@prom.codes/memory-mcp 0.10.0 → 0.10.1

package/dist/bin.js

@@ -876,6 +876,7 @@ var DEFAULT_BASE2 = "https://api.prom.codes";
 var DEFAULT_NAME = "prometheus";
 var DEFAULT_MODEL2 = "prometheus";
 var DEFAULT_BATCH4 = 100;
+var DEFAULT_BATCH_CHARS2 = 4e5;
 var DEFAULT_RETRIES4 = 4;
 var DEFAULT_BACKOFF4 = 250;
 function sleep4(ms, signal) {
@@ -907,6 +908,7 @@ var PrometheusRerankProvider = class {
   #apiKey;
   #url;
   #batchSize;
+  #maxBatchChars;
   #maxRetries;
   #retryBaseMs;
   #fetch;
@@ -924,6 +926,7 @@ var PrometheusRerankProvider = class {
     this.#apiKey = opts.apiKey;
     this.#url = `${(opts.baseUrl ?? DEFAULT_BASE2).replace(/\/+$/, "")}/rerank`;
     this.#batchSize = opts.batchSize ?? DEFAULT_BATCH4;
+    this.#maxBatchChars = opts.maxBatchChars ?? DEFAULT_BATCH_CHARS2;
     this.#maxRetries = opts.maxRetries ?? DEFAULT_RETRIES4;
     this.#retryBaseMs = opts.retryBaseMs ?? DEFAULT_BACKOFF4;
     this.#fetch = opts.fetch ?? fetch;
@@ -937,21 +940,44 @@ var PrometheusRerankProvider = class {
       return [];
     const all = new Array(candidates.length);
     let cursor = 0;
-    for (let start = 0; start < candidates.length; start += this.#batchSize) {
-      const slice = candidates.slice(start, start + this.#batchSize);
+    let start = 0;
+    while (start < candidates.length) {
+      const end = this.#batchEnd(query, candidates, start);
+      const slice = candidates.slice(start, end);
       const scored = await this.#rerankBatch(query, slice, opts?.signal);
       for (const hit of scored) {
         const globalIndex = start + hit.localIndex;
         const cand = candidates[globalIndex];
         all[cursor++] = { id: cand.id, index: globalIndex, score: hit.score };
       }
+      start = end;
     }
+    all.length = cursor;
     all.sort((a, b) => b.score - a.score);
     if (opts?.topK !== void 0 && opts.topK >= 0 && opts.topK < all.length) {
       return all.slice(0, opts.topK);
     }
     return all;
   }
+  /**
+   * Greedy batch cut honouring both the document-count cap and the
+   * char budget (query + documents must stay under the proxy limit).
+   * Always advances by at least one candidate, even if a single
+   * document alone exceeds the budget (the proxy applies upstream
+   * truncation; refusing to send it would strand the rest).
+   */
+  #batchEnd(query, candidates, start) {
+    let chars = query.length;
+    let end = start;
+    while (end < candidates.length && end - start < this.#batchSize) {
+      const len = candidates[end].text.length;
+      if (end > start && chars + len > this.#maxBatchChars)
+        break;
+      chars += len;
+      end += 1;
+    }
+    return end;
+  }
   async #rerankBatch(query, batch, signal) {
     const init = {
       method: "POST",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@prom.codes/memory-mcp",
-  "version": "0.10.0",
+  "version": "0.10.1",
   "description": "prom.codes Memory — persistent, local-first agent memory as an MCP server.",
   "type": "module",
   "bin": {