unrag 0.2.6 → 0.2.7

package/registry/embedding/vertex.ts

@@ -1,7 +1,16 @@
-import { embed, embedMany } from "ai";
+import { embed, embedMany, type EmbeddingModel } from "ai";
 import type { EmbeddingProvider } from "../core/types";
 import { requireOptional } from "./_shared";
 
+/**
+ * Vertex AI provider module interface.
+ */
+interface VertexModule {
+  vertex: {
+    embeddingModel: (model: string) => EmbeddingModel<string>;
+  };
+}
+
 export type VertexEmbeddingTaskType =
   | "SEMANTIC_SIMILARITY"
   | "CLASSIFICATION"
@@ -47,7 +56,7 @@ const buildProviderOptions = (config: VertexEmbeddingConfig) => {
 export const createVertexEmbeddingProvider = (
   config: VertexEmbeddingConfig = {}
 ): EmbeddingProvider => {
-  const { vertex } = requireOptional<any>({
+  const { vertex } = requireOptional<VertexModule>({
     id: "@ai-sdk/google-vertex",
     installHint: "bun add @ai-sdk/google-vertex",
     providerName: "vertex",

package/registry/embedding/voyage.ts

@@ -2,6 +2,17 @@ import { embed, embedMany, type EmbeddingModel } from "ai";
 import type { EmbeddingProvider, ImageEmbeddingInput } from "../core/types";
 import { requireOptional } from "./_shared";
 
+/**
+ * Voyage AI provider module interface.
+ */
+interface VoyageModule {
+  voyage: {
+    embeddingModel?: (model: string) => EmbeddingModel<string>;
+    textEmbeddingModel?: (model: string) => EmbeddingModel<string>;
+    multimodalEmbeddingModel?: (model: string) => EmbeddingModel<unknown>;
+  };
+}
+
 type BaseConfig = {
   model?: string;
   timeoutMs?: number;
@@ -44,7 +55,7 @@ const defaultImageValue = (input: ImageEmbeddingInput) => {
 export const createVoyageEmbeddingProvider = (
   config: VoyageEmbeddingConfig = {}
 ): EmbeddingProvider => {
-  const { voyage } = requireOptional<any>({
+  const { voyage } = requireOptional<VoyageModule>({
     id: "voyage-ai-provider",
     installHint: "bun add voyage-ai-provider",
     providerName: "voyage",
@@ -57,15 +68,14 @@ export const createVoyageEmbeddingProvider = (
     (type === "multimodal" ? DEFAULT_MULTIMODAL_MODEL : DEFAULT_TEXT_MODEL);
   const timeoutMs = config.timeoutMs;
 
-  const voyageProvider = voyage as any;
   const textEmbeddingModel =
     type === "multimodal"
       ? undefined
-      : typeof voyageProvider.embeddingModel === "function"
-        ? voyageProvider.embeddingModel(model)
-        : voyageProvider.textEmbeddingModel(model);
+      : typeof voyage.embeddingModel === "function"
+        ? voyage.embeddingModel(model)
+        : voyage.textEmbeddingModel?.(model);
   const multimodalEmbeddingModel =
-    type === "multimodal" ? (voyage as any).multimodalEmbeddingModel(model) : undefined;
+    type === "multimodal" ? voyage.multimodalEmbeddingModel?.(model) : undefined;
 
   // AI SDK 6 types only accept string inputs; cast multimodal models/values.
   const multimodalModel = multimodalEmbeddingModel as unknown as EmbeddingModel;

package/registry/extractors/audio-transcribe/index.ts

@@ -1,7 +1,21 @@
-import { experimental_transcribe as transcribe } from "ai";
-import type { AssetExtractor } from "../../core/types";
+import { experimental_transcribe as transcribe, type TranscriptionModel } from "ai";
+import type { AssetExtractor, ExtractedTextItem } from "../../core/types";
 import { getAssetBytes } from "../_shared/fetch";
 
+/**
+ * Model reference type that accepts both string gateway IDs and TranscriptionModel instances.
+ */
+type TranscriptionModelRef = string | TranscriptionModel;
+
+/**
+ * Transcription segment from the AI SDK.
+ */
+interface TranscriptionSegment {
+  text?: string;
+  startSecond?: number;
+  endSecond?: number;
+}
+
 /**
  * Audio transcription via the AI SDK `transcribe()` API.
  */
@@ -25,43 +39,45 @@ export function createAudioTranscribeExtractor(): AssetExtractor {
       const abortSignal = AbortSignal.timeout(cfg.timeoutMs);
 
       const result = await transcribe({
-        model: cfg.model as any,
+        model: cfg.model as TranscriptionModelRef,
         audio: bytes,
         abortSignal,
       });
 
-      const segments: any[] = Array.isArray((result as any)?.segments)
-        ? (result as any).segments
+      const segments: TranscriptionSegment[] = Array.isArray(result.segments)
+        ? result.segments
         : [];
 
       if (segments.length > 0) {
+        const textItems: ExtractedTextItem[] = segments
+          .map((s, i) => {
+            const t = String(s?.text ?? "").trim();
+            if (!t) return null;
+            const start = Number(s?.startSecond ?? NaN);
+            const end = Number(s?.endSecond ?? NaN);
+            return {
+              label: `segment-${i + 1}`,
+              content: t,
+              ...(Number.isFinite(start) && Number.isFinite(end)
+                ? { timeRangeSec: [start, end] as [number, number] }
+                : {}),
+            };
+          })
+          .filter((item): item is ExtractedTextItem => item !== null);
+
         return {
-          texts: segments
-            .map((s, i) => {
-              const t = String(s?.text ?? "").trim();
-              if (!t) return null;
-              const start = Number(s?.startSecond ?? NaN);
-              const end = Number(s?.endSecond ?? NaN);
-              return {
-                label: `segment-${i + 1}`,
-                content: t,
-                ...(Number.isFinite(start) && Number.isFinite(end)
-                  ? { timeRangeSec: [start, end] as [number, number] }
-                  : {}),
-              };
-            })
-            .filter(Boolean) as any,
+          texts: textItems,
           diagnostics: {
             model: cfg.model,
             seconds:
-              typeof (result as any)?.durationInSeconds === "number"
-                ? (result as any).durationInSeconds
+              typeof result.durationInSeconds === "number"
+                ? result.durationInSeconds
                 : undefined,
           },
         };
       }
 
-      const text = String((result as any)?.text ?? "").trim();
+      const text = (result.text ?? "").trim();
       if (!text) return { texts: [], diagnostics: { model: cfg.model } };
 
       return {

package/registry/extractors/file-docx/index.ts

@@ -3,6 +3,13 @@ import { getAssetBytes } from "../_shared/fetch";
 import { extFromFilename, normalizeMediaType } from "../_shared/media";
 import { capText } from "../_shared/text";
 
+/**
+ * Minimal mammoth module interface.
+ */
+interface MammothModule {
+  extractRawText(options: { arrayBuffer: ArrayBuffer }): Promise<{ value?: string }>;
+}
+
 const DOCX_MEDIA =
   "application/vnd.openxmlformats-officedocument.wordprocessingml.document";
 
@@ -33,7 +40,7 @@ export function createFileDocxExtractor(): AssetExtractor {
       });
 
       // Dynamic import so the core package can be used without mammoth unless this extractor is installed.
-      const mammoth: any = await import("mammoth");
+      const mammoth = (await import("mammoth")) as MammothModule;
       const arrayBuffer = bytes.buffer.slice(
         bytes.byteOffset,
         bytes.byteOffset + bytes.byteLength

package/registry/extractors/file-pptx/index.ts

@@ -3,6 +3,27 @@ import { getAssetBytes } from "../_shared/fetch";
 import { extFromFilename, normalizeMediaType } from "../_shared/media";
 import { capText } from "../_shared/text";
 
+/**
+ * Zip file entry interface.
+ */
+interface ZipFile {
+  async(type: "string"): Promise<string>;
+}
+
+/**
+ * JSZip instance interface.
+ */
+interface JSZipInstance {
+  files: Record<string, ZipFile>;
+}
+
+/**
+ * JSZip constructor interface.
+ */
+interface JSZipConstructor {
+  loadAsync(data: Uint8Array): Promise<JSZipInstance>;
+}
+
 const PPTX_MEDIA =
   "application/vnd.openxmlformats-officedocument.presentationml.presentation";
 
@@ -41,7 +62,7 @@ export function createFilePptxExtractor(): AssetExtractor {
       });
 
       // Dynamic import to avoid hard dependency unless installed.
-      const JSZip: any = (await import("jszip")).default;
+      const JSZip = (await import("jszip")).default as unknown as JSZipConstructor;
       const zip = await JSZip.loadAsync(bytes);
 
       const slidePaths = Object.keys(zip.files).filter((p) =>

package/registry/extractors/file-xlsx/index.ts

@@ -3,6 +3,29 @@ import { getAssetBytes } from "../_shared/fetch";
 import { extFromFilename, normalizeMediaType } from "../_shared/media";
 import { capText } from "../_shared/text";
 
+/**
+ * XLSX sheet interface.
+ */
+type XLSXSheet = unknown;
+
+/**
+ * XLSX workbook interface.
+ */
+interface XLSXWorkbook {
+  SheetNames?: string[];
+  Sheets?: Record<string, XLSXSheet>;
+}
+
+/**
+ * Minimal xlsx module interface.
+ */
+interface XLSXModule {
+  read(data: Buffer, options: { type: string }): XLSXWorkbook;
+  utils: {
+    sheet_to_csv(sheet: XLSXSheet): string;
+  };
+}
+
 const XLSX_MEDIA =
   "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet";
 
@@ -32,7 +55,7 @@ export function createFileXlsxExtractor(): AssetExtractor {
         defaultMediaType: XLSX_MEDIA,
       });
 
-      const xlsx: any = await import("xlsx");
+      const xlsx = (await import("xlsx")) as XLSXModule;
       const wb = xlsx.read(Buffer.from(bytes), { type: "buffer" });
 
       const parts: string[] = [];

package/registry/extractors/image-caption-llm/index.ts

@@ -1,9 +1,14 @@
-import { generateText } from "ai";
+import { generateText, type LanguageModel } from "ai";
 import type { AssetExtractor } from "../../core/types";
 import { getAssetBytes } from "../_shared/fetch";
 import { normalizeMediaType } from "../_shared/media";
 import { capText } from "../_shared/text";
 
+/**
+ * Model reference type that accepts both string gateway IDs and LanguageModel instances.
+ */
+type ModelRef = string | LanguageModel;
+
 /**
  * Caption generation for images via a vision-capable LLM.
  *
@@ -29,7 +34,7 @@ export function createImageCaptionLlmExtractor(): AssetExtractor {
       const abortSignal = AbortSignal.timeout(cfg.timeoutMs);
 
       const result = await generateText({
-        model: cfg.model as any,
+        model: cfg.model as ModelRef,
         abortSignal,
         messages: [
           {
@@ -46,7 +51,7 @@ export function createImageCaptionLlmExtractor(): AssetExtractor {
         ],
       });
 
-      const caption = String((result as any)?.text ?? "").trim();
+      const caption = (result.text ?? "").trim();
       if (!caption) return { texts: [], diagnostics: { model: cfg.model } };
 
       return {

package/registry/extractors/image-ocr/index.ts

@@ -1,13 +1,18 @@
-import { generateText } from "ai";
+import { generateText, type LanguageModel } from "ai";
 import type { AssetExtractor } from "../../core/types";
 import { getAssetBytes } from "../_shared/fetch";
 import { normalizeMediaType } from "../_shared/media";
 import { capText } from "../_shared/text";
 
+/**
+ * Model reference type that accepts both string gateway IDs and LanguageModel instances.
+ */
+type ModelRef = string | LanguageModel;
+
 /**
  * Image OCR via a vision-capable LLM.
  *
- * This extractor is intended for screenshots, charts, diagrams, and any image with embedded text.
+ * This extractor is intended for screenshots, charts, diagrams, and images with embedded text.
  */
 export function createImageOcrExtractor(): AssetExtractor {
   return {
@@ -29,7 +34,7 @@ export function createImageOcrExtractor(): AssetExtractor {
       const abortSignal = AbortSignal.timeout(cfg.timeoutMs);
 
       const result = await generateText({
-        model: cfg.model as any,
+        model: cfg.model as ModelRef,
         abortSignal,
         messages: [
           {
@@ -46,7 +51,7 @@ export function createImageOcrExtractor(): AssetExtractor {
         ],
       });
 
-      const text = String((result as any)?.text ?? "").trim();
+      const text = (result.text ?? "").trim();
       if (!text) return { texts: [], diagnostics: { model: cfg.model } };
 
       return {

package/registry/extractors/pdf-llm/index.ts

@@ -1,9 +1,14 @@
-import { generateText } from "ai";
+import { generateText, type LanguageModel } from "ai";
 import type { AssetData, AssetExtractor, AssetFetchConfig } from "../../core/types";
 import { getAssetBytes } from "../_shared/fetch";
 import { normalizeMediaType } from "../_shared/media";
 import { capText } from "../_shared/text";
 
+/**
+ * Model reference type that accepts both string gateway IDs and LanguageModel instances.
+ */
+type ModelRef = string | LanguageModel;
+
 async function getPdfBytes(args: {
   data: AssetData;
   fetchConfig: AssetFetchConfig;
@@ -49,8 +54,8 @@ export function createPdfLlmExtractor(): AssetExtractor {
       const abortSignal = AbortSignal.timeout(llm.timeoutMs);
 
       const result = await generateText({
-        // Intentionally allow string model ids for AI Gateway usage.
-        model: llm.model as any,
+        // String model IDs are supported for AI Gateway routing.
+        model: llm.model as ModelRef,
         abortSignal,
         messages: [
           {
@@ -68,7 +73,7 @@ export function createPdfLlmExtractor(): AssetExtractor {
         ],
       });
 
-      const text = String((result as any)?.text ?? "").trim();
+      const text = (result.text ?? "").trim();
       if (!text) return { texts: [], diagnostics: { model: llm.model } };
 
       const capped = capText(text, llm.maxOutputChars);

package/registry/extractors/pdf-text-layer/index.ts

@@ -2,6 +2,27 @@ import type { AssetExtractor } from "../../core/types";
 import { getAssetBytes } from "../_shared/fetch";
 import { capText } from "../_shared/text";
 
+/**
+ * Text content item from pdfjs-dist.
+ */
+interface PdfTextItem {
+  str?: string;
+}
+
+/**
+ * Minimal pdfjs-dist module interface.
+ */
+interface PdfJsModule {
+  getDocument(params: { data: Uint8Array }): {
+    promise: Promise<{
+      numPages: number;
+      getPage(pageNum: number): Promise<{
+        getTextContent(): Promise<{ items?: PdfTextItem[] }>;
+      }>;
+    }>;
+  };
+}
+
 /**
  * Fast/cheap PDF extraction using the PDF's built-in text layer.
  *
@@ -29,7 +50,7 @@ export function createPdfTextLayerExtractor(): AssetExtractor {
       });
 
       // Dynamic import so the core package can be used without pdfjs unless this extractor is installed.
-      const pdfjs: any = await import("pdfjs-dist/legacy/build/pdf.mjs");
+      const pdfjs = (await import("pdfjs-dist/legacy/build/pdf.mjs")) as PdfJsModule;
 
       const doc = await pdfjs.getDocument({ data: bytes }).promise;
       const totalPages: number = Number(doc?.numPages ?? 0);
@@ -42,7 +63,7 @@ export function createPdfTextLayerExtractor(): AssetExtractor {
       for (let pageNum = 1; pageNum <= maxPages; pageNum++) {
         const page = await doc.getPage(pageNum);
         const textContent = await page.getTextContent();
-        const items: any[] = Array.isArray(textContent?.items)
+        const items: PdfTextItem[] = Array.isArray(textContent?.items)
           ? textContent.items
           : [];
         const pageText = items

package/registry/extractors/video-frames/index.ts

@@ -1,4 +1,4 @@
-import { generateText } from "ai";
+import { generateText, type LanguageModel } from "ai";
 import { spawn } from "node:child_process";
 import { mkdir, readdir, readFile, rm, writeFile } from "node:fs/promises";
 import os from "node:os";
@@ -7,6 +7,11 @@ import type { AssetExtractor } from "../../core/types";
 import { getAssetBytes } from "../_shared/fetch";
 import { capText } from "../_shared/text";
 
+/**
+ * Model reference type that accepts both string gateway IDs and LanguageModel instances.
+ */
+type ModelRef = string | LanguageModel;
+
 const run = async (cmd: string, args: string[], opts: { cwd: string }) => {
   return await new Promise<{ stdout: string; stderr: string }>((resolve, reject) => {
     const child = spawn(cmd, args, { cwd: opts.cwd, stdio: ["ignore", "pipe", "pipe"] });
@@ -87,7 +92,7 @@ export function createVideoFramesExtractor(): AssetExtractor {
 
           const imgBytes = await readFile(path.join(tmpDir, f));
           const result = await generateText({
-            model: cfg.model as any,
+            model: cfg.model as ModelRef,
             abortSignal: abortPerFrame(cfg.timeoutMs),
             messages: [
               {
@@ -100,7 +105,7 @@ export function createVideoFramesExtractor(): AssetExtractor {
             ],
           });
 
-          const t = String((result as any)?.text ?? "").trim();
+          const t = (result.text ?? "").trim();
           if (!t) continue;
 
           const capped = capText(t, cfg.maxOutputChars - totalChars);

package/registry/extractors/video-transcribe/index.ts

@@ -1,7 +1,21 @@
-import { experimental_transcribe as transcribe } from "ai";
-import type { AssetExtractor } from "../../core/types";
+import { experimental_transcribe as transcribe, type TranscriptionModel } from "ai";
+import type { AssetExtractor, ExtractedTextItem } from "../../core/types";
 import { getAssetBytes } from "../_shared/fetch";
 
+/**
+ * Model reference type that accepts both string gateway IDs and TranscriptionModel instances.
+ */
+type TranscriptionModelRef = string | TranscriptionModel;
+
+/**
+ * Transcription segment from the AI SDK.
+ */
+interface TranscriptionSegment {
+  text?: string;
+  startSecond?: number;
+  endSecond?: number;
+}
+
 /**
  * Video transcription by sending the video file to the AI SDK transcription API.
  *
@@ -28,43 +42,45 @@ export function createVideoTranscribeExtractor(): AssetExtractor {
       const abortSignal = AbortSignal.timeout(cfg.timeoutMs);
 
       const result = await transcribe({
-        model: cfg.model as any,
-        audio: bytes as any,
+        model: cfg.model as TranscriptionModelRef,
+        audio: bytes,
         abortSignal,
       });
 
-      const segments: any[] = Array.isArray((result as any)?.segments)
-        ? (result as any).segments
+      const segments: TranscriptionSegment[] = Array.isArray(result.segments)
+        ? result.segments
         : [];
 
       if (segments.length > 0) {
+        const textItems: ExtractedTextItem[] = segments
+          .map((s, i) => {
+            const t = String(s?.text ?? "").trim();
+            if (!t) return null;
+            const start = Number(s?.startSecond ?? NaN);
+            const end = Number(s?.endSecond ?? NaN);
+            return {
+              label: `segment-${i + 1}`,
+              content: t,
+              ...(Number.isFinite(start) && Number.isFinite(end)
+                ? { timeRangeSec: [start, end] as [number, number] }
+                : {}),
+            };
+          })
+          .filter((item): item is ExtractedTextItem => item !== null);
+
         return {
-          texts: segments
-            .map((s, i) => {
-              const t = String(s?.text ?? "").trim();
-              if (!t) return null;
-              const start = Number(s?.startSecond ?? NaN);
-              const end = Number(s?.endSecond ?? NaN);
-              return {
-                label: `segment-${i + 1}`,
-                content: t,
-                ...(Number.isFinite(start) && Number.isFinite(end)
-                  ? { timeRangeSec: [start, end] as [number, number] }
-                  : {}),
-              };
-            })
-            .filter(Boolean) as any,
+          texts: textItems,
           diagnostics: {
             model: cfg.model,
             seconds:
-              typeof (result as any)?.durationInSeconds === "number"
-                ? (result as any).durationInSeconds
+              typeof result.durationInSeconds === "number"
+                ? result.durationInSeconds
                 : undefined,
           },
         };
       }
 
-      const text = String((result as any)?.text ?? "").trim();
+      const text = (result.text ?? "").trim();
       if (!text) return { texts: [], diagnostics: { model: cfg.model } };
 
       return {

package/registry/manifest.json

@@ -33,7 +33,7 @@
       "inputModes": ["file", "url", "buffer"],
       "output": "text (markdown)",
       "docsPath": "/docs/extractors/pdf/llm",
-      "deps": { "ai": "^5.0.113" },
+      "deps": { "ai": "^6.0.3" },
       "devDeps": {},
       "factory": "createPdfLlmExtractor",
       "assetProcessingFlagKeys": ["pdf_llmExtraction"]
@@ -70,7 +70,7 @@
       "inputModes": ["file", "url", "buffer"],
       "output": "text",
       "docsPath": "/docs/extractors/image/ocr",
-      "deps": { "ai": "^5.0.113" },
+      "deps": { "ai": "^6.0.0" },
       "devDeps": {},
       "factory": "createImageOcrExtractor",
       "assetProcessingFlagKeys": ["image_ocr"]
@@ -88,7 +88,7 @@
       "inputModes": ["file", "url", "buffer"],
       "output": "caption",
       "docsPath": "/docs/extractors/image/caption-llm",
-      "deps": { "ai": "^5.0.113" },
+      "deps": { "ai": "^6.0.0" },
       "devDeps": {},
       "factory": "createImageCaptionLlmExtractor",
       "assetProcessingFlagKeys": ["image_captionLlm"]
@@ -106,7 +106,7 @@
       "inputModes": ["file", "url", "buffer"],
       "output": "transcript",
       "docsPath": "/docs/extractors/audio/transcribe",
-      "deps": { "ai": "^5.0.113" },
+      "deps": { "ai": "^6.0.0" },
       "devDeps": {},
       "factory": "createAudioTranscribeExtractor",
       "assetProcessingFlagKeys": ["audio_transcription"]
@@ -124,7 +124,7 @@
       "inputModes": ["file", "url", "buffer"],
       "output": "transcript",
       "docsPath": "/docs/extractors/video/transcribe",
-      "deps": { "ai": "^5.0.113" },
+      "deps": { "ai": "^6.0.0" },
       "devDeps": {},
       "factory": "createVideoTranscribeExtractor",
       "assetProcessingFlagKeys": ["video_transcription"]
@@ -143,7 +143,7 @@
       "inputModes": ["file", "url", "buffer"],
       "output": "frame descriptions",
       "docsPath": "/docs/extractors/video/frames",
-      "deps": { "ai": "^5.0.113" },
+      "deps": { "ai": "^6.0.0" },
       "devDeps": {},
       "factory": "createVideoFramesExtractor",
       "assetProcessingFlagKeys": ["video_frames"]

package/registry/store/drizzle-postgres-pgvector/store.ts

@@ -1,9 +1,26 @@
 import { documents, chunks, embeddings } from "./schema";
 import type { Chunk, VectorStore } from "../../core/types";
 import { eq, like, sql, type SQL } from "drizzle-orm";
-import type { PgDatabase } from "drizzle-orm/pg-core";
-
-type DrizzleDb = PgDatabase<any, any, any>;
+import type { PgDatabase, PgQueryResultHKT } from "drizzle-orm/pg-core";
+
+/**
+ * Accepts any Drizzle Postgres database instance regardless of schema type.
+ */
+type DrizzleDb = PgDatabase<PgQueryResultHKT, Record<string, unknown>>;
+
+/**
+ * Query row type for vector similarity search results.
+ */
+interface QueryRow {
+  id: string;
+  document_id: string;
+  source_id: string;
+  idx: number;
+  content: string;
+  token_count: number;
+  metadata: Record<string, unknown> | null;
+  score: number;
+}
 
 const sanitizeMetadata = (metadata: unknown) => {
   if (metadata === undefined) {
@@ -133,11 +150,11 @@ export const createDrizzleVectorStore = (db: DrizzleDb): VectorStore => ({
       `
     );
 
-    const rows = Array.isArray(result)
-      ? result
-      : ((result as { rows?: unknown[] }).rows ?? []);
+    const rows: QueryRow[] = Array.isArray(result)
+      ? (result as QueryRow[])
+      : ((result as { rows?: QueryRow[] }).rows ?? []);
 
-    return (rows as Array<Record<string, unknown>>).map((row) => ({
+    return rows.map((row) => ({
       id: String(row.id),
       documentId: String(row.document_id),
       sourceId: String(row.source_id),