@blockrun/clawrouter 0.6.1 → 0.6.3

package/dist/index.js

@@ -2320,6 +2320,34 @@ var DEFAULT_REQUEST_TIMEOUT_MS = 18e4;
 var DEFAULT_PORT = 8402;
 var MAX_FALLBACK_ATTEMPTS = 3;
 var HEALTH_CHECK_TIMEOUT_MS = 2e3;
+var RATE_LIMIT_COOLDOWN_MS = 6e4;
+var rateLimitedModels = /* @__PURE__ */ new Map();
+function isRateLimited(modelId) {
+  const hitTime = rateLimitedModels.get(modelId);
+  if (!hitTime) return false;
+  const elapsed = Date.now() - hitTime;
+  if (elapsed >= RATE_LIMIT_COOLDOWN_MS) {
+    rateLimitedModels.delete(modelId);
+    return false;
+  }
+  return true;
+}
+function markRateLimited(modelId) {
+  rateLimitedModels.set(modelId, Date.now());
+  console.log(`[ClawRouter] Model ${modelId} rate-limited, will deprioritize for 60s`);
+}
+function prioritizeNonRateLimited(models) {
+  const available = [];
+  const rateLimited = [];
+  for (const model of models) {
+    if (isRateLimited(model)) {
+      rateLimited.push(model);
+    } else {
+      available.push(model);
+    }
+  }
+  return [...available, ...rateLimited];
+}
 var BALANCE_CHECK_BUFFER = 1.5;
 function getProxyPort() {
   const envPort = process.env.BLOCKRUN_PROXY_PORT;
@@ -2909,6 +2937,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         );
       }
       modelsToTry = contextFiltered.slice(0, MAX_FALLBACK_ATTEMPTS);
+      modelsToTry = prioritizeNonRateLimited(modelsToTry);
     } else {
       modelsToTry = modelId ? [modelId] : [];
     }
@@ -2941,6 +2970,9 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         status: result.errorStatus || 500
       };
       if (result.isProviderError && !isLastAttempt) {
+        if (result.errorStatus === 429) {
+          markRateLimited(tryModel);
+        }
         console.log(
           `[ClawRouter] Provider error from ${tryModel}, trying fallback: ${result.errorBody?.slice(0, 100)}`
         );