@claritylabs/cl-sdk 0.7.5 → 0.8.0

package/dist/index.d.mts

@@ -12,7 +12,16 @@ type GenerateText = (params: {
     text: string;
     usage?: TokenUsage;
 }>;
-/** Callback to generate a typed object from a prompt + Zod schema. Provider-agnostic. */
+/**
+ * Callback to generate a typed object from a prompt + Zod schema. Provider-agnostic.
+ *
+ * The extraction pipeline passes document content via `providerOptions`:
+ * - `providerOptions.pdfBase64` — base64-encoded PDF to include as document context
+ * - `providerOptions.images` — `Array<{ imageBase64: string; mimeType: string }>` page images
+ *
+ * Your callback should check for these fields and include them as multi-part
+ * message content (e.g. file/image parts) when calling your AI provider.
+ */
 type GenerateObject<T = unknown> = (params: {
     prompt: string;
     system?: string;

package/dist/index.d.ts

@@ -12,7 +12,16 @@ type GenerateText = (params: {
     text: string;
     usage?: TokenUsage;
 }>;
-/** Callback to generate a typed object from a prompt + Zod schema. Provider-agnostic. */
+/**
+ * Callback to generate a typed object from a prompt + Zod schema. Provider-agnostic.
+ *
+ * The extraction pipeline passes document content via `providerOptions`:
+ * - `providerOptions.pdfBase64` — base64-encoded PDF to include as document context
+ * - `providerOptions.images` — `Array<{ imageBase64: string; mimeType: string }>` page images
+ *
+ * Your callback should check for these fields and include them as multi-part
+ * message content (e.g. file/image parts) when calling your AI provider.
+ */
 type GenerateObject<T = unknown> = (params: {
     prompt: string;
     system?: string;

package/dist/index.js

@@ -1758,19 +1758,28 @@ async function runExtractor(params) {
     maxTokens = 4096,
     providerOptions
   } = params;
-  const pagesPdf = await extractPageRange(pdfBase64, startPage, endPage);
-  const fullPrompt = convertPdfToImages ? `${prompt}
-
-[Document pages ${startPage}-${endPage} are provided as images above.]` : `${prompt}
+  const extractorProviderOptions = { ...providerOptions };
+  let fullPrompt;
+  if (convertPdfToImages) {
+    const images = await convertPdfToImages(pdfBase64, startPage, endPage);
+    extractorProviderOptions.images = images;
+    fullPrompt = `${prompt}
+
+[Document pages ${startPage}-${endPage} are provided as images.]`;
+  } else {
+    const pagesPdf = await extractPageRange(pdfBase64, startPage, endPage);
+    extractorProviderOptions.pdfBase64 = pagesPdf;
+    fullPrompt = `${prompt}
 
-[Document pages ${startPage}-${endPage} are provided as a PDF file above.]`;
+[Document pages ${startPage}-${endPage} are provided as a PDF file.]`;
+  }
   const strictSchema = toStrictSchema(schema);
   const result = await withRetry(
     () => generateObject({
       prompt: fullPrompt,
       schema: strictSchema,
       maxTokens,
-      providerOptions
+      providerOptions: extractorProviderOptions
     })
   );
   return {
@@ -3590,7 +3599,7 @@ function createExtractor(config) {
           prompt: buildClassifyPrompt(),
           schema: ClassifyResultSchema,
           maxTokens: 512,
-          providerOptions
+          providerOptions: { ...providerOptions, pdfBase64 }
         },
         {
           fallback: { documentType: "policy", policyTypes: ["other"], confidence: 0 },
@@ -3634,7 +3643,7 @@ function createExtractor(config) {
           prompt: buildPlanPrompt(templateHints),
           schema: ExtractionPlanSchema,
           maxTokens: 2048,
-          providerOptions
+          providerOptions: { ...providerOptions, pdfBase64 }
         },
         {
           fallback: {