@melaya/runner 1.0.50 → 1.0.52

package/dist/connection.js

@@ -580,30 +580,49 @@ export async function connect(opts) {
                 "--embedder-model", payload.embedderModel,
             ];
             if (ingestMode === "cloud_files") {
-                // Server bundled file bytes into the event. Unpack them to a
-                // temp dir + a small manifest JSON so the Python script can
-                // walk them with the same code path as Mode B reads from disk.
+                // Pull each file from the server via an ack-based `rag:fetch-file`
+                // round-trip. One file per Socket.IO frame keeps every message
+                // small regardless of total corpus size. The server reads from
+                // its loopback builder API and returns base64 bytes.
                 const filesDir = join(workDir, "cloud-files");
                 mkdirSync(filesDir, { recursive: true });
                 const { writeFileSync } = await import("fs");
-                const filesArr = payload.cloudFiles || [];
-                for (const f of filesArr) {
+                const manifestFiles = payload.cloudManifest?.files || [];
+                const fetched = [];
+                let failed = 0;
+                for (const f of manifestFiles) {
+                    const resp = await new Promise((resolve) => {
+                        const timer = setTimeout(() => resolve({ ok: false, error: "ack timeout (60s)" }), 60_000);
+                        socket.emit("rag:fetch-file", { session_id: sid, filename: f.name }, (r) => {
+                            clearTimeout(timer);
+                            resolve(r || { ok: false, error: "empty ack" });
+                        });
+                    });
+                    if (!resp.ok || !resp.contentB64) {
+                        failed += 1;
+                        console.log(chalk.red(`  [rag] fetch failed ${f.name}: ${resp.error || "unknown"}`));
+                        emitProgress("file_failed", { file: f.name, error: resp.error || "fetch failed" });
+                        continue;
+                    }
                     try {
-                        writeFileSync(join(filesDir, f.name), Buffer.from(f.contentB64, "base64"));
+                        writeFileSync(join(filesDir, f.name), Buffer.from(resp.contentB64, "base64"));
+                        fetched.push({ name: f.name, sha256: f.sha256, ext: f.ext });
                     }
                     catch (e) {
-                        console.log(chalk.red(`  [rag] failed to unpack ${f.name}: ${e?.message || e}`));
+                        failed += 1;
+                        console.log(chalk.red(`  [rag] write failed ${f.name}: ${e?.message || e}`));
+                        emitProgress("file_failed", { file: f.name, error: String(e?.message || e) });
                     }
                 }
                 const manifestPath = join(workDir, "cloud-manifest.json");
                 writeFileSync(manifestPath, JSON.stringify({
-                    files: filesArr.map(f => ({ name: f.name, sha256: f.sha256, ext: f.ext })),
+                    files: fetched,
                     stored_embedder: payload.cloudManifest?.stored_embedder || {},
                     previously_ingested: payload.cloudManifest?.previously_ingested || {},
                 }, null, 2));
                 ingestArgs.push("--files-dir", filesDir);
                 ingestArgs.push("--manifest-json", manifestPath);
-                console.log(chalk.gray(`  [rag] unpacked ${filesArr.length} cloud file(s) → ${filesDir}`));
+                console.log(chalk.gray(`  [rag] pulled ${fetched.length}/${manifestFiles.length} cloud file(s)${failed ? ` (${failed} failed)` : ""} → ${filesDir}`));
             }
             else {
                 ingestArgs.push("--folder", payload.localFolderPath || "");

package/dist/localRagIngest.py

@@ -484,7 +484,10 @@ async def _ingest_cloud_files(args) -> int:
                 first_emit = False
                 continue
             vectors_resp = await embedder(text=texts)
-            vectors = [eb.embedding for eb in vectors_resp.embeddings]
+            # agentscope's embedding adapters already unwrap to a flat list
+            # of vectors (list[list[float]]) via _openai_embedding.py L104 /
+            # _ollama_embedding.py — do NOT re-access `.embedding`.
+            vectors = list(vectors_resp.embeddings)
             if len(vectors) != len(texts):
                 raise RuntimeError(
                     f"embedder returned {len(vectors)} vectors for {len(texts)} chunks"

package/localRagIngest.py

@@ -484,7 +484,10 @@ async def _ingest_cloud_files(args) -> int:
                 first_emit = False
                 continue
             vectors_resp = await embedder(text=texts)
-            vectors = [eb.embedding for eb in vectors_resp.embeddings]
+            # agentscope's embedding adapters already unwrap to a flat list
+            # of vectors (list[list[float]]) via _openai_embedding.py L104 /
+            # _ollama_embedding.py — do NOT re-access `.embedding`.
+            vectors = list(vectors_resp.embeddings)
             if len(vectors) != len(texts):
                 raise RuntimeError(
                     f"embedder returned {len(vectors)} vectors for {len(texts)} chunks"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@melaya/runner",
-  "version": "1.0.50",
+  "version": "1.0.52",
   "description": "Run Melaya AI pipelines locally with your own LM Studio or Ollama models",
   "license": "UNLICENSED",
   "private": false,