crw-sdk 0.22.0 → 0.23.0

package/README.md

@@ -61,11 +61,13 @@ CRW_LOCAL=1 node app.js
 ¹ Local search needs a SearXNG URL configured on the engine.
 
 ```ts
-// Structured extraction:
-const data = await crw.extract({
+// Structured extraction (async job → per-URL results array):
+const results = await crw.extract({
   urls: ["https://example.com"],
   schema: { type: "object", properties: { title: { type: "string" } } },
 });
+// results: [{ url, status, data, error, llmUsage }]
+for (const r of results) if (r.status === "completed") console.log(r.url, r.data);
 
 // Parse a PDF:
 import { readFileSync } from "node:fs";

package/dist/cjs/client.js

@@ -170,19 +170,27 @@ class CrwClient {
         Object.assign(args, rest);
         return this.localTransport().toolCall("crw_parse_file", args);
     }
-    /** Structured LLM extraction across URLs (HTTP mode only). */
+    /**
+     * Native multi-URL structured extraction (HTTP mode only). Starts an async
+     * `/v1/extract` job and polls until complete, returning a per-URL results
+     * array (`[{ url, status, data, error, llmUsage }]`) in request order.
+     */
     async extract(opts) {
         if (!this.apiUrl)
             throw new errors_js_1.CrwError(httpOnlyHint("extract", "LLM extract job endpoint"));
-        const { urls, prompt, schema, systemPrompt, pollInterval = 2, timeout = 120 } = opts;
+        const { urls, prompt, schema, llmApiKey, llmProvider, llmModel, pollInterval = 2, timeout = 120 } = opts;
         const body = { urls: [...urls] };
         if (prompt !== undefined)
             body.prompt = prompt;
         if (schema !== undefined)
             body.schema = schema;
-        if (systemPrompt !== undefined)
-            body.systemPrompt = systemPrompt;
-        const start = await this.httpRequest("POST", "/v2/extract", body, { raw: true });
+        if (llmApiKey !== undefined)
+            body.llmApiKey = llmApiKey;
+        if (llmProvider !== undefined)
+            body.llmProvider = llmProvider;
+        if (llmModel !== undefined)
+            body.llmModel = llmModel;
+        const start = await this.httpRequest("POST", "/v1/extract", body, { raw: true });
         const jobId = start.id;
         if (!jobId)
             throw new errors_js_1.CrwError(`extract did not return job ID: ${JSON.stringify(start)}`);
@@ -190,9 +198,9 @@ class CrwClient {
         for (;;) {
             if (Date.now() > deadline)
                 throw new errors_js_1.CrwTimeoutError(`Extract ${jobId} timed out after ${timeout}s`);
-            const status = await this.httpRequest("GET", `/v2/extract/${jobId}`, undefined, { raw: true, checkSuccess: false });
+            const status = await this.httpRequest("GET", `/v1/extract/${jobId}`, undefined, { raw: true, checkSuccess: false });
             if (status.status === "completed")
-                return status.data ?? {};
+                return status.results ?? [];
             if (status.status === "failed")
                 throw new errors_js_1.CrwError(`Extract failed: ${status.error ?? "unknown"}`);
             await sleep(pollInterval * 1000);

package/dist/esm/client.d.ts

@@ -37,7 +37,11 @@ export declare class CrwClient {
      * Parse a document (PDF) into markdown / structured JSON. Works in both modes.
      */
     parseFile(content: Uint8Array, opts?: ParseFileOptions): Promise<ParseResult>;
-    /** Structured LLM extraction across URLs (HTTP mode only). */
+    /**
+     * Native multi-URL structured extraction (HTTP mode only). Starts an async
+     * `/v1/extract` job and polls until complete, returning a per-URL results
+     * array (`[{ url, status, data, error, llmUsage }]`) in request order.
+     */
     extract(opts: ExtractOptions): Promise<ExtractResult>;
     /** Scrape many URLs in one async batch job (HTTP mode only). */
     batchScrape(urls: string[], opts?: BatchScrapeOptions): Promise<BatchResult>;

package/dist/esm/client.js

@@ -167,19 +167,27 @@ export class CrwClient {
         Object.assign(args, rest);
         return this.localTransport().toolCall("crw_parse_file", args);
     }
-    /** Structured LLM extraction across URLs (HTTP mode only). */
+    /**
+     * Native multi-URL structured extraction (HTTP mode only). Starts an async
+     * `/v1/extract` job and polls until complete, returning a per-URL results
+     * array (`[{ url, status, data, error, llmUsage }]`) in request order.
+     */
     async extract(opts) {
         if (!this.apiUrl)
             throw new CrwError(httpOnlyHint("extract", "LLM extract job endpoint"));
-        const { urls, prompt, schema, systemPrompt, pollInterval = 2, timeout = 120 } = opts;
+        const { urls, prompt, schema, llmApiKey, llmProvider, llmModel, pollInterval = 2, timeout = 120 } = opts;
         const body = { urls: [...urls] };
         if (prompt !== undefined)
             body.prompt = prompt;
         if (schema !== undefined)
             body.schema = schema;
-        if (systemPrompt !== undefined)
-            body.systemPrompt = systemPrompt;
-        const start = await this.httpRequest("POST", "/v2/extract", body, { raw: true });
+        if (llmApiKey !== undefined)
+            body.llmApiKey = llmApiKey;
+        if (llmProvider !== undefined)
+            body.llmProvider = llmProvider;
+        if (llmModel !== undefined)
+            body.llmModel = llmModel;
+        const start = await this.httpRequest("POST", "/v1/extract", body, { raw: true });
         const jobId = start.id;
         if (!jobId)
             throw new CrwError(`extract did not return job ID: ${JSON.stringify(start)}`);
@@ -187,9 +195,9 @@ export class CrwClient {
         for (;;) {
             if (Date.now() > deadline)
                 throw new CrwTimeoutError(`Extract ${jobId} timed out after ${timeout}s`);
-            const status = await this.httpRequest("GET", `/v2/extract/${jobId}`, undefined, { raw: true, checkSuccess: false });
+            const status = await this.httpRequest("GET", `/v1/extract/${jobId}`, undefined, { raw: true, checkSuccess: false });
             if (status.status === "completed")
-                return status.data ?? {};
+                return status.results ?? [];
             if (status.status === "failed")
                 throw new CrwError(`Extract failed: ${status.error ?? "unknown"}`);
             await sleep(pollInterval * 1000);

package/dist/esm/types.d.ts

@@ -60,7 +60,10 @@ export interface ExtractOptions {
     urls: string[];
     prompt?: string;
     schema?: Json;
-    systemPrompt?: string;
+    /** BYOK: use your own LLM key/provider/model instead of the server's. */
+    llmApiKey?: string;
+    llmProvider?: string;
+    llmModel?: string;
     pollInterval?: number;
     timeout?: number;
 }
@@ -80,7 +83,8 @@ export type ScrapeResult = Json;
 export type CrawlResult = Json[];
 export type SearchResult = Json | Json[];
 export type ParseResult = Json;
-export type ExtractResult = Json;
+/** Native `/v1/extract` returns one result object per URL, in request order. */
+export type ExtractResult = Json[];
 export type BatchResult = Json[];
 export type Capabilities = Json;
 export type DiffResult = Json;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "crw-sdk",
-  "version": "0.22.0",
+  "version": "0.23.0",
   "description": "TypeScript/JavaScript SDK for CRW — scrape, crawl, map, search, parse, and extract any website",
   "license": "MIT",
   "homepage": "https://github.com/us/crw",