@blockrun/clawrouter 0.9.36 → 0.9.38

package/dist/cli.js

@@ -1201,8 +1201,6 @@ var DEFAULT_ROUTING_CONFIG = {
       primary: "moonshot/kimi-k2.5",
       // $0.50/$2.40 - best quality/price for simple tasks
       fallback: [
-        "minimax/minimax-m2.5",
-        // $0.30/$1.20 - cheap with reasoning
         "google/gemini-2.5-flash",
         // 1M context, cost-effective
         "nvidia/gpt-oss-120b",
@@ -1214,8 +1212,6 @@ var DEFAULT_ROUTING_CONFIG = {
       primary: "xai/grok-code-fast-1",
       // Code specialist, $0.20/$1.50
       fallback: [
-        "minimax/minimax-m2.5",
-        // $0.30/$1.20 - cheap with reasoning
         "google/gemini-2.5-flash",
         // 1M context, cost-effective
         "deepseek/deepseek-chat",
@@ -1230,10 +1226,7 @@ var DEFAULT_ROUTING_CONFIG = {
         "google/gemini-2.5-flash",
         // CRITICAL: 1M context, cheap failsafe before expensive models
         "google/gemini-2.5-pro",
-        "minimax/minimax-m2.5",
-        // $0.30/$1.20 - cheap with reasoning
         "deepseek/deepseek-chat",
-        // Another cheap option
         "xai/grok-4-0709",
         "openai/gpt-5.2",
         // Newer and cheaper input than gpt-4o
@@ -1245,8 +1238,6 @@ var DEFAULT_ROUTING_CONFIG = {
       primary: "xai/grok-4-1-fast-reasoning",
       // Upgraded Grok 4.1 reasoning $0.20/$0.50
       fallback: [
-        "minimax/minimax-m2.5",
-        // $0.30/$1.20 - reasoning capable
         "deepseek/deepseek-reasoner",
         // Cheap reasoning model
         "openai/o4-mini",
@@ -1260,22 +1251,22 @@ var DEFAULT_ROUTING_CONFIG = {
     SIMPLE: {
       primary: "nvidia/gpt-oss-120b",
       // FREE! $0.00/$0.00
-      fallback: ["google/gemini-2.5-flash", "deepseek/deepseek-chat", "minimax/minimax-m2.5"]
+      fallback: ["google/gemini-2.5-flash", "deepseek/deepseek-chat"]
     },
     MEDIUM: {
       primary: "google/gemini-2.5-flash",
       // $0.15/$0.60 - cheapest capable
-      fallback: ["deepseek/deepseek-chat", "nvidia/gpt-oss-120b", "minimax/minimax-m2.5"]
+      fallback: ["deepseek/deepseek-chat", "nvidia/gpt-oss-120b"]
     },
     COMPLEX: {
       primary: "google/gemini-2.5-flash",
       // $0.15/$0.60 - 1M context handles complexity
-      fallback: ["deepseek/deepseek-chat", "xai/grok-4-0709", "minimax/minimax-m2.5"]
+      fallback: ["deepseek/deepseek-chat", "xai/grok-4-0709"]
     },
     REASONING: {
       primary: "xai/grok-4-1-fast-reasoning",
-      // $0.20/$0.50 - was MORE expensive than AUTO!
-      fallback: ["deepseek/deepseek-reasoner", "minimax/minimax-m2.5"]
+      // $0.20/$0.50
+      fallback: ["deepseek/deepseek-reasoner"]
     }
   },
   // Premium tier configs - best quality (blockrun/premium)
@@ -1326,8 +1317,6 @@ var DEFAULT_ROUTING_CONFIG = {
       primary: "moonshot/kimi-k2.5",
       // Cheaper than Haiku ($0.5/$2.4 vs $1/$5), larger context
       fallback: [
-        "minimax/minimax-m2.5",
-        // $0.30/$1.20 - agentic capable, cheaper than kimi
         "claude-haiku-4.5",
         "xai/grok-4-1-fast-non-reasoning",
         "openai/gpt-4o-mini"
@@ -1337,8 +1326,6 @@ var DEFAULT_ROUTING_CONFIG = {
       primary: "xai/grok-code-fast-1",
       // Code specialist for agentic coding
       fallback: [
-        "minimax/minimax-m2.5",
-        // $0.30/$1.20 - agentic capable
         "moonshot/kimi-k2.5",
         "claude-haiku-4.5",
         "claude-sonnet-4"
@@ -1349,8 +1336,6 @@ var DEFAULT_ROUTING_CONFIG = {
       fallback: [
         "claude-opus-4",
         // Latest Opus - best agentic
-        "minimax/minimax-m2.5",
-        // $0.30/$1.20 - cheap agentic fallback
         "openai/gpt-5.2",
         "google/gemini-3-pro-preview",
         "xai/grok-4-0709"
@@ -1361,8 +1346,6 @@ var DEFAULT_ROUTING_CONFIG = {
       // Strong tool use + reasoning for agentic tasks
       fallback: [
         "claude-opus-4",
-        "minimax/minimax-m2.5",
-        // $0.30/$1.20 - reasoning + agentic
         "xai/grok-4-1-fast-reasoning",
         "deepseek/deepseek-reasoner"
       ]
@@ -3621,6 +3604,7 @@ var ROUTING_PROFILES = /* @__PURE__ */ new Set([
 ]);
 var FREE_MODEL = "nvidia/gpt-oss-120b";
 var MAX_MESSAGES = 200;
+var CONTEXT_LIMIT_KB = 5120;
 var HEARTBEAT_INTERVAL_MS = 2e3;
 var DEFAULT_REQUEST_TIMEOUT_MS = 18e4;
 var MAX_FALLBACK_ATTEMPTS = 5;
@@ -4007,15 +3991,28 @@ function normalizeMessagesForThinking(messages) {
   return hasChanges ? normalized : messages;
 }
 function truncateMessages(messages) {
-  if (!messages || messages.length <= MAX_MESSAGES) return messages;
+  if (!messages || messages.length <= MAX_MESSAGES) {
+    return {
+      messages,
+      wasTruncated: false,
+      originalCount: messages?.length ?? 0,
+      truncatedCount: messages?.length ?? 0
+    };
+  }
   const systemMsgs = messages.filter((m) => m.role === "system");
   const conversationMsgs = messages.filter((m) => m.role !== "system");
   const maxConversation = MAX_MESSAGES - systemMsgs.length;
   const truncatedConversation = conversationMsgs.slice(-maxConversation);
+  const result = [...systemMsgs, ...truncatedConversation];
   console.log(
-    `[ClawRouter] Truncated messages: ${messages.length} \u2192 ${systemMsgs.length + truncatedConversation.length} (kept ${systemMsgs.length} system + ${truncatedConversation.length} recent)`
+    `[ClawRouter] Truncated messages: ${messages.length} \u2192 ${result.length} (kept ${systemMsgs.length} system + ${truncatedConversation.length} recent)`
   );
-  return [...systemMsgs, ...truncatedConversation];
+  return {
+    messages: result,
+    wasTruncated: true,
+    originalCount: messages.length,
+    truncatedCount: result.length
+  };
 }
 var KIMI_BLOCK_RE = /<[｜|][^<>]*begin[^<>]*[｜|]>[\s\S]*?<[｜|][^<>]*end[^<>]*[｜|]>/gi;
 var KIMI_TOKEN_RE = /<[｜|][^<>]*[｜|]>/g;
@@ -4348,7 +4345,8 @@ async function tryModelRequest(upstreamUrl, method, headers, body, modelId, maxT
       parsed.messages = normalizeMessageRoles(parsed.messages);
     }
     if (Array.isArray(parsed.messages)) {
-      parsed.messages = truncateMessages(parsed.messages);
+      const truncationResult = truncateMessages(parsed.messages);
+      parsed.messages = truncationResult.messages;
     }
     if (Array.isArray(parsed.messages)) {
       parsed.messages = sanitizeToolIds(parsed.messages);
@@ -4422,6 +4420,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
     bodyChunks.push(Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk));
   }
   let body = Buffer.concat(bodyChunks);
+  const originalContextSizeKB = Math.ceil(body.length / 1024);
   let routingDecision;
   let isStreaming = false;
   let modelId = "";
@@ -4529,7 +4528,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
             const systemPrompt = typeof systemMsg?.content === "string" ? systemMsg.content : void 0;
             const tools = parsed.tools;
             const hasTools = Array.isArray(tools) && tools.length > 0;
-            if (hasTools) {
+            if (hasTools && tools) {
               console.log(
                 `[ClawRouter] Tools detected (${tools.length}), agentic mode via keywords`
               );
@@ -4674,7 +4673,9 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
     res.writeHead(200, {
       "content-type": "text/event-stream",
       "cache-control": "no-cache",
-      connection: "keep-alive"
+      connection: "keep-alive",
+      "x-context-used-kb": String(originalContextSizeKB),
+      "x-context-limit-kb": String(CONTEXT_LIMIT_KB)
     });
     headersSentEarly = true;
     safeWrite(res, ": heartbeat\n\n");
@@ -4837,7 +4838,11 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
           completedAt: Date.now()
         });
       } else {
-        res.writeHead(errStatus, { "Content-Type": "application/json" });
+        res.writeHead(errStatus, {
+          "Content-Type": "application/json",
+          "x-context-used-kb": String(originalContextSizeKB),
+          "x-context-limit-kb": String(CONTEXT_LIMIT_KB)
+        });
         res.end(transformedErr);
         deduplicator.complete(dedupKey, {
           status: errStatus,
@@ -4963,6 +4968,8 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
           return;
         responseHeaders[key] = value;
       });
+      responseHeaders["x-context-used-kb"] = String(originalContextSizeKB);
+      responseHeaders["x-context-limit-kb"] = String(CONTEXT_LIMIT_KB);
       res.writeHead(upstream.status, responseHeaders);
       if (upstream.body) {
         const reader = upstream.body.getReader();