membot 0.4.0 → 0.4.1

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "membot",
-	"version": "0.4.0",
+	"version": "0.4.1",
 	"description": "Versioned context store with hybrid search for AI agents. Stdio + HTTP MCP server and CLI.",
 	"type": "module",
 	"exports": {

package/src/constants.ts

@@ -20,6 +20,15 @@ export const ENV = {
 export const EMBEDDING_MODEL = "Xenova/bge-small-en-v1.5";
 export const EMBEDDING_DIMENSION = 384;
 
+/**
+ * Max chunks fed to the feature-extraction pipeline in one forward pass.
+ * ONNX/WASM allocates activations linearly with batch size, so a single
+ * unbounded call OOMs (`std::bad_alloc`) on large files — a 168-chunk file
+ * was the original repro. 16 is comfortably within the WASM heap for
+ * bge-small-en-v1.5 at 512 tokens and still amortizes the per-call overhead.
+ */
+export const EMBEDDING_BATCH_SIZE = 16;
+
 export const DEFAULTS = {
 	CHUNKER_MODE: "deterministic" as const,
 	CHUNKER_TARGET_CHARS: 4_000,

package/src/ingest/embedder.ts

@@ -1,7 +1,7 @@
 import { existsSync } from "node:fs";
 import { join } from "node:path";
 import { env, type FeatureExtractionPipeline, pipeline } from "@huggingface/transformers";
-import { EMBEDDING_DIMENSION, EMBEDDING_MODEL } from "../constants.ts";
+import { EMBEDDING_BATCH_SIZE, EMBEDDING_DIMENSION, EMBEDDING_MODEL } from "../constants.ts";
 import { HelpfulError } from "../errors.ts";
 import { logger } from "../output/logger.ts";
 
@@ -67,20 +67,29 @@ async function getPipeline(model: string): Promise<FeatureExtractionPipeline> {
  * Embed an array of texts to L2-normalized vectors with the configured
  * model. Throws a HelpfulError when the model's dimension doesn't match
  * EMBEDDING_DIMENSION (the value baked into the DB schema).
+ *
+ * Inputs are sliced into windows of EMBEDDING_BATCH_SIZE so a single
+ * forward pass never has to allocate activations for arbitrarily many
+ * chunks — large files (hundreds of chunks) otherwise OOM the WASM heap.
  */
 export async function embed(texts: string[], model: string = EMBEDDING_MODEL): Promise<number[][]> {
 	if (texts.length === 0) return [];
 	const extractor = await getPipeline(model);
-	const output = await extractor(texts, { pooling: "mean", normalize: true });
-	const data = output.tolist() as number[][];
-	if (data[0] && data[0].length !== EMBEDDING_DIMENSION) {
-		throw new HelpfulError({
-			kind: "internal_error",
-			message: `embedding model ${model} returned ${data[0].length}-dim vectors, expected ${EMBEDDING_DIMENSION}`,
-			hint: `Set config.embedding_model to a ${EMBEDDING_DIMENSION}-dim model (default: ${EMBEDDING_MODEL}).`,
-		});
+	const out: number[][] = [];
+	for (let i = 0; i < texts.length; i += EMBEDDING_BATCH_SIZE) {
+		const slice = texts.slice(i, i + EMBEDDING_BATCH_SIZE);
+		const output = await extractor(slice, { pooling: "mean", normalize: true });
+		const data = output.tolist() as number[][];
+		if (out.length === 0 && data[0] && data[0].length !== EMBEDDING_DIMENSION) {
+			throw new HelpfulError({
+				kind: "internal_error",
+				message: `embedding model ${model} returned ${data[0].length}-dim vectors, expected ${EMBEDDING_DIMENSION}`,
+				hint: `Set config.embedding_model to a ${EMBEDDING_DIMENSION}-dim model (default: ${EMBEDDING_MODEL}).`,
+			});
+		}
+		for (const vec of data) out.push(vec);
 	}
-	return data;
+	return out;
 }
 
 /** Embed a single text — convenience wrapper for query-time embedding. */