@blockrun/clawrouter 0.8.14 → 0.8.15

package/README.md

@@ -60,6 +60,7 @@ Done! Smart routing (`blockrun/auto`) is now your default model.
 **📖 Full Windows Guide:** [docs/windows-installation.md](docs/windows-installation.md)
 
 **Quick Summary:**
+
 - ✅ ClawRouter code is Windows-compatible
 - ❌ OpenClaw CLI has a `spawn EINVAL` bug on Windows
 - ✅ Works perfectly on **Linux** and **macOS**

package/dist/cli.js

@@ -1150,7 +1150,11 @@ var DEFAULT_ROUTING_CONFIG = {
     SIMPLE: {
       primary: "moonshot/kimi-k2.5",
       // Cheaper than Haiku ($0.5/$2.4 vs $1/$5), larger context
-      fallback: ["anthropic/claude-haiku-4.5", "xai/grok-4-fast-non-reasoning", "openai/gpt-4o-mini"]
+      fallback: [
+        "anthropic/claude-haiku-4.5",
+        "xai/grok-4-fast-non-reasoning",
+        "openai/gpt-4o-mini"
+      ]
     },
     MEDIUM: {
       primary: "xai/grok-code-fast-1",
@@ -1662,6 +1666,11 @@ function getModelContextWindow(modelId) {
   const model = BLOCKRUN_MODELS.find((m) => m.id === normalized);
   return model?.contextWindow;
 }
+function isReasoningModel(modelId) {
+  const normalized = modelId.replace("blockrun/", "");
+  const model = BLOCKRUN_MODELS.find((m) => m.id === normalized);
+  return model?.reasoning ?? false;
+}
 
 // src/logger.ts
 import { appendFile, mkdir } from "fs/promises";
@@ -2815,7 +2824,8 @@ async function tryModelRequest(upstreamUrl, method, headers, body, modelId, maxT
     if (isGoogleModel(modelId) && Array.isArray(parsed.messages)) {
       parsed.messages = normalizeMessagesForGoogle(parsed.messages);
     }
-    if (parsed.thinking && Array.isArray(parsed.messages)) {
+    const hasThinkingEnabled = !!(parsed.thinking || parsed.extended_thinking || isReasoningModel(modelId));
+    if (hasThinkingEnabled && Array.isArray(parsed.messages)) {
       parsed.messages = normalizeMessagesForThinking(parsed.messages);
     }
     requestBody = Buffer.from(JSON.stringify(parsed));
@@ -3057,7 +3067,11 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
       modelsToTry = contextFiltered.slice(0, MAX_FALLBACK_ATTEMPTS);
       modelsToTry = prioritizeNonRateLimited(modelsToTry);
     } else {
-      modelsToTry = modelId ? [modelId] : [];
+      if (modelId && modelId !== FREE_MODEL) {
+        modelsToTry = [modelId, FREE_MODEL];
+      } else {
+        modelsToTry = modelId ? [modelId] : [];
+      }
     }
     let upstream;
     let lastError;
@@ -3136,7 +3150,9 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
           const parsed = JSON.parse(transformedErr);
           errPayload = JSON.stringify(parsed);
         } catch {
-          errPayload = JSON.stringify({ error: { message: rawErrBody, type: "provider_error", status: errStatus } });
+          errPayload = JSON.stringify({
+            error: { message: rawErrBody, type: "provider_error", status: errStatus }
+          });
         }
         const errEvent = `data: ${errPayload}
 
@@ -3240,7 +3256,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
                     index,
                     delta: {},
                     logprobs: null,
-                    finish_reason: choice.finish_reason ?? "stop"
+                    finish_reason: toolCalls && toolCalls.length > 0 ? "tool_calls" : choice.finish_reason ?? "stop"
                   }
                 ]
               };