@babav/knowledge-core-client 0.6.0 → 0.8.0

package/dist/index.d.ts

@@ -73,7 +73,6 @@ export interface QueryResponse {
     groundedness: Groundedness | null;
     usage: Record<string, number | null>;
     standalone_query: string | null;
-    hops: string[];
 }
 export interface Citation {
     quote: string;
@@ -241,12 +240,6 @@ declare class HttpBase {
     protected pageAll<T>(path: string, query?: RequestOpts["query"]): Promise<T[]>;
 }
 export interface StreamHandlers {
-    onHop?: (d: {
-        n: number;
-        subquery: string;
-        status: string;
-        retrieval_contents?: number;
-    }) => void;
     onSources?: (d: {
         retrieval_contents: RetrievalContent[];
     }) => void;
@@ -301,9 +294,11 @@ export declare class KnowledgeCoreClient extends HttpBase {
             limit?: number;
             cursor?: string;
         }) => Promise<Page<Document>>;
-        /** Upload a file and ingest it ASYNCHRONOUSLY — resolves with the created
-         *  document at status `pending` (HTTP 202); track via documents.get() polling
-         *  the status to `indexed`/`failed`, or the tenant result webhook. */
+        /** Ingest a document ASYNCHRONOUSLY — resolves with the created document at
+         *  status `pending` (HTTP 202); track via documents.get() polling to
+         *  `indexed`/`failed`, or the tenant result webhook. Auto-routes by size: files
+         *  over ~30 MB go via the signed-URL upload path (Cloud Run caps requests at
+         *  ~32 MB), smaller files via one multipart request. One method, any size. */
         ingestDocument: (id: UUID, a: {
             file: FileData;
             filename: string;

package/dist/index.js

@@ -148,6 +148,16 @@ function toBlob(data, contentType) {
         return data;
     return new Blob([data], { type: contentType ?? "application/octet-stream" });
 }
+function fileSize(data) {
+    if (data instanceof Blob)
+        return data.size;
+    if (data instanceof ArrayBuffer)
+        return data.byteLength;
+    return data.byteLength;
+}
+// Cloud Run caps request bodies at ~32 MB; over this, ingestDocument auto-routes to
+// the signed-URL upload path instead of a single multipart request.
+const MULTIPART_MAX_BYTES = 30 * 1024 * 1024;
 // ---------------------------------------------------------------------------
 // Tenant client (data ops) — use a TENANT key
 // ---------------------------------------------------------------------------
@@ -203,10 +213,14 @@ export class KnowledgeCoreClient extends HttpBase {
         countDocuments: (id) => this.request("GET", `/v1/corpora/${id}/documents/count`),
         /** Documents in the corpus whose filename contains `q` (case-insensitive), paginated. */
         searchDocuments: (id, q, opts) => this.request("GET", `/v1/corpora/${id}/documents/search`, { query: { q, ...opts } }),
-        /** Upload a file and ingest it ASYNCHRONOUSLY — resolves with the created
-         *  document at status `pending` (HTTP 202); track via documents.get() polling
-         *  the status to `indexed`/`failed`, or the tenant result webhook. */
+        /** Ingest a document ASYNCHRONOUSLY — resolves with the created document at
+         *  status `pending` (HTTP 202); track via documents.get() polling to
+         *  `indexed`/`failed`, or the tenant result webhook. Auto-routes by size: files
+         *  over ~30 MB go via the signed-URL upload path (Cloud Run caps requests at
+         *  ~32 MB), smaller files via one multipart request. One method, any size. */
         ingestDocument: (id, a) => {
+            if (fileSize(a.file) > MULTIPART_MAX_BYTES)
+                return this.corpora.uploadDocument(id, a);
             const fd = new FormData();
             fd.append("file", toBlob(a.file, a.content_type), a.filename);
             if (a.custom_metadata)
@@ -354,9 +368,6 @@ function dispatchSse(frame, h) {
     const data = safeJson(dataLines.join("\n"));
     h.onEvent?.(event, data);
     switch (event) {
-        case "hop":
-            h.onHop?.(data);
-            break;
         case "retrieval_contents":
             h.onSources?.(data);
             break;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@babav/knowledge-core-client",
-  "version": "0.6.0",
+  "version": "0.8.0",
   "description": "TypeScript client for the Babav Knowledge Core API (Deno + Node 18+, zero deps).",
   "type": "module",
   "main": "./dist/index.js",

package/src/index.ts

@@ -85,7 +85,6 @@ export interface QueryResponse {
   groundedness: Groundedness | null;
   usage: Record<string, number | null>;
   standalone_query: string | null;
-  hops: string[];
 }
 
 export interface Citation {
@@ -364,11 +363,20 @@ function toBlob(data: FileData, contentType?: string): Blob {
   return new Blob([data as BlobPart], { type: contentType ?? "application/octet-stream" });
 }
 
+function fileSize(data: FileData): number {
+  if (data instanceof Blob) return data.size;
+  if (data instanceof ArrayBuffer) return data.byteLength;
+  return (data as Uint8Array).byteLength;
+}
+
+// Cloud Run caps request bodies at ~32 MB; over this, ingestDocument auto-routes to
+// the signed-URL upload path instead of a single multipart request.
+const MULTIPART_MAX_BYTES = 30 * 1024 * 1024;
+
 // ---------------------------------------------------------------------------
 // SSE handlers
 // ---------------------------------------------------------------------------
 export interface StreamHandlers {
-  onHop?: (d: { n: number; subquery: string; status: string; retrieval_contents?: number }) => void;
   onSources?: (d: { retrieval_contents: RetrievalContent[] }) => void;
   onToken?: (text: string) => void;
   onFinal?: (d: QueryResponse) => void;
@@ -440,10 +448,13 @@ export class KnowledgeCoreClient extends HttpBase {
     /** Documents in the corpus whose filename contains `q` (case-insensitive), paginated. */
     searchDocuments: (id: UUID, q: string, opts?: { limit?: number; cursor?: string }) =>
       this.request<Page<Document>>("GET", `/v1/corpora/${id}/documents/search`, { query: { q, ...opts } }),
-    /** Upload a file and ingest it ASYNCHRONOUSLY — resolves with the created
-     *  document at status `pending` (HTTP 202); track via documents.get() polling
-     *  the status to `indexed`/`failed`, or the tenant result webhook. */
+    /** Ingest a document ASYNCHRONOUSLY — resolves with the created document at
+     *  status `pending` (HTTP 202); track via documents.get() polling to
+     *  `indexed`/`failed`, or the tenant result webhook. Auto-routes by size: files
+     *  over ~30 MB go via the signed-URL upload path (Cloud Run caps requests at
+     *  ~32 MB), smaller files via one multipart request. One method, any size. */
     ingestDocument: (id: UUID, a: { file: FileData; filename: string; content_type?: string; visibility?: Visibility; folder_id?: UUID; custom_metadata?: Record<string, unknown> }) => {
+      if (fileSize(a.file) > MULTIPART_MAX_BYTES) return this.corpora.uploadDocument(id, a);
       const fd = new FormData();
       fd.append("file", toBlob(a.file, a.content_type), a.filename);
       if (a.custom_metadata) fd.append("custom_metadata", JSON.stringify(a.custom_metadata));
@@ -612,7 +623,6 @@ function dispatchSse(frame: string, h: StreamHandlers): void {
   const data = safeJson(dataLines.join("\n"));
   h.onEvent?.(event, data);
   switch (event) {
-    case "hop": h.onHop?.(data as never); break;
     case "retrieval_contents": h.onSources?.(data as never); break;
     case "token": h.onToken?.((data as { text: string }).text); break;
     case "final": h.onFinal?.(data as QueryResponse); break;