⚡ Bolt: Optimize ZimaOS and Ollama model fetching

.jules/bolt.md

@@ -0,0 +1,3 @@
+## 2026-05-06 - Parallelizing fallback sequences trades network for latency
+**Learning:** Converting sequential `for...of` retry/fallback loops into parallel `Promise.all` queries executes all options concurrently. While this achieves the lowest possible latency for the function, it generates redundant HTTP requests to secondary pathways if the primary succeeds.
+**Action:** For internal or gateway endpoints where minimal latency is favored over network overhead (like ZimaOS local probes), `Promise.all` is beneficial. Ensure to iterate correctly through the results array in the original order to preserve state variables dependent on sequence.

src/lib/forge-openai-surface.ts

@@ -118,8 +118,8 @@ export async function collectZimaOSV1ModelEntries(
   let lastError: string | undefined;
   let lastRaw: unknown = null;
 
-  for (const path of V1_MODELS_PATHS) {
-    const r = await fetchZimaOSJson(email, path);
+  const results = await Promise.all(V1_MODELS_PATHS.map(path => fetchZimaOSJson(email, path)));
+  for (const r of results) {
     lastStatus = r.status;
     lastError = r.error;
     lastRaw = r.data;