@od-oneapp/ai-platform 0.1.0

package/dist/registry-CsD3iTIx.mjs

@@ -0,0 +1,190 @@
+//#region src/models/registry.ts
+/**
+* Semantic model names - use these in code, not provider-specific IDs
+*/
+let Model = /* @__PURE__ */ function(Model) {
+	Model["OPUS"] = "opus";
+	Model["SONNET"] = "sonnet";
+	Model["HAIKU"] = "haiku";
+	Model["GPT_4_TURBO"] = "gpt-4-turbo";
+	Model["GPT_4O"] = "gpt-4o";
+	Model["GPT_4O_MINI"] = "gpt-4o-mini";
+	Model["O1_PREVIEW"] = "o1-preview";
+	Model["O1_PRO"] = "o1-pro";
+	Model["GEMINI_15_FLASH"] = "gemini-1.5-flash";
+	Model["GEMINI_15_PRO"] = "gemini-1.5-pro";
+	Model["GEMINI_20_FLASH"] = "gemini-2.0-flash";
+	Model["GROK_BETA"] = "grok-beta";
+	Model["PERPLEXITY"] = "perplexity";
+	return Model;
+}({});
+/**
+* Model ID mapping - provider-specific IDs
+* Update here when model versions change
+*/
+const MODEL_IDS = {
+	[Model.OPUS]: "claude-opus-4-20250514",
+	[Model.SONNET]: "claude-sonnet-4-20250514",
+	[Model.HAIKU]: "claude-haiku-4-20250514",
+	[Model.GPT_4_TURBO]: "gpt-4-turbo",
+	[Model.GPT_4O]: "gpt-4o",
+	[Model.GPT_4O_MINI]: "gpt-4o-mini",
+	[Model.O1_PREVIEW]: "o1-preview",
+	[Model.O1_PRO]: "o1-pro",
+	[Model.GEMINI_15_FLASH]: "gemini-1.5-flash",
+	[Model.GEMINI_15_PRO]: "gemini-1.5-pro",
+	[Model.GEMINI_20_FLASH]: "gemini-2.0-flash-exp",
+	[Model.GROK_BETA]: "grok-beta",
+	[Model.PERPLEXITY]: "perplexity-sonar"
+};
+/**
+* Model metadata registry
+*/
+const MODEL_METADATA = {
+	[Model.OPUS]: {
+		id: MODEL_IDS[Model.OPUS],
+		provider: "anthropic",
+		tier: "powerful",
+		contextWindow: 2e5,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: 15,
+		costPer1kOutput: 75
+	},
+	[Model.SONNET]: {
+		id: MODEL_IDS[Model.SONNET],
+		provider: "anthropic",
+		tier: "balanced",
+		contextWindow: 2e5,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: 3,
+		costPer1kOutput: 15
+	},
+	[Model.HAIKU]: {
+		id: MODEL_IDS[Model.HAIKU],
+		provider: "anthropic",
+		tier: "fast",
+		contextWindow: 2e5,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: .8,
+		costPer1kOutput: 4
+	},
+	[Model.GPT_4_TURBO]: {
+		id: MODEL_IDS[Model.GPT_4_TURBO],
+		provider: "openai",
+		tier: "powerful",
+		contextWindow: 128e3,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: 10,
+		costPer1kOutput: 30
+	},
+	[Model.GPT_4O]: {
+		id: MODEL_IDS[Model.GPT_4O],
+		provider: "openai",
+		tier: "balanced",
+		contextWindow: 128e3,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: 2.5,
+		costPer1kOutput: 10
+	},
+	[Model.GPT_4O_MINI]: {
+		id: MODEL_IDS[Model.GPT_4O_MINI],
+		provider: "openai",
+		tier: "fast",
+		contextWindow: 128e3,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: .15,
+		costPer1kOutput: .6
+	},
+	[Model.O1_PREVIEW]: {
+		id: MODEL_IDS[Model.O1_PREVIEW],
+		provider: "openai",
+		tier: "reasoning",
+		contextWindow: 128e3,
+		supportsTools: false,
+		supportsStreaming: false,
+		costPer1kInput: 15,
+		costPer1kOutput: 60
+	},
+	[Model.O1_PRO]: {
+		id: MODEL_IDS[Model.O1_PRO],
+		provider: "openai",
+		tier: "reasoning",
+		contextWindow: 128e3,
+		supportsTools: false,
+		supportsStreaming: false,
+		costPer1kInput: 60,
+		costPer1kOutput: 240
+	},
+	[Model.GEMINI_15_FLASH]: {
+		id: MODEL_IDS[Model.GEMINI_15_FLASH],
+		provider: "google",
+		tier: "fast",
+		contextWindow: 1e6,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: .075,
+		costPer1kOutput: .3
+	},
+	[Model.GEMINI_15_PRO]: {
+		id: MODEL_IDS[Model.GEMINI_15_PRO],
+		provider: "google",
+		tier: "balanced",
+		contextWindow: 2e6,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: 1.25,
+		costPer1kOutput: 5
+	},
+	[Model.GEMINI_20_FLASH]: {
+		id: MODEL_IDS[Model.GEMINI_20_FLASH],
+		provider: "google",
+		tier: "fast",
+		contextWindow: 1e6,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: .075,
+		costPer1kOutput: .3
+	},
+	[Model.GROK_BETA]: {
+		id: MODEL_IDS[Model.GROK_BETA],
+		provider: "xai",
+		tier: "balanced",
+		contextWindow: 128e3,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: 5,
+		costPer1kOutput: 15
+	},
+	[Model.PERPLEXITY]: {
+		id: MODEL_IDS[Model.PERPLEXITY],
+		provider: "perplexity",
+		tier: "balanced",
+		contextWindow: 127072,
+		supportsTools: true,
+		supportsStreaming: true,
+		costPer1kInput: 1,
+		costPer1kOutput: 1
+	}
+};
+/**
+* Get provider-specific model ID
+*/
+function getModelId(model) {
+	return MODEL_IDS[model];
+}
+/**
+* Get model metadata
+*/
+function getModelMetadata(model) {
+	return MODEL_METADATA[model];
+}
+
+//#endregion
+export { getModelMetadata as i, Model as n, getModelId as r, MODEL_METADATA as t };
+//# sourceMappingURL=registry-CsD3iTIx.mjs.map

package/dist/registry-CsD3iTIx.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"registry-CsD3iTIx.mjs","names":[],"sources":["../src/models/registry.ts"],"sourcesContent":["import type { ModelMetadata } from './types';\n\n/**\n * Semantic model names - use these in code, not provider-specific IDs\n */\nexport enum Model {\n  // Anthropic\n  OPUS = 'opus',\n  SONNET = 'sonnet',\n  HAIKU = 'haiku',\n\n  // OpenAI\n  GPT_4_TURBO = 'gpt-4-turbo',\n  GPT_4O = 'gpt-4o',\n  GPT_4O_MINI = 'gpt-4o-mini',\n  O1_PREVIEW = 'o1-preview',\n  O1_PRO = 'o1-pro',\n\n  // Google\n  GEMINI_15_FLASH = 'gemini-1.5-flash',\n  GEMINI_15_PRO = 'gemini-1.5-pro',\n  GEMINI_20_FLASH = 'gemini-2.0-flash',\n\n  // X AI\n  GROK_BETA = 'grok-beta',\n\n  // Perplexity\n  PERPLEXITY = 'perplexity',\n}\n\n/**\n * Model ID mapping - provider-specific IDs\n * Update here when model versions change\n */\nconst MODEL_IDS: Record<Model, string> = {\n  // Anthropic\n  [Model.OPUS]: 'claude-opus-4-20250514',\n  [Model.SONNET]: 'claude-sonnet-4-20250514',\n  [Model.HAIKU]: 'claude-haiku-4-20250514',\n\n  // OpenAI\n  [Model.GPT_4_TURBO]: 'gpt-4-turbo',\n  [Model.GPT_4O]: 'gpt-4o',\n  [Model.GPT_4O_MINI]: 'gpt-4o-mini',\n  [Model.O1_PREVIEW]: 'o1-preview',\n  [Model.O1_PRO]: 'o1-pro',\n\n  // Google\n  [Model.GEMINI_15_FLASH]: 'gemini-1.5-flash',\n  [Model.GEMINI_15_PRO]: 'gemini-1.5-pro',\n  [Model.GEMINI_20_FLASH]: 'gemini-2.0-flash-exp',\n\n  // X AI\n  [Model.GROK_BETA]: 'grok-beta',\n\n  // Perplexity\n  [Model.PERPLEXITY]: 'perplexity-sonar',\n};\n\n/**\n * Model metadata registry\n */\nexport const MODEL_METADATA: Record<Model, ModelMetadata> = {\n  [Model.OPUS]: {\n    id: MODEL_IDS[Model.OPUS],\n    provider: 'anthropic',\n    tier: 'powerful',\n    contextWindow: 200_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 15,\n    costPer1kOutput: 75,\n  },\n  [Model.SONNET]: {\n    id: MODEL_IDS[Model.SONNET],\n    provider: 'anthropic',\n    tier: 'balanced',\n    contextWindow: 200_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 3,\n    costPer1kOutput: 15,\n  },\n  [Model.HAIKU]: {\n    id: MODEL_IDS[Model.HAIKU],\n    provider: 'anthropic',\n    tier: 'fast',\n    contextWindow: 200_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 0.8,\n    costPer1kOutput: 4,\n  },\n  [Model.GPT_4_TURBO]: {\n    id: MODEL_IDS[Model.GPT_4_TURBO],\n    provider: 'openai',\n    tier: 'powerful',\n    contextWindow: 128_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 10,\n    costPer1kOutput: 30,\n  },\n  [Model.GPT_4O]: {\n    id: MODEL_IDS[Model.GPT_4O],\n    provider: 'openai',\n    tier: 'balanced',\n    contextWindow: 128_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 2.5,\n    costPer1kOutput: 10,\n  },\n  [Model.GPT_4O_MINI]: {\n    id: MODEL_IDS[Model.GPT_4O_MINI],\n    provider: 'openai',\n    tier: 'fast',\n    contextWindow: 128_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 0.15,\n    costPer1kOutput: 0.6,\n  },\n  [Model.O1_PREVIEW]: {\n    id: MODEL_IDS[Model.O1_PREVIEW],\n    provider: 'openai',\n    tier: 'reasoning',\n    contextWindow: 128_000,\n    supportsTools: false,\n    supportsStreaming: false,\n    costPer1kInput: 15,\n    costPer1kOutput: 60,\n  },\n  [Model.O1_PRO]: {\n    id: MODEL_IDS[Model.O1_PRO],\n    provider: 'openai',\n    tier: 'reasoning',\n    contextWindow: 128_000,\n    supportsTools: false,\n    supportsStreaming: false,\n    costPer1kInput: 60,\n    costPer1kOutput: 240,\n  },\n  [Model.GEMINI_15_FLASH]: {\n    id: MODEL_IDS[Model.GEMINI_15_FLASH],\n    provider: 'google',\n    tier: 'fast',\n    contextWindow: 1_000_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 0.075,\n    costPer1kOutput: 0.3,\n  },\n  [Model.GEMINI_15_PRO]: {\n    id: MODEL_IDS[Model.GEMINI_15_PRO],\n    provider: 'google',\n    tier: 'balanced',\n    contextWindow: 2_000_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 1.25,\n    costPer1kOutput: 5,\n  },\n  [Model.GEMINI_20_FLASH]: {\n    id: MODEL_IDS[Model.GEMINI_20_FLASH],\n    provider: 'google',\n    tier: 'fast',\n    contextWindow: 1_000_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 0.075,\n    costPer1kOutput: 0.3,\n  },\n  [Model.GROK_BETA]: {\n    id: MODEL_IDS[Model.GROK_BETA],\n    provider: 'xai',\n    tier: 'balanced',\n    contextWindow: 128_000,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 5,\n    costPer1kOutput: 15,\n  },\n  [Model.PERPLEXITY]: {\n    id: MODEL_IDS[Model.PERPLEXITY],\n    provider: 'perplexity',\n    tier: 'balanced',\n    contextWindow: 127_072,\n    supportsTools: true,\n    supportsStreaming: true,\n    costPer1kInput: 1,\n    costPer1kOutput: 1,\n  },\n};\n\n/**\n * Get provider-specific model ID\n */\nexport function getModelId(model: Model): string {\n  return MODEL_IDS[model];\n}\n\n/**\n * Get model metadata\n */\nexport function getModelMetadata(model: Model): ModelMetadata {\n  return MODEL_METADATA[model];\n}\n"],"mappings":";;;;AAKA,IAAY,wCAAL;AAEL;AACA;AACA;AAGA;AACA;AACA;AACA;AACA;AAGA;AACA;AACA;AAGA;AAGA;;;;;;;AAOF,MAAM,YAAmC;EAEtC,MAAM,OAAO;EACb,MAAM,SAAS;EACf,MAAM,QAAQ;EAGd,MAAM,cAAc;EACpB,MAAM,SAAS;EACf,MAAM,cAAc;EACpB,MAAM,aAAa;EACnB,MAAM,SAAS;EAGf,MAAM,kBAAkB;EACxB,MAAM,gBAAgB;EACtB,MAAM,kBAAkB;EAGxB,MAAM,YAAY;EAGlB,MAAM,aAAa;CACrB;;;;AAKD,MAAa,iBAA+C;EACzD,MAAM,OAAO;EACZ,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,SAAS;EACd,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,QAAQ;EACb,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,cAAc;EACnB,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,SAAS;EACd,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,cAAc;EACnB,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,aAAa;EAClB,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,SAAS;EACd,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,kBAAkB;EACvB,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,gBAAgB;EACrB,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,kBAAkB;EACvB,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,YAAY;EACjB,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;EACA,MAAM,aAAa;EAClB,IAAI,UAAU,MAAM;EACpB,UAAU;EACV,MAAM;EACN,eAAe;EACf,eAAe;EACf,mBAAmB;EACnB,gBAAgB;EAChB,iBAAiB;EAClB;CACF;;;;AAKD,SAAgB,WAAW,OAAsB;AAC/C,QAAO,UAAU;;;;;AAMnB,SAAgB,iBAAiB,OAA6B;AAC5D,QAAO,eAAe"}

package/dist/registry-DVPWzkXR.d.mts

@@ -0,0 +1,36 @@
+//#region src/models/types.d.ts
+type Gateway = 'cloudflare' | 'vercel' | 'direct';
+type ModelTier = 'fast' | 'balanced' | 'powerful' | 'reasoning';
+interface ModelMetadata {
+  readonly id: string;
+  readonly provider: 'anthropic' | 'openai' | 'google' | 'perplexity' | 'xai';
+  readonly tier: ModelTier;
+  readonly contextWindow: number;
+  readonly supportsTools: boolean;
+  readonly supportsStreaming: boolean;
+  readonly costPer1kInput: number;
+  readonly costPer1kOutput: number;
+}
+//#endregion
+//#region src/models/registry.d.ts
+declare enum Model {
+  OPUS = "opus",
+  SONNET = "sonnet",
+  HAIKU = "haiku",
+  GPT_4_TURBO = "gpt-4-turbo",
+  GPT_4O = "gpt-4o",
+  GPT_4O_MINI = "gpt-4o-mini",
+  O1_PREVIEW = "o1-preview",
+  O1_PRO = "o1-pro",
+  GEMINI_15_FLASH = "gemini-1.5-flash",
+  GEMINI_15_PRO = "gemini-1.5-pro",
+  GEMINI_20_FLASH = "gemini-2.0-flash",
+  GROK_BETA = "grok-beta",
+  PERPLEXITY = "perplexity"
+}
+declare const MODEL_METADATA: Record<Model, ModelMetadata>;
+declare function getModelId(model: Model): string;
+declare function getModelMetadata(model: Model): ModelMetadata;
+//#endregion
+export { Gateway as a, getModelMetadata as i, Model as n, ModelMetadata as o, getModelId as r, ModelTier as s, MODEL_METADATA as t };
+//# sourceMappingURL=registry-DVPWzkXR.d.mts.map

package/dist/registry-DVPWzkXR.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"registry-DVPWzkXR.d.mts","names":[],"sources":["../src/models/types.ts","../src/models/registry.ts"],"mappings":";KAGY,OAAA;AAAA,KAKA,SAAA;AAAA,UAKK,aAAA;EAAA,SACN,EAAA;EAAA,SACA,QAAA;EAAA,SACA,IAAA,EAAM,SAAA;EAAA,SACN,aAAA;EAAA,SACA,aAAA;EAAA,SACA,iBAAA;EAAA,SACA,cAAA;EAAA,SACA,eAAA;AAAA;;;aChBC,KAAA;EAEV,IAAA;EACA,MAAA;EACA,KAAA;EAGA,WAAA;EACA,MAAA;EACA,WAAA;EACA,UAAA;EACA,MAAA;EAGA,eAAA;EACA,aAAA;EACA,eAAA;EAGA,SAAA;EAGA,UAAA;AAAA;AAAA,cAmCW,cAAA,EAAgB,MAAA,CAAO,KAAA,EAAO,aAAA;AAAA,iBAwI3B,UAAA,CAAW,KAAA,EAAO,KAAA;AAAA,iBAOlB,gBAAA,CAAiB,KAAA,EAAO,KAAA,GAAQ,aAAA"}

package/dist/reranking-BpWYhYzl.d.mts

@@ -0,0 +1,72 @@
+import { z } from "zod";
+
+//#region src/grounding/embed/reranking.d.ts
+interface RerankDocument {
+  id: string;
+  text: string;
+  metadata?: Record<string, unknown>;
+  originalScore?: number;
+}
+interface RerankResult {
+  document: RerankDocument;
+  relevanceScore: number;
+  originalIndex: number;
+  newIndex: number;
+}
+interface RerankOptions {
+  query: string;
+  documents: RerankDocument[];
+  topK?: number;
+  minScore?: number;
+  returnAllResults?: boolean;
+  model?: string;
+}
+interface RerankProvider {
+  readonly name: string;
+  rerank(options: RerankOptions): Promise<RerankResult[]>;
+  isAvailable(): Promise<boolean>;
+}
+interface CohereRerankConfig {
+  apiKey: string;
+  model?: 'rerank-english-v3.0' | 'rerank-multilingual-v3.0' | 'rerank-english-v2.0';
+  baseUrl?: string;
+  maxRetries?: number;
+  timeout?: number;
+}
+interface CrossEncoderConfig {
+  modelName: string;
+  endpoint: string;
+  apiKey?: string;
+  batchSize?: number;
+  timeout?: number;
+}
+declare function createCohereReranker(config: CohereRerankConfig): RerankProvider;
+declare function createCrossEncoderReranker(config: CrossEncoderConfig): RerankProvider;
+declare function reciprocalRankFusion(rankings: RerankResult[][], options?: {
+  k?: number;
+  topK?: number;
+}): RerankResult[];
+interface RerankPipelineConfig {
+  provider: RerankProvider;
+  fallbackProvider?: RerankProvider;
+  useRRFWithOriginal?: boolean;
+  rrfK?: number;
+  enableCache?: boolean;
+  cacheTTL?: number;
+}
+declare function createRerankPipeline(config: RerankPipelineConfig): {
+  rerank(options: RerankOptions): Promise<RerankResult[]>;
+  isAvailable(): Promise<boolean>;
+  clearCache(): void;
+};
+declare function vectorResultsToDocuments(results: Array<{
+  id: string;
+  content: string;
+  score: number;
+  metadata?: Record<string, unknown>;
+}>): RerankDocument[];
+declare function calculateNDCG(results: RerankResult[], relevanceLabels: Map<string, number>, k?: number): number;
+declare function calculateMRR(results: RerankResult[], relevantIds: Set<string>): number;
+//#endregion
+export { RerankPipelineConfig as a, calculateMRR as c, createCrossEncoderReranker as d, createRerankPipeline as f, RerankOptions as i, calculateNDCG as l, vectorResultsToDocuments as m, CrossEncoderConfig as n, RerankProvider as o, reciprocalRankFusion as p, RerankDocument as r, RerankResult as s, CohereRerankConfig as t, createCohereReranker as u };
+//# sourceMappingURL=reranking-BpWYhYzl.d.mts.map

package/dist/reranking-BpWYhYzl.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"reranking-BpWYhYzl.d.mts","names":[],"sources":["../src/grounding/embed/reranking.ts"],"mappings":";;;UAmBiB,cAAA;EAEf,EAAA;EAEA,IAAA;EAEA,QAAA,GAAW,MAAA;EAEX,aAAA;AAAA;AAAA,UAMe,YAAA;EAEf,QAAA,EAAU,cAAA;EAEV,cAAA;EAEA,aAAA;EAEA,QAAA;AAAA;AAAA,UAMe,aAAA;EAEf,KAAA;EAEA,SAAA,EAAW,cAAA;EAEX,IAAA;EAEA,QAAA;EAEA,gBAAA;EAEA,KAAA;AAAA;AAAA,UAMe,cAAA;EAAA,SAEN,IAAA;EAET,MAAA,CAAO,OAAA,EAAS,aAAA,GAAgB,OAAA,CAAQ,YAAA;EAExC,WAAA,IAAe,OAAA;AAAA;AAAA,UAMA,kBAAA;EACf,MAAA;EACA,KAAA;EACA,OAAA;EACA,UAAA;EACA,OAAA;AAAA;AAAA,UAMe,kBAAA;EACf,SAAA;EACA,QAAA;EACA,MAAA;EACA,SAAA;EACA,OAAA;AAAA;AAAA,iBA4Cc,oBAAA,CAAqB,MAAA,EAAQ,kBAAA,GAAqB,cAAA;AAAA,iBAqHlD,0BAAA,CAA2B,MAAA,EAAQ,kBAAA,GAAqB,cAAA;AAAA,iBAsGxD,oBAAA,CACd,QAAA,EAAU,YAAA,MACV,OAAA;EAAW,CAAA;EAAY,IAAA;AAAA,IACtB,YAAA;AAAA,UA4Cc,oBAAA;EAEf,QAAA,EAAU,cAAA;EAEV,gBAAA,GAAmB,cAAA;EAEnB,kBAAA;EAEA,IAAA;EAEA,WAAA;EAEA,QAAA;AAAA;AAAA,iBAsBc,oBAAA,CAAqB,MAAA,EAAQ,oBAAA;kBAkBnB,aAAA,GAAgB,OAAA,CAAQ,YAAA;iBAmEzB,OAAA;;;iBAsBT,wBAAA,CACd,OAAA,EAAS,KAAA;EACP,EAAA;EACA,OAAA;EACA,KAAA;EACA,QAAA,GAAW,MAAA;AAAA,KAEZ,cAAA;AAAA,iBAaa,aAAA,CACd,OAAA,EAAS,YAAA,IACT,eAAA,EAAiB,GAAA,kBACjB,CAAA;AAAA,iBA4Bc,YAAA,CAAa,OAAA,EAAS,YAAA,IAAgB,WAAA,EAAa,GAAA"}

package/dist/reranking-Ck8aKZW7.mjs

@@ -0,0 +1,331 @@
+import { z } from "zod";
+import { logWarn } from "@od-oneapp/shared/logs";
+
+//#region src/grounding/embed/reranking.ts
+/**
+* Reranking Module - AI SDK v6 Compatible
+*
+* Provides document reranking capabilities for RAG pipelines using
+* Cohere, cross-encoders, and other reranking providers.
+*
+* @module @od-oneapp/ai-platform/grounding/embed/reranking
+*/
+const RerankDocumentSchema = z.object({
+	id: z.string(),
+	text: z.string(),
+	metadata: z.record(z.string(), z.unknown()).optional(),
+	originalScore: z.number().optional()
+});
+const RerankOptionsSchema = z.object({
+	query: z.string().min(1),
+	documents: z.array(RerankDocumentSchema).min(1),
+	topK: z.number().int().positive().optional(),
+	minScore: z.number().min(0).max(1).optional(),
+	returnAllResults: z.boolean().optional(),
+	model: z.string().optional()
+});
+/**
+* Create a Cohere reranker provider
+*
+* @example
+* ```typescript
+* const reranker = createCohereReranker({
+*   apiKey: process.env.COHERE_API_KEY!,
+*   model: 'rerank-english-v3.0',
+* });
+*
+* const results = await reranker.rerank({
+*   query: 'What is machine learning?',
+*   documents: docs,
+*   topK: 5,
+* });
+* ```
+*/
+function createCohereReranker(config) {
+	const { apiKey, model = "rerank-english-v3.0", baseUrl = "https://api.cohere.ai/v1", maxRetries = 3, timeout = 3e4 } = config;
+	return {
+		name: "cohere",
+		async rerank(options) {
+			const { query, documents, topK, minScore = 0, returnAllResults = false } = options;
+			if (documents.length === 0) return [];
+			const docTexts = documents.map((d) => d.text);
+			let lastError = null;
+			for (let attempt = 0; attempt < maxRetries; attempt++) try {
+				const controller = new AbortController();
+				const timeoutId = setTimeout(() => controller.abort(), timeout);
+				const response = await fetch(`${baseUrl}/rerank`, {
+					method: "POST",
+					headers: {
+						Authorization: `Bearer ${apiKey}`,
+						"Content-Type": "application/json"
+					},
+					body: JSON.stringify({
+						query,
+						documents: docTexts,
+						model: options.model || model,
+						top_n: topK || documents.length,
+						return_documents: false
+					}),
+					signal: controller.signal
+				});
+				clearTimeout(timeoutId);
+				if (!response.ok) {
+					const errorText = await response.text();
+					throw new Error(`Cohere API error: ${response.status} - ${errorText}`);
+				}
+				const results = (await response.json()).results.map((r, newIndex) => {
+					const document = documents[r.index];
+					if (!document) return null;
+					return {
+						document,
+						relevanceScore: r.relevance_score,
+						originalIndex: r.index,
+						newIndex
+					};
+				}).filter((entry) => entry !== null && (returnAllResults || entry.relevanceScore >= minScore));
+				return topK ? results.slice(0, topK) : results;
+			} catch (error) {
+				lastError = error;
+				if (attempt < maxRetries - 1) await new Promise((resolve) => setTimeout(resolve, Math.pow(2, attempt) * 1e3));
+			}
+			throw lastError || /* @__PURE__ */ new Error("Cohere rerank failed after retries");
+		},
+		async isAvailable() {
+			try {
+				return (await fetch(`${baseUrl}/check-api-key`, {
+					method: "POST",
+					headers: {
+						Authorization: `Bearer ${apiKey}`,
+						"Content-Type": "application/json"
+					}
+				})).ok;
+			} catch {
+				return false;
+			}
+		}
+	};
+}
+/**
+* Create a cross-encoder reranker (for self-hosted models)
+*
+* @example
+* ```typescript
+* const reranker = createCrossEncoderReranker({
+*   modelName: 'cross-encoder/ms-marco-MiniLM-L-6-v2',
+*   endpoint: 'http://localhost:8080/rerank',
+* });
+* ```
+*/
+function createCrossEncoderReranker(config) {
+	const { modelName, endpoint, apiKey, batchSize = 32, timeout = 6e4 } = config;
+	return {
+		name: `cross-encoder:${modelName}`,
+		async rerank(options) {
+			const { query, documents, topK, minScore = 0, returnAllResults = false } = options;
+			if (documents.length === 0) return [];
+			const allScores = [];
+			for (let i = 0; i < documents.length; i += batchSize) {
+				const batch = documents.slice(i, i + batchSize);
+				const controller = new AbortController();
+				const timeoutId = setTimeout(() => controller.abort(), timeout);
+				const headers = { "Content-Type": "application/json" };
+				if (apiKey) headers["Authorization"] = `Bearer ${apiKey}`;
+				const response = await fetch(endpoint, {
+					method: "POST",
+					headers,
+					body: JSON.stringify({
+						model: modelName,
+						query,
+						documents: batch.map((d) => d.text)
+					}),
+					signal: controller.signal
+				});
+				clearTimeout(timeoutId);
+				if (!response.ok) throw new Error(`Cross-encoder API error: ${response.status}`);
+				const data = await response.json();
+				allScores.push(...data.scores);
+			}
+			const results = documents.map((doc, index) => {
+				const score = allScores[index];
+				if (score === void 0) return null;
+				return {
+					document: doc,
+					relevanceScore: score,
+					originalIndex: index,
+					newIndex: -1
+				};
+			}).filter((entry) => entry !== null).sort((a, b) => b.relevanceScore - a.relevanceScore).map((r, newIndex) => ({
+				...r,
+				newIndex
+			})).filter((r) => returnAllResults || r.relevanceScore >= minScore);
+			return topK ? results.slice(0, topK) : results;
+		},
+		async isAvailable() {
+			try {
+				return (await fetch(endpoint.replace("/rerank", "/health"), { method: "GET" })).ok;
+			} catch {
+				return false;
+			}
+		}
+	};
+}
+/**
+* Combine multiple ranking results using Reciprocal Rank Fusion
+*
+* RRF is useful when combining results from different retrieval methods
+* (e.g., BM25 + vector search + reranker).
+*
+* @example
+* ```typescript
+* const combined = reciprocalRankFusion([
+*   vectorSearchResults,
+*   bm25Results,
+*   rerankResults,
+* ], { k: 60 });
+* ```
+*/
+function reciprocalRankFusion(rankings, options = {}) {
+	const { k = 60, topK } = options;
+	const scoreMap = /* @__PURE__ */ new Map();
+	for (const ranking of rankings) for (let rank = 0; rank < ranking.length; rank++) {
+		const result = ranking[rank];
+		if (!result) continue;
+		const docId = result.document.id;
+		const rrfScore = 1 / (k + rank + 1);
+		const existing = scoreMap.get(docId);
+		if (existing) existing.score += rrfScore;
+		else scoreMap.set(docId, {
+			score: rrfScore,
+			document: result.document
+		});
+	}
+	const results = Array.from(scoreMap.values()).sort((a, b) => b.score - a.score).map((item, index) => ({
+		document: item.document,
+		relevanceScore: item.score,
+		originalIndex: -1,
+		newIndex: index
+	}));
+	return topK ? results.slice(0, topK) : results;
+}
+/**
+* Create a reranking pipeline with fallback and caching
+*
+* @example
+* ```typescript
+* const pipeline = createRerankPipeline({
+*   provider: createCohereReranker({ apiKey: '...' }),
+*   fallbackProvider: createCrossEncoderReranker({ ... }),
+*   useRRFWithOriginal: true,
+*   enableCache: true,
+* });
+*
+* const results = await pipeline.rerank({
+*   query: 'machine learning',
+*   documents: docs,
+*   topK: 10,
+* });
+* ```
+*/
+function createRerankPipeline(config) {
+	const { provider, fallbackProvider, useRRFWithOriginal = false, rrfK = 60, enableCache = false, cacheTTL = 300 } = config;
+	const cache = /* @__PURE__ */ new Map();
+	function getCacheKey(query, docIds) {
+		return `${query}:${docIds.sort().join(",")}`;
+	}
+	return {
+		async rerank(options) {
+			const { query, documents, topK } = options;
+			if (enableCache) {
+				const cacheKey = getCacheKey(query, documents.map((d) => d.id));
+				const cached = cache.get(cacheKey);
+				if (cached && Date.now() - cached.timestamp < cacheTTL * 1e3) return topK ? cached.results.slice(0, topK) : cached.results;
+			}
+			let rerankResults;
+			try {
+				rerankResults = await provider.rerank(options);
+			} catch (error) {
+				if (fallbackProvider) {
+					logWarn("Primary reranker failed, using fallback", { error: error instanceof Error ? error : new Error(String(error)) });
+					rerankResults = await fallbackProvider.rerank(options);
+				} else throw error;
+			}
+			if (useRRFWithOriginal && documents.some((d) => d.originalScore !== void 0)) {
+				const originalRanking = [...documents].sort((a, b) => (b.originalScore || 0) - (a.originalScore || 0)).map((doc, index) => ({
+					document: doc,
+					relevanceScore: doc.originalScore || 0,
+					originalIndex: documents.indexOf(doc),
+					newIndex: index
+				}));
+				rerankResults = reciprocalRankFusion([rerankResults, originalRanking], {
+					k: rrfK,
+					topK
+				});
+			}
+			if (enableCache) {
+				const cacheKey = getCacheKey(query, documents.map((d) => d.id));
+				cache.set(cacheKey, {
+					results: rerankResults,
+					timestamp: Date.now()
+				});
+				const now = Date.now();
+				for (const [key, value] of cache.entries()) if (now - value.timestamp > cacheTTL * 1e3) cache.delete(key);
+			}
+			return rerankResults;
+		},
+		async isAvailable() {
+			if (await provider.isAvailable()) return true;
+			if (fallbackProvider) return fallbackProvider.isAvailable();
+			return false;
+		},
+		clearCache() {
+			cache.clear();
+		}
+	};
+}
+/**
+* Convert vector search results to RerankDocuments
+*/
+function vectorResultsToDocuments(results) {
+	return results.map((r) => ({
+		id: r.id,
+		text: r.content,
+		metadata: r.metadata,
+		originalScore: r.score
+	}));
+}
+/**
+* Calculate Normalized Discounted Cumulative Gain (NDCG)
+* for evaluating reranking quality
+*/
+function calculateNDCG(results, relevanceLabels, k) {
+	const resultsToScore = k ? results.slice(0, k) : results;
+	let dcg = 0;
+	for (let i = 0; i < resultsToScore.length; i++) {
+		const result = resultsToScore[i];
+		if (!result) continue;
+		const relevance = relevanceLabels.get(result.document.id) || 0;
+		dcg += (Math.pow(2, relevance) - 1) / Math.log2(i + 2);
+	}
+	const idealOrder = Array.from(relevanceLabels.values()).sort((a, b) => b - a);
+	let idcg = 0;
+	const idealLength = k ? Math.min(k, idealOrder.length) : idealOrder.length;
+	for (let i = 0; i < idealLength; i++) {
+		const idealScore = idealOrder[i] ?? 0;
+		idcg += (Math.pow(2, idealScore) - 1) / Math.log2(i + 2);
+	}
+	return idcg > 0 ? dcg / idcg : 0;
+}
+/**
+* Calculate Mean Reciprocal Rank (MRR)
+*/
+function calculateMRR(results, relevantIds) {
+	for (let i = 0; i < results.length; i++) {
+		const result = results[i];
+		if (result && relevantIds.has(result.document.id)) return 1 / (i + 1);
+	}
+	return 0;
+}
+
+//#endregion
+export { createRerankPipeline as a, createCrossEncoderReranker as i, calculateNDCG as n, reciprocalRankFusion as o, createCohereReranker as r, vectorResultsToDocuments as s, calculateMRR as t };
+//# sourceMappingURL=reranking-Ck8aKZW7.mjs.map