Nomad changes

docs/snippets/wasm/plugins/stateful_plugin.md

@@ -0,0 +1,79 @@
+# Stateful Post-Processor Plugin
+
+Create a stateful post-processor that maintains state across multiple extraction calls.
+
+```typescript title="WASM"
+import init, { registerPostProcessor, extractBytes } from "kreuzberg-wasm";
+
+await init();
+
+// Create a stateful post-processor using a closure
+function createStatefulProcessor() {
+  const state = {
+    extractionCount: 0,
+    totalChars: 0,
+    lastResult: null,
+  };
+
+  return {
+    processingStage: () => "post-extraction",
+    process: (extractionResult) => {
+      // Update state
+      state.extractionCount++;
+      state.totalChars += extractionResult.text?.length || 0;
+      state.lastResult = extractionResult;
+
+      // Enrich result with statistics
+      const enriched = {
+        ...extractionResult,
+        metadata: {
+          ...extractionResult.metadata,
+          extractionIndex: state.extractionCount,
+          cumulativeChars: state.totalChars,
+          averageDocLength: Math.round(state.totalChars / state.extractionCount),
+        },
+      };
+
+      console.log(
+        `[Extraction ${state.extractionCount}] ${enriched.text?.length || 0} chars, cumulative: ${state.totalChars}`,
+      );
+
+      return enriched;
+    },
+
+    // Optional: expose state for inspection
+    getState: () => state,
+  };
+}
+
+// Register the stateful processor
+const statefulProcessor = createStatefulProcessor();
+registerPostProcessor(statefulProcessor);
+
+// Multiple extractions use the same state
+async function processMultipleDocs() {
+  const docs = [
+    new Uint8Array([
+      /* Doc 1 */
+    ]),
+    new Uint8Array([
+      /* Doc 2 */
+    ]),
+    new Uint8Array([
+      /* Doc 3 */
+    ]),
+  ];
+
+  const results = [];
+  for (const docBytes of docs) {
+    const result = await extractBytes(docBytes, "application/pdf", {});
+    results.push(result);
+  }
+
+  return results;
+}
+
+await processMultipleDocs();
+```
+
+Stateful processors can track metrics across multiple extractions or maintain context.