type: "base64";
+            media_type: string;
+            data: string;
+        };
+    }>;
+}
+interface ClaudeTool {
+    name: string;
+    description: string;
+    input_schema: {
+        type: string;
+        properties: Record<string, any>;
+        required?: string[];
+        additionalProperties?: boolean;
+        $schema?: string;
+    };
+}
+interface VertexClaudeRequest {
+    anthropic_version: "vertex-2023-10-16";
+    messages: ClaudeMessage[];
+    max_tokens: number;
+    stream?: boolean;
+    temperature?: number;
+    top_p?: number;
+    top_k?: number;
+    tools?: ClaudeTool[];
+    tool_choice?: "auto" | "none" | {
+        type: "tool";
+        name: string;
+    };
+}
+export declare function buildRequestBody(request: UnifiedChatRequest): VertexClaudeRequest;
+export declare function transformRequestOut(request: Record<string, any>): UnifiedChatRequest;
+export declare function transformResponseOut(response: Response, providerName: string, logger?: any): Promise<Response>;
+export {};

package/package.json

@@ -1,14 +1,14 @@
 {
   "name": "@wengine-ai/llms",
-  "version": "2.3.231",
+  "version": "2.3.234",
   "description": "A universal LLM API transformation server",
   "main": "dist/cjs/server.cjs",
   "module": "dist/esm/server.mjs",
-  "types": "./dist/index.d.ts",
+  "types": "./dist/server.d.ts",
   "type": "module",
   "exports": {
     ".": {
-      "types": "./dist/index.d.ts",
+      "types": "./dist/server.d.ts",
       "import": "./dist/esm/server.mjs",
       "require": "./dist/cjs/server.cjs"
     }
@@ -49,14 +49,21 @@
     "openai": "^5.6.0",
     "tiktoken": "^1.0.21",
     "undici": "^7.10.0",
-    "uuid": "^11.1.0"
+    "uuid": "^11.1.0",
+    "@fastify/multipart": "^9.0.0",
+    "@fastify/static": "^8.2.0",
+    "@wengine-ai/claude-code-router-shared": "workspace:*",
+    "adm-zip": "^0.5.16",
+    "better-sqlite3": "^12.10.0",
+    "rotating-file-stream": "^3.2.7"
   },
   "devDependencies": {
     "@types/node": "^24.0.15",
-    "@wengine-ai/claude-code-router-shared": "workspace:*",
     "esbuild": "^0.25.1",
     "tsx": "^4.20.3",
     "typescript": "^5.8.2",
-    "vitest": "^4.1.9"
+    "vitest": "^4.1.9",
+    "@types/adm-zip": "^0.5.7",
+    "@types/better-sqlite3": "^7.6.13"
   }
 }

package/vitest.config.ts

@@ -8,6 +8,7 @@ export default defineConfig({
   resolve: {
     alias: {
       "@": path.resolve(__dirname, "src"),
+      "@wengine-ai/claude-code-router-shared": path.resolve(__dirname, "../shared/src/index.ts"),
     },
   },
 });

package/dist/index.d.ts

@@ -1,175 +0,0 @@
-// Main type declarations for @wengine-ai/llms
-
-// Re-export plugin declarations (pluginManager, tokenSpeedPlugin, etc.)
-export * from "./plugins";
-
-// Health store types
-export interface ProviderHealthState {
-  provider: string;
-  model: string;
-  status: 'closed' | 'open' | 'half-open';
-  failureCount: number;
-  successCount: number;
-  lastFailureTime: number;
-  lastProbeTime: number;
-  lastError?: string;
-}
-
-export interface HealthPoolConfig {
-  enabled?: boolean;
-  failureThreshold?: number;
-  probeIntervalMinutes?: number;
-  halfOpenSuccessThreshold?: number;
-}
-
-export class ProviderHealthStore {
-  constructor(config?: HealthPoolConfig);
-  recordSuccess(provider: string, model: string): void;
-  recordFailure(provider: string, model: string, error?: string): void;
-  getState(provider: string, model: string): ProviderHealthState | undefined;
-  isAvailable(provider: string, model: string): boolean;
-  getPriority(provider: string, model: string): number;
-  needsProbe(state: ProviderHealthState): boolean;
-  markProbeAttempt(provider: string, model: string): void;
-  clear(): void;
-  getAllStates(): ProviderHealthState[];
-  getFailPoolModels(): string[];
-  getHalfOpenModels(): string[];
-  stopProbeTimer(): void;
-}
-
-export declare function getHealthStore(config?: HealthPoolConfig): ProviderHealthStore;
-export declare function resetHealthStore(): void;
-
-// Core exports
-export interface RouterScenarioType {
-  'default': 'default';
-  'background': 'background';
-  'think': 'think';
-  'longContext': 'longContext';
-  'extendedContext': 'extendedContext';
-  'webSearch': 'webSearch';
-  'modelMapping': 'modelMapping';
-  'image': 'image';
-}
-
-export interface RouterFallbackConfig {
-  default?: string[];
-  background?: string[];
-  think?: string[];
-  longContext?: string[];
-  webSearch?: string[];
-  modelMapping?: string[];
-  image?: string[];
-}
-
-export class ConfigService {
-  get<T>(key: string): T | undefined;
-  getAll(): Record<string, any>;
-}
-
-export class ProviderService {
-  getProvider(name: string): any | undefined;
-}
-
-export class TransformerService {
-  getTransformersWithEndpoint(): Array<{ transformer: any }>;
-}
-
-export class TokenizerService {
-  countTokens(request: any, config?: any): Promise<{ tokenCount: number }>;
-}
-
-export function router(req: any, reply: any, context: any): Promise<void>;
-export function calculateTokenCount(messages: any[], system: any, tools: any[]): number;
-export function searchProjectBySession(sessionId: string): Promise<string | null>;
-
-// Session cache
-export const sessionUsageCache: any;
-
-// SSE utilities
-export class SSEParserTransform {}
-export class SSESerializerTransform {}
-export function rewriteStream(stream: any, handler: any): any;
-
-// Rate limit types
-export interface RateLimitInfo {
-  provider: string;
-  remaining: number | null;
-  limit: number | null;
-  reset: number | null;
-  capturedAt: number;
-}
-
-export declare function getAllRateLimitInfo(): RateLimitInfo[];
-export declare function getRateLimitInfo(providerName: string): RateLimitInfo | undefined;
-export declare function initRateLimitPersistence(): void;
-
-// Active probe types
-export interface ActiveProbeConfig {
-  enabled?: boolean;
-  quotaProbeIntervalMinutes?: number;
-  probeTimeoutMs?: number;
-  initialDelayMs?: number;
-  excludeProviders?: string[];
-}
-
-export interface LLMProvider {
-  name: string;
-  baseUrl: string;
-  apiKey: string;
-  models: string[];
-  transformer?: any;
-}
-
-export class ActiveProbeService {
-  constructor(
-    getProviders: () => LLMProvider[],
-    config?: ActiveProbeConfig,
-    getHttpsProxy?: () => string | undefined,
-    logger?: any
-  );
-  start(): void;
-  stop(): void;
-  probeProviderManually(providerName: string): Promise<boolean>;
-}
-
-export declare function getActiveProbeService(
-  getProviders: () => LLMProvider[],
-  config?: ActiveProbeConfig,
-  getHttpsProxy?: () => string | undefined,
-  logger?: any
-): ActiveProbeService;
-
-export declare function startActiveProbe(
-  getProviders: () => LLMProvider[],
-  config?: ActiveProbeConfig,
-  getHttpsProxy?: () => string | undefined,
-  logger?: any
-): ActiveProbeService;
-
-export declare function stopActiveProbe(): void;
-export declare function resetActiveProbeService(): void;
-
-// Runtime debug log toggle
-export declare function setRuntimeDebugLog(enabled: boolean): void;
-export declare function getRuntimeDebugLog(): boolean;
-
-// Quota
-export interface QuotaAdapter {
-  checkQuota(providerName: string, model: string): Promise<boolean>;
-}
-export interface ProviderQuotaResult {
-  totalBalance?: number;
-  usedBalance?: number;
-  limitDaily?: number;
-  usedDailyBalance?: number;
-}
-export interface StoredQuotaResult extends ProviderQuotaResult {
-  updatedAt: number;
-}
-export declare function getQuotaAdapter(providerName: string): QuotaAdapter | undefined;
-export declare function getAllQuotaResults(): Record<string, StoredQuotaResult>;
-export declare function getQuotaResult(providerName: string): StoredQuotaResult | undefined;
-export declare function storeQuotaResult(providerName: string, result: ProviderQuotaResult): void;
-export declare function initQuotaStorePersistence(): void;

package/dist/plugins.d.ts

@@ -1,42 +0,0 @@
-// Plugin type declarations
-
-export interface PluginMetadata {
-  name: string;
-  version?: string;
-  description?: string;
-}
-
-export interface CCRPluginOptions {
-  enabled?: boolean;
-}
-
-export class CCRPlugin {
-  name: string;
-  metadata: PluginMetadata;
-  options: CCRPluginOptions;
-}
-
-export const pluginManager: {
-  registerPlugin(plugin: CCRPlugin, options?: any): void;
-  enablePlugin(name: string, fastify: any): Promise<void>;
-  enablePlugins(fastify: any): Promise<void>;
-  getPlugins(): PluginMetadata[];
-  getPlugin(name: string): CCRPlugin | undefined;
-  hasPlugin(name: string): boolean;
-  isPluginEnabled(name: string): boolean;
-  setPluginEnabled(name: string, enabled: boolean): void;
-  removePlugin(name: string): void;
-  clear(): void;
-};
-
-export const tokenSpeedPlugin: CCRPlugin;
-
-export function getTokenSpeedStats(sessionId: string): {
-  ttft: number | null;
-  tokensPerSecond: number | null;
-};
-
-export function getGlobalTokenSpeedStats(): {
-  avgTtft: number | null;
-  avgTokensPerSecond: number | null;
-};

package/dist/provider-health.d.ts

@@ -1,39 +0,0 @@
-// Type declarations for @wengine-ai/llms
-// Manually maintained to avoid path alias issues
-
-export interface ProviderHealthState {
-  provider: string;
-  model: string;
-  status: 'closed' | 'open' | 'half-open';
-  failureCount: number;
-  successCount: number;
-  lastFailureTime: number;
-  lastProbeTime: number;
-  lastError?: string;
-}
-
-export interface HealthPoolConfig {
-  enabled?: boolean;
-  failureThreshold?: number;
-  probeIntervalMinutes?: number;
-  halfOpenSuccessThreshold?: number;
-}
-
-export class ProviderHealthStore {
-  constructor(config?: HealthPoolConfig);
-  recordSuccess(provider: string, model: string): void;
-  recordFailure(provider: string, model: string, error?: string): void;
-  getState(provider: string, model: string): ProviderHealthState | undefined;
-  isAvailable(provider: string, model: string): boolean;
-  getPriority(provider: string, model: string): number;
-  needsProbe(state: ProviderHealthState): boolean;
-  markProbeAttempt(provider: string, model: string): void;
-  clear(): void;
-  getAllStates(): ProviderHealthState[];
-  getFailPoolModels(): string[];
-  getHalfOpenModels(): string[];
-  stopProbeTimer(): void;
-}
-
-export function getHealthStore(config?: HealthPoolConfig): ProviderHealthStore;
-export function resetHealthStore(): void;