@blockrun/clawrouter 0.11.13 → 0.11.14

package/dist/cli.js

@@ -11,13 +11,13 @@ import { x402Client } from "@x402/fetch";
 // src/payment-preauth.ts
 import { x402HTTPClient } from "@x402/fetch";
 var DEFAULT_TTL_MS = 36e5;
-function createPayFetchWithPreAuth(baseFetch, client, ttlMs = DEFAULT_TTL_MS) {
+function createPayFetchWithPreAuth(baseFetch, client, ttlMs = DEFAULT_TTL_MS, options) {
   const httpClient = new x402HTTPClient(client);
   const cache = /* @__PURE__ */ new Map();
   return async (input, init) => {
     const request = new Request(input, init);
     const urlPath = new URL(request.url).pathname;
-    const cached = cache.get(urlPath);
+    const cached = !options?.skipPreAuth ? cache.get(urlPath) : void 0;
     if (cached && Date.now() - cached.cachedAt < ttlMs) {
       try {
         const payload2 = await client.createPaymentPayload(cached.paymentRequired);
@@ -157,20 +157,18 @@ function scoreAgenticTask(text, keywords) {
   };
 }
 function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
-  const text = `${systemPrompt ?? ""} ${prompt}`.toLowerCase();
   const userText = prompt.toLowerCase();
   const dimensions = [
-    // Original 8 dimensions
+    // Token count uses total estimated tokens (system + user) — context size matters for model selection
     scoreTokenCount(estimatedTokens, config.tokenCountThresholds),
     scoreKeywordMatch(
-      text,
+      userText,
       config.codeKeywords,
       "codePresence",
       "code",
       { low: 1, high: 2 },
       { none: 0, low: 0.5, high: 1 }
     ),
-    // Reasoning markers use USER prompt only — system prompt "step by step" shouldn't trigger reasoning
     scoreKeywordMatch(
       userText,
       config.reasoningKeywords,
@@ -180,7 +178,7 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       { none: 0, low: 0.7, high: 1 }
     ),
     scoreKeywordMatch(
-      text,
+      userText,
       config.technicalKeywords,
       "technicalTerms",
       "technical",
@@ -188,7 +186,7 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       { none: 0, low: 0.5, high: 1 }
     ),
     scoreKeywordMatch(
-      text,
+      userText,
       config.creativeKeywords,
       "creativeMarkers",
       "creative",
@@ -196,18 +194,18 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       { none: 0, low: 0.5, high: 0.7 }
     ),
     scoreKeywordMatch(
-      text,
+      userText,
       config.simpleKeywords,
       "simpleIndicators",
       "simple",
       { low: 1, high: 2 },
       { none: 0, low: -1, high: -1 }
     ),
-    scoreMultiStep(text),
+    scoreMultiStep(userText),
     scoreQuestionComplexity(prompt),
     // 6 new dimensions
     scoreKeywordMatch(
-      text,
+      userText,
       config.imperativeVerbs,
       "imperativeVerbs",
       "imperative",
@@ -215,7 +213,7 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       { none: 0, low: 0.3, high: 0.5 }
     ),
     scoreKeywordMatch(
-      text,
+      userText,
       config.constraintIndicators,
       "constraintCount",
       "constraints",
@@ -223,7 +221,7 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       { none: 0, low: 0.3, high: 0.7 }
     ),
     scoreKeywordMatch(
-      text,
+      userText,
       config.outputFormatKeywords,
       "outputFormat",
       "format",
@@ -231,7 +229,7 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       { none: 0, low: 0.4, high: 0.7 }
     ),
     scoreKeywordMatch(
-      text,
+      userText,
       config.referenceKeywords,
       "referenceComplexity",
       "references",
@@ -239,7 +237,7 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       { none: 0, low: 0.3, high: 0.5 }
     ),
     scoreKeywordMatch(
-      text,
+      userText,
       config.negationKeywords,
       "negationComplexity",
       "negation",
@@ -247,7 +245,7 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       { none: 0, low: 0.3, high: 0.5 }
     ),
     scoreKeywordMatch(
-      text,
+      userText,
       config.domainSpecificKeywords,
       "domainSpecificity",
       "domain-specific",
@@ -279,7 +277,8 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
       tier: "REASONING",
       confidence: Math.max(confidence2, 0.85),
       signals,
-      agenticScore
+      agenticScore,
+      dimensions
     };
   }
   const { simpleMedium, mediumComplex, complexReasoning } = config.tierBoundaries;
@@ -303,9 +302,9 @@ function classifyByRules(prompt, systemPrompt, estimatedTokens, config) {
   }
   const confidence = calibrateConfidence(distanceFromBoundary, config.confidenceSteepness);
   if (confidence < config.confidenceThreshold) {
-    return { score: weightedScore, tier: null, confidence, signals, agenticScore };
+    return { score: weightedScore, tier: null, confidence, signals, agenticScore, dimensions };
   }
-  return { score: weightedScore, tier, confidence, signals, agenticScore };
+  return { score: weightedScore, tier, confidence, signals, agenticScore, dimensions };
 }
 function calibrateConfidence(distance, steepness) {
   return 1 / (1 + Math.exp(-steepness * distance));
@@ -313,7 +312,9 @@ function calibrateConfidence(distance, steepness) {
 
 // src/router/selector.ts
 var BASELINE_MODEL_ID = "anthropic/claude-opus-4.6";
-function selectModel(tier, confidence, method, reasoning, tierConfigs, modelPricing, estimatedInputTokens, maxOutputTokens, routingProfile) {
+var BASELINE_INPUT_PRICE = 5;
+var BASELINE_OUTPUT_PRICE = 25;
+function selectModel(tier, confidence, method, reasoning, tierConfigs, modelPricing, estimatedInputTokens, maxOutputTokens, routingProfile, agenticScore) {
   const tierConfig = tierConfigs[tier];
   const model = tierConfig.primary;
   const pricing = modelPricing.get(model);
@@ -323,8 +324,8 @@ function selectModel(tier, confidence, method, reasoning, tierConfigs, modelPric
   const outputCost = maxOutputTokens / 1e6 * outputPrice;
   const costEstimate = inputCost + outputCost;
   const opusPricing = modelPricing.get(BASELINE_MODEL_ID);
-  const opusInputPrice = opusPricing?.inputPrice ?? 0;
-  const opusOutputPrice = opusPricing?.outputPrice ?? 0;
+  const opusInputPrice = opusPricing?.inputPrice ?? BASELINE_INPUT_PRICE;
+  const opusOutputPrice = opusPricing?.outputPrice ?? BASELINE_OUTPUT_PRICE;
   const baselineInput = estimatedInputTokens / 1e6 * opusInputPrice;
   const baselineOutput = maxOutputTokens / 1e6 * opusOutputPrice;
   const baselineCost = baselineInput + baselineOutput;
@@ -337,7 +338,8 @@ function selectModel(tier, confidence, method, reasoning, tierConfigs, modelPric
     reasoning,
     costEstimate,
     baselineCost,
-    savings
+    savings,
+    ...agenticScore !== void 0 && { agenticScore }
   };
 }
 function getFallbackChain(tier, tierConfigs) {
@@ -352,14 +354,24 @@ function calculateModelCost(model, modelPricing, estimatedInputTokens, maxOutput
   const outputCost = maxOutputTokens / 1e6 * outputPrice;
   const costEstimate = inputCost + outputCost;
   const opusPricing = modelPricing.get(BASELINE_MODEL_ID);
-  const opusInputPrice = opusPricing?.inputPrice ?? 0;
-  const opusOutputPrice = opusPricing?.outputPrice ?? 0;
+  const opusInputPrice = opusPricing?.inputPrice ?? BASELINE_INPUT_PRICE;
+  const opusOutputPrice = opusPricing?.outputPrice ?? BASELINE_OUTPUT_PRICE;
   const baselineInput = estimatedInputTokens / 1e6 * opusInputPrice;
   const baselineOutput = maxOutputTokens / 1e6 * opusOutputPrice;
   const baselineCost = baselineInput + baselineOutput;
   const savings = routingProfile === "premium" ? 0 : baselineCost > 0 ? Math.max(0, (baselineCost - costEstimate) / baselineCost) : 0;
   return { costEstimate, baselineCost, savings };
 }
+function filterByToolCalling(models, hasTools, supportsToolCalling2) {
+  if (!hasTools) return models;
+  const filtered = models.filter(supportsToolCalling2);
+  return filtered.length > 0 ? filtered : models;
+}
+function filterByVision(models, hasVision, supportsVision2) {
+  if (!hasVision) return models;
+  const filtered = models.filter(supportsVision2);
+  return filtered.length > 0 ? filtered : models;
+}
 function getFallbackChainFiltered(tier, tierConfigs, estimatedTotalTokens, getContextWindow) {
   const fullChain = getFallbackChain(tier, tierConfigs);
   const filtered = fullChain.filter((modelId) => {
@@ -1415,18 +1427,18 @@ var DEFAULT_ROUTING_CONFIG = {
       ]
     },
     MEDIUM: {
-      primary: "xai/grok-code-fast-1",
-      // Code specialist, $0.20/$1.50
+      primary: "moonshot/kimi-k2.5",
+      // $0.50/$2.40 - strong tool use, proper function call format
       fallback: [
+        "deepseek/deepseek-chat",
         "google/gemini-2.5-flash-lite",
         // 1M context, ultra cheap ($0.10/$0.40)
-        "deepseek/deepseek-chat",
         "xai/grok-4-1-fast-non-reasoning"
         // Upgraded Grok 4.1
       ]
     },
     COMPLEX: {
-      primary: "google/gemini-3.1-pro-preview",
+      primary: "google/gemini-3.1-pro",
       // Newest Gemini 3.1 - upgraded from 3.0
       fallback: [
         "google/gemini-2.5-flash-lite",
@@ -1486,7 +1498,7 @@ var DEFAULT_ROUTING_CONFIG = {
       fallback: [
         "anthropic/claude-haiku-4.5",
         "google/gemini-2.5-flash-lite",
-        "xai/grok-code-fast-1"
+        "deepseek/deepseek-chat"
       ]
     },
     MEDIUM: {
@@ -1506,7 +1518,7 @@ var DEFAULT_ROUTING_CONFIG = {
         "openai/gpt-5.2-codex",
         "anthropic/claude-opus-4.6",
         "anthropic/claude-sonnet-4.6",
-        "google/gemini-3.1-pro-preview",
+        "google/gemini-3.1-pro",
         // Newest Gemini
         "google/gemini-3-pro-preview",
         "moonshot/kimi-k2.5"
@@ -1537,9 +1549,13 @@ var DEFAULT_ROUTING_CONFIG = {
       ]
     },
     MEDIUM: {
-      primary: "xai/grok-code-fast-1",
-      // Code specialist for agentic coding
-      fallback: ["moonshot/kimi-k2.5", "anthropic/claude-haiku-4.5", "claude-sonnet-4"]
+      primary: "moonshot/kimi-k2.5",
+      // $0.50/$2.40 - strong tool use, handles function calls correctly
+      fallback: [
+        "anthropic/claude-haiku-4.5",
+        "deepseek/deepseek-chat",
+        "xai/grok-4-1-fast-non-reasoning"
+      ]
     },
     COMPLEX: {
       primary: "anthropic/claude-sonnet-4.6",
@@ -1547,7 +1563,7 @@ var DEFAULT_ROUTING_CONFIG = {
         "anthropic/claude-opus-4.6",
         // Latest Opus - best agentic
         "openai/gpt-5.2",
-        "google/gemini-3.1-pro-preview",
+        "google/gemini-3.1-pro",
         // Newest Gemini
         "google/gemini-3-pro-preview",
         "xai/grok-4-0709"
@@ -1579,7 +1595,7 @@ function route(prompt, systemPrompt, maxOutputTokens, options) {
   const ruleResult = classifyByRules(prompt, systemPrompt, estimatedTokens, config.scoring);
   const { routingProfile } = options;
   let tierConfigs;
-  let profileSuffix = "";
+  let profileSuffix;
   if (routingProfile === "eco" && config.ecoTiers) {
     tierConfigs = config.ecoTiers;
     profileSuffix = " | eco";
@@ -1594,6 +1610,7 @@ function route(prompt, systemPrompt, maxOutputTokens, options) {
     tierConfigs = useAgenticTiers ? config.agenticTiers : config.tiers;
     profileSuffix = useAgenticTiers ? " | agentic" : "";
   }
+  const agenticScoreValue = ruleResult.agenticScore;
   if (estimatedTokens > config.overrides.maxTokensForceComplex) {
     return selectModel(
       "COMPLEX",
@@ -1604,7 +1621,8 @@ function route(prompt, systemPrompt, maxOutputTokens, options) {
       modelPricing,
       estimatedTokens,
       maxOutputTokens,
-      routingProfile
+      routingProfile,
+      agenticScoreValue
     );
   }
   const hasStructuredOutput = systemPrompt ? /json|structured|schema/i.test(systemPrompt) : false;
@@ -1638,7 +1656,8 @@ function route(prompt, systemPrompt, maxOutputTokens, options) {
     modelPricing,
     estimatedTokens,
     maxOutputTokens,
-    routingProfile
+    routingProfile,
+    agenticScoreValue
   );
 }
 
@@ -1686,6 +1705,8 @@ var MODEL_ALIASES = {
   // Google
   gemini: "google/gemini-2.5-pro",
   flash: "google/gemini-2.5-flash",
+  "gemini-3.1-pro-preview": "google/gemini-3.1-pro",
+  "google/gemini-3.1-pro-preview": "google/gemini-3.1-pro",
   // xAI
   grok: "xai/grok-3",
   "grok-fast": "xai/grok-4-fast-reasoning",
@@ -1759,7 +1780,8 @@ var BLOCKRUN_MODELS = [
     maxOutput: 128e3,
     reasoning: true,
     vision: true,
-    agentic: true
+    agentic: true,
+    toolCalling: true
   },
   {
     id: "openai/gpt-5-mini",
@@ -1768,7 +1790,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.25,
     outputPrice: 2,
     contextWindow: 2e5,
-    maxOutput: 65536
+    maxOutput: 65536,
+    toolCalling: true
   },
   {
     id: "openai/gpt-5-nano",
@@ -1777,7 +1800,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.05,
     outputPrice: 0.4,
     contextWindow: 128e3,
-    maxOutput: 32768
+    maxOutput: 32768,
+    toolCalling: true
   },
   {
     id: "openai/gpt-5.2-pro",
@@ -1787,7 +1811,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 168,
     contextWindow: 4e5,
     maxOutput: 128e3,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   // OpenAI Codex Family
   {
@@ -1798,7 +1823,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 14,
     contextWindow: 128e3,
     maxOutput: 32e3,
-    agentic: true
+    agentic: true,
+    toolCalling: true
   },
   // OpenAI GPT-4 Family
   {
@@ -1809,7 +1835,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 8,
     contextWindow: 128e3,
     maxOutput: 16384,
-    vision: true
+    vision: true,
+    toolCalling: true
   },
   {
     id: "openai/gpt-4.1-mini",
@@ -1818,7 +1845,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.4,
     outputPrice: 1.6,
     contextWindow: 128e3,
-    maxOutput: 16384
+    maxOutput: 16384,
+    toolCalling: true
   },
   {
     id: "openai/gpt-4.1-nano",
@@ -1827,7 +1855,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.1,
     outputPrice: 0.4,
     contextWindow: 128e3,
-    maxOutput: 16384
+    maxOutput: 16384,
+    toolCalling: true
   },
   {
     id: "openai/gpt-4o",
@@ -1838,7 +1867,8 @@ var BLOCKRUN_MODELS = [
     contextWindow: 128e3,
     maxOutput: 16384,
     vision: true,
-    agentic: true
+    agentic: true,
+    toolCalling: true
   },
   {
     id: "openai/gpt-4o-mini",
@@ -1847,7 +1877,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.15,
     outputPrice: 0.6,
     contextWindow: 128e3,
-    maxOutput: 16384
+    maxOutput: 16384,
+    toolCalling: true
   },
   // OpenAI O-series (Reasoning)
   {
@@ -1858,7 +1889,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 60,
     contextWindow: 2e5,
     maxOutput: 1e5,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   {
     id: "openai/o1-mini",
@@ -1868,7 +1900,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 4.4,
     contextWindow: 128e3,
     maxOutput: 65536,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   {
     id: "openai/o3",
@@ -1878,7 +1911,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 8,
     contextWindow: 2e5,
     maxOutput: 1e5,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   {
     id: "openai/o3-mini",
@@ -1888,7 +1922,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 4.4,
     contextWindow: 128e3,
     maxOutput: 65536,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   {
     id: "openai/o4-mini",
@@ -1898,7 +1933,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 4.4,
     contextWindow: 128e3,
     maxOutput: 65536,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   // Anthropic - all Claude models excel at agentic workflows
   // Use newest versions (4.6) with full provider prefix
@@ -1910,7 +1946,9 @@ var BLOCKRUN_MODELS = [
     outputPrice: 5,
     contextWindow: 2e5,
     maxOutput: 8192,
-    agentic: true
+    vision: true,
+    agentic: true,
+    toolCalling: true
   },
   {
     id: "anthropic/claude-sonnet-4.6",
@@ -1921,7 +1959,9 @@ var BLOCKRUN_MODELS = [
     contextWindow: 2e5,
     maxOutput: 64e3,
     reasoning: true,
-    agentic: true
+    vision: true,
+    agentic: true,
+    toolCalling: true
   },
   {
     id: "anthropic/claude-opus-4.6",
@@ -1932,19 +1972,22 @@ var BLOCKRUN_MODELS = [
     contextWindow: 2e5,
     maxOutput: 32e3,
     reasoning: true,
-    agentic: true
+    vision: true,
+    agentic: true,
+    toolCalling: true
   },
   // Google
   {
-    id: "google/gemini-3.1-pro-preview",
-    name: "Gemini 3.1 Pro Preview",
+    id: "google/gemini-3.1-pro",
+    name: "Gemini 3.1 Pro",
     version: "3.1",
     inputPrice: 2,
     outputPrice: 12,
     contextWindow: 105e4,
     maxOutput: 65536,
     reasoning: true,
-    vision: true
+    vision: true,
+    toolCalling: true
   },
   {
     id: "google/gemini-3-pro-preview",
@@ -1955,7 +1998,8 @@ var BLOCKRUN_MODELS = [
     contextWindow: 105e4,
     maxOutput: 65536,
     reasoning: true,
-    vision: true
+    vision: true,
+    toolCalling: true
   },
   {
     id: "google/gemini-3-flash-preview",
@@ -1965,7 +2009,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 3,
     contextWindow: 1e6,
     maxOutput: 65536,
-    vision: true
+    vision: true,
+    toolCalling: true
   },
   {
     id: "google/gemini-2.5-pro",
@@ -1976,7 +2021,8 @@ var BLOCKRUN_MODELS = [
     contextWindow: 105e4,
     maxOutput: 65536,
     reasoning: true,
-    vision: true
+    vision: true,
+    toolCalling: true
   },
   {
     id: "google/gemini-2.5-flash",
@@ -1985,7 +2031,9 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.3,
     outputPrice: 2.5,
     contextWindow: 1e6,
-    maxOutput: 65536
+    maxOutput: 65536,
+    vision: true,
+    toolCalling: true
   },
   {
     id: "google/gemini-2.5-flash-lite",
@@ -1994,7 +2042,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.1,
     outputPrice: 0.4,
     contextWindow: 1e6,
-    maxOutput: 65536
+    maxOutput: 65536,
+    toolCalling: true
   },
   // DeepSeek
   {
@@ -2004,7 +2053,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.28,
     outputPrice: 0.42,
     contextWindow: 128e3,
-    maxOutput: 8192
+    maxOutput: 8192,
+    toolCalling: true
   },
   {
     id: "deepseek/deepseek-reasoner",
@@ -2014,7 +2064,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 0.42,
     contextWindow: 128e3,
     maxOutput: 8192,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   // Moonshot / Kimi - optimized for agentic workflows
   {
@@ -2027,7 +2078,8 @@ var BLOCKRUN_MODELS = [
     maxOutput: 8192,
     reasoning: true,
     vision: true,
-    agentic: true
+    agentic: true,
+    toolCalling: true
   },
   // xAI / Grok
   {
@@ -2038,7 +2090,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 15,
     contextWindow: 131072,
     maxOutput: 16384,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   // grok-3-fast removed - too expensive ($5/$25), use grok-4-fast instead
   {
@@ -2048,7 +2101,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.3,
     outputPrice: 0.5,
     contextWindow: 131072,
-    maxOutput: 16384
+    maxOutput: 16384,
+    toolCalling: true
   },
   // xAI Grok 4 Family - Ultra-cheap fast models
   {
@@ -2059,7 +2113,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 0.5,
     contextWindow: 131072,
     maxOutput: 16384,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   {
     id: "xai/grok-4-fast-non-reasoning",
@@ -2068,7 +2123,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.2,
     outputPrice: 0.5,
     contextWindow: 131072,
-    maxOutput: 16384
+    maxOutput: 16384,
+    toolCalling: true
   },
   {
     id: "xai/grok-4-1-fast-reasoning",
@@ -2078,7 +2134,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 0.5,
     contextWindow: 131072,
     maxOutput: 16384,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   {
     id: "xai/grok-4-1-fast-non-reasoning",
@@ -2087,7 +2144,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.2,
     outputPrice: 0.5,
     contextWindow: 131072,
-    maxOutput: 16384
+    maxOutput: 16384,
+    toolCalling: true
   },
   {
     id: "xai/grok-code-fast-1",
@@ -2096,9 +2154,10 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.2,
     outputPrice: 1.5,
     contextWindow: 131072,
-    maxOutput: 16384,
-    agentic: true
-    // Good for coding tasks
+    maxOutput: 16384
+    // toolCalling intentionally omitted: outputs tool calls as plain text JSON,
+    // not OpenAI-compatible structured function calls. Will be skipped when
+    // request has tools to prevent the "talking to itself" bug.
   },
   {
     id: "xai/grok-4-0709",
@@ -2108,7 +2167,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 1.5,
     contextWindow: 131072,
     maxOutput: 16384,
-    reasoning: true
+    reasoning: true,
+    toolCalling: true
   },
   {
     id: "xai/grok-2-vision",
@@ -2118,7 +2178,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 10,
     contextWindow: 131072,
     maxOutput: 16384,
-    vision: true
+    vision: true,
+    toolCalling: true
   },
   // MiniMax
   {
@@ -2130,7 +2191,8 @@ var BLOCKRUN_MODELS = [
     contextWindow: 204800,
     maxOutput: 16384,
     reasoning: true,
-    agentic: true
+    agentic: true,
+    toolCalling: true
   },
   // NVIDIA - Free/cheap models
   {
@@ -2141,6 +2203,8 @@ var BLOCKRUN_MODELS = [
     outputPrice: 0,
     contextWindow: 128e3,
     maxOutput: 16384
+    // toolCalling intentionally omitted: free model, structured function
+    // calling support unverified. Excluded from tool-heavy routing paths.
   },
   {
     id: "nvidia/kimi-k2.5",
@@ -2149,7 +2213,8 @@ var BLOCKRUN_MODELS = [
     inputPrice: 0.55,
     outputPrice: 2.5,
     contextWindow: 262144,
-    maxOutput: 16384
+    maxOutput: 16384,
+    toolCalling: true
   }
 ];
 function toOpenClawModel(m) {
@@ -2178,6 +2243,16 @@ var OPENCLAW_MODELS = [
   ...BLOCKRUN_MODELS.map(toOpenClawModel),
   ...ALIAS_MODELS
 ];
+function supportsToolCalling(modelId) {
+  const normalized = modelId.replace("blockrun/", "");
+  const model = BLOCKRUN_MODELS.find((m) => m.id === normalized);
+  return model?.toolCalling ?? false;
+}
+function supportsVision(modelId) {
+  const normalized = modelId.replace("blockrun/", "");
+  const model = BLOCKRUN_MODELS.find((m) => m.id === normalized);
+  return model?.vision ?? false;
+}
 function getModelContextWindow(modelId) {
   const normalized = modelId.replace("blockrun/", "");
   const model = BLOCKRUN_MODELS.find((m) => m.id === normalized);
@@ -2828,6 +2903,102 @@ var BalanceMonitor = class {
   }
 };
 
+// src/solana-balance.ts
+import { address as solAddress, createSolanaRpc } from "@solana/kit";
+var SOLANA_USDC_MINT = "EPjFWdd5AufqSSqeM2qN1xzybapC8G4wEGGkZwyTDt1v";
+var SOLANA_DEFAULT_RPC = "https://api.mainnet-beta.solana.com";
+var BALANCE_TIMEOUT_MS = 1e4;
+var CACHE_TTL_MS2 = 3e4;
+var SolanaBalanceMonitor = class {
+  rpc;
+  walletAddress;
+  cachedBalance = null;
+  cachedAt = 0;
+  constructor(walletAddress, rpcUrl) {
+    this.walletAddress = walletAddress;
+    const url = rpcUrl || process["env"].CLAWROUTER_SOLANA_RPC_URL || SOLANA_DEFAULT_RPC;
+    this.rpc = createSolanaRpc(url);
+  }
+  async checkBalance() {
+    const now = Date.now();
+    if (this.cachedBalance !== null && now - this.cachedAt < CACHE_TTL_MS2) {
+      return this.buildInfo(this.cachedBalance);
+    }
+    const balance = await this.fetchBalance();
+    this.cachedBalance = balance;
+    this.cachedAt = now;
+    return this.buildInfo(balance);
+  }
+  deductEstimated(amountMicros) {
+    if (this.cachedBalance !== null && this.cachedBalance >= amountMicros) {
+      this.cachedBalance -= amountMicros;
+    }
+  }
+  invalidate() {
+    this.cachedBalance = null;
+    this.cachedAt = 0;
+  }
+  async refresh() {
+    this.invalidate();
+    return this.checkBalance();
+  }
+  /**
+   * Check if balance is sufficient for an estimated cost.
+   */
+  async checkSufficient(estimatedCostMicros) {
+    const info = await this.checkBalance();
+    if (info.balance >= estimatedCostMicros) {
+      return { sufficient: true, info };
+    }
+    const shortfall = estimatedCostMicros - info.balance;
+    return {
+      sufficient: false,
+      info,
+      shortfall: this.formatUSDC(shortfall)
+    };
+  }
+  /**
+   * Format USDC amount (in micros) as "$X.XX".
+   */
+  formatUSDC(amountMicros) {
+    const dollars = Number(amountMicros) / 1e6;
+    return `$${dollars.toFixed(2)}`;
+  }
+  getWalletAddress() {
+    return this.walletAddress;
+  }
+  async fetchBalance() {
+    const owner = solAddress(this.walletAddress);
+    const mint = solAddress(SOLANA_USDC_MINT);
+    const controller = new AbortController();
+    const timer = setTimeout(() => controller.abort(), BALANCE_TIMEOUT_MS);
+    try {
+      const response = await this.rpc.getTokenAccountsByOwner(owner, { mint }, { encoding: "jsonParsed" }).send({ abortSignal: controller.signal });
+      if (response.value.length === 0) return 0n;
+      let total = 0n;
+      for (const account of response.value) {
+        const parsed = account.account.data;
+        total += BigInt(parsed.parsed.info.tokenAmount.amount);
+      }
+      return total;
+    } catch (err) {
+      throw new Error(`Failed to fetch Solana USDC balance: ${err instanceof Error ? err.message : String(err)}`);
+    } finally {
+      clearTimeout(timer);
+    }
+  }
+  buildInfo(balance) {
+    const dollars = Number(balance) / 1e6;
+    return {
+      balance,
+      balanceUSD: `$${dollars.toFixed(2)}`,
+      isLow: balance < 1000000n,
+      isEmpty: balance < 100n,
+      walletAddress: this.walletAddress
+    };
+  }
+};
+
 // src/auth.ts
 import { writeFile, mkdir as mkdir2 } from "fs/promises";
 import { join as join4 } from "path";
@@ -2884,7 +3055,9 @@ async function loadSavedWallet() {
     console.error(`[ClawRouter] \u2717 CRITICAL: Wallet file exists but has invalid format!`);
     console.error(`[ClawRouter]   File: ${WALLET_FILE}`);
     console.error(`[ClawRouter]   Expected: 0x followed by 64 hex characters (66 chars total)`);
-    console.error(`[ClawRouter]   To fix: restore your backup key or set BLOCKRUN_WALLET_KEY env var`);
+    console.error(
+      `[ClawRouter]   To fix: restore your backup key or set BLOCKRUN_WALLET_KEY env var`
+    );
     throw new Error(
       `Wallet file at ${WALLET_FILE} is corrupted or has wrong format. Refusing to auto-generate new wallet to protect existing funds. Restore your backup key or set BLOCKRUN_WALLET_KEY environment variable.`
     );
@@ -2897,7 +3070,8 @@ async function loadSavedWallet() {
         `[ClawRouter] \u2717 Failed to read wallet file: ${err instanceof Error ? err.message : String(err)}`
       );
       throw new Error(
-        `Cannot read wallet file at ${WALLET_FILE}: ${err instanceof Error ? err.message : String(err)}. Refusing to auto-generate new wallet to protect existing funds. Fix file permissions or set BLOCKRUN_WALLET_KEY environment variable.`
+        `Cannot read wallet file at ${WALLET_FILE}: ${err instanceof Error ? err.message : String(err)}. Refusing to auto-generate new wallet to protect existing funds. Fix file permissions or set BLOCKRUN_WALLET_KEY environment variable.`,
+        { cause: err }
       );
     }
   }
@@ -2938,7 +3112,8 @@ async function generateAndSaveWallet() {
     console.log(`[ClawRouter] Wallet saved and verified at ${WALLET_FILE}`);
   } catch (err) {
     throw new Error(
-      `Failed to verify wallet file after creation: ${err instanceof Error ? err.message : String(err)}`
+      `Failed to verify wallet file after creation: ${err instanceof Error ? err.message : String(err)}`,
+      { cause: err }
     );
   }
   console.log(`[ClawRouter]`);
@@ -3767,8 +3942,9 @@ function shouldCompress(messages) {
 }
 
 // src/session.ts
+import { createHash as createHash3 } from "crypto";
 var DEFAULT_SESSION_CONFIG = {
-  enabled: false,
+  enabled: true,
   timeoutMs: 30 * 60 * 1e3,
   // 30 minutes
   headerName: "x-session-id"
@@ -3823,7 +3999,10 @@ var SessionStore = class {
         tier,
         createdAt: now,
         lastUsedAt: now,
-        requestCount: 1
+        requestCount: 1,
+        recentHashes: [],
+        strikes: 0,
+        escalated: false
       });
     }
   }
@@ -3875,6 +4054,43 @@ var SessionStore = class {
       }
     }
   }
+  /**
+   * Record a request content hash and detect repetitive patterns.
+   * Returns true if escalation should be triggered (3+ consecutive similar requests).
+   */
+  recordRequestHash(sessionId, hash) {
+    const entry = this.sessions.get(sessionId);
+    if (!entry) return false;
+    const prev = entry.recentHashes;
+    if (prev.length > 0 && prev[prev.length - 1] === hash) {
+      entry.strikes++;
+    } else {
+      entry.strikes = 0;
+    }
+    entry.recentHashes.push(hash);
+    if (entry.recentHashes.length > 3) {
+      entry.recentHashes.shift();
+    }
+    return entry.strikes >= 2 && !entry.escalated;
+  }
+  /**
+   * Escalate session to next tier. Returns the new model/tier or null if already at max.
+   */
+  escalateSession(sessionId, tierConfigs) {
+    const entry = this.sessions.get(sessionId);
+    if (!entry) return null;
+    const TIER_ORDER = ["SIMPLE", "MEDIUM", "COMPLEX", "REASONING"];
+    const currentIdx = TIER_ORDER.indexOf(entry.tier);
+    if (currentIdx < 0 || currentIdx >= TIER_ORDER.length - 1) return null;
+    const nextTier = TIER_ORDER[currentIdx + 1];
+    const nextConfig = tierConfigs[nextTier];
+    if (!nextConfig) return null;
+    entry.model = nextConfig.primary;
+    entry.tier = nextTier;
+    entry.strikes = 0;
+    entry.escalated = true;
+    return { model: nextConfig.primary, tier: nextTier };
+  }
   /**
    * Stop the cleanup interval.
    */
@@ -3895,6 +4111,17 @@ function getSessionId(headers, headerName = DEFAULT_SESSION_CONFIG.headerName) {
   }
   return void 0;
 }
+function deriveSessionId(messages) {
+  const firstUser = messages.find((m) => m.role === "user");
+  if (!firstUser) return void 0;
+  const content = typeof firstUser.content === "string" ? firstUser.content : JSON.stringify(firstUser.content);
+  return createHash3("sha256").update(content).digest("hex").slice(0, 8);
+}
+function hashRequestContent(lastUserContent, toolCallNames) {
+  const normalized = lastUserContent.replace(/\s+/g, " ").trim().slice(0, 500);
+  const toolSuffix = toolCallNames?.length ? `|tools:${toolCallNames.sort().join(",")}` : "";
+  return createHash3("sha256").update(normalized + toolSuffix).digest("hex").slice(0, 12);
+}
 
 // src/updater.ts
 var NPM_REGISTRY = "https://registry.npmjs.org/@blockrun/clawrouter/latest";
@@ -4652,6 +4879,27 @@ async function proxyPartnerRequest(req, res, apiBase, payFetch) {
   }).catch(() => {
   });
 }
+async function uploadDataUriToHost(dataUri) {
+  const match = dataUri.match(/^data:(image\/\w+);base64,(.+)$/);
+  if (!match) throw new Error("Invalid data URI format");
+  const [, mimeType, b64Data] = match;
+  const ext = mimeType === "image/jpeg" ? "jpg" : mimeType.split("/")[1] ?? "png";
+  const buffer = Buffer.from(b64Data, "base64");
+  const blob = new Blob([buffer], { type: mimeType });
+  const form = new FormData();
+  form.append("reqtype", "fileupload");
+  form.append("fileToUpload", blob, `image.${ext}`);
+  const resp = await fetch("https://catbox.moe/user/api.php", {
+    method: "POST",
+    body: form
+  });
+  if (!resp.ok) throw new Error(`catbox.moe upload failed: HTTP ${resp.status}`);
+  const result = await resp.text();
+  if (result.startsWith("https://")) {
+    return result.trim();
+  }
+  throw new Error(`catbox.moe upload failed: ${result}`);
+}
 async function startProxy(options) {
   const walletKey = typeof options.wallet === "string" ? options.wallet : options.wallet.key;
   const solanaPrivateKeyBytes = typeof options.wallet === "string" ? void 0 : options.wallet.solanaPrivateKeyBytes;
@@ -4666,7 +4914,6 @@ async function startProxy(options) {
   const existingProxy = await checkExistingProxy(listenPort);
   if (existingProxy) {
     const account2 = privateKeyToAccount3(walletKey);
-    const balanceMonitor2 = new BalanceMonitor(account2.address);
     const baseUrl2 = `http://127.0.0.1:${listenPort}`;
     if (existingProxy.wallet !== account2.address) {
       console.warn(
@@ -4691,6 +4938,7 @@ async function startProxy(options) {
       const solanaSigner = await createKeyPairSignerFromPrivateKeyBytes(solanaPrivateKeyBytes);
       reuseSolanaAddress = solanaSigner.address;
     }
+    const balanceMonitor2 = paymentChain === "solana" && reuseSolanaAddress ? new SolanaBalanceMonitor(reuseSolanaAddress) : new BalanceMonitor(account2.address);
     options.onReady?.(listenPort);
     return {
       port: listenPort,
@@ -4721,8 +4969,10 @@ async function startProxy(options) {
     const chain = network.startsWith("eip155") ? "Base (EVM)" : network.startsWith("solana") ? "Solana" : network;
     console.log(`[ClawRouter] Payment signed on ${chain} (${network})`);
   });
-  const payFetch = createPayFetchWithPreAuth(fetch, x402);
-  const balanceMonitor = new BalanceMonitor(account.address);
+  const payFetch = createPayFetchWithPreAuth(fetch, x402, void 0, {
+    skipPreAuth: paymentChain === "solana"
+  });
+  const balanceMonitor = paymentChain === "solana" && solanaAddress ? new SolanaBalanceMonitor(solanaAddress) : new BalanceMonitor(account.address);
   const routingConfig = mergeRoutingConfig(options.routingConfig);
   const modelPricing = buildModelPricing();
   const routerOpts = {
@@ -5075,14 +5325,19 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
   }
   let body = Buffer.concat(bodyChunks);
   const originalContextSizeKB = Math.ceil(body.length / 1024);
+  const debugMode = req.headers["x-clawrouter-debug"] !== "false";
   let routingDecision;
+  let hasTools = false;
+  let hasVision = false;
   let isStreaming = false;
   let modelId = "";
   let maxTokens = 4096;
   let routingProfile = null;
   let accumulatedContent = "";
+  let responseInputTokens;
   const isChatCompletion = req.url?.includes("/chat/completions");
   const sessionId = getSessionId(req.headers);
+  let effectiveSessionId = sessionId;
   if (isChatCompletion && body.length > 0) {
     try {
       const parsed = JSON.parse(body.toString());
@@ -5090,10 +5345,12 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
       modelId = parsed.model || "";
       maxTokens = parsed.max_tokens || 4096;
       let bodyModified = false;
-      if (sessionId && Array.isArray(parsed.messages)) {
-        const messages = parsed.messages;
-        const lastUserMsg = [...messages].reverse().find((m) => m.role === "user");
-        const lastContent = typeof lastUserMsg?.content === "string" ? lastUserMsg.content : "";
+      const parsedMessages = Array.isArray(parsed.messages) ? parsed.messages : [];
+      const lastUserMsg = [...parsedMessages].reverse().find((m) => m.role === "user");
+      const rawLastContent = lastUserMsg?.content;
+      const lastContent = typeof rawLastContent === "string" ? rawLastContent : Array.isArray(rawLastContent) ? rawLastContent.filter((b) => b.type === "text").map((b) => b.text ?? "").join(" ") : "";
+      if (sessionId && parsedMessages.length > 0) {
+        const messages = parsedMessages;
         if (sessionJournal.needsContext(lastContent)) {
           const journalText = sessionJournal.format(sessionId);
           if (journalText) {
@@ -5114,6 +5371,303 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
           }
         }
       }
+      if (lastContent.startsWith("/debug")) {
+        const debugPrompt = lastContent.slice("/debug".length).trim() || "hello";
+        const messages = parsed.messages;
+        const systemMsg = messages?.find((m) => m.role === "system");
+        const systemPrompt = typeof systemMsg?.content === "string" ? systemMsg.content : void 0;
+        const fullText = `${systemPrompt ?? ""} ${debugPrompt}`;
+        const estimatedTokens = Math.ceil(fullText.length / 4);
+        const normalizedModel2 = typeof parsed.model === "string" ? parsed.model.trim().toLowerCase() : "";
+        const profileName = normalizedModel2.replace("blockrun/", "");
+        const debugProfile = ["free", "eco", "auto", "premium"].includes(profileName) ? profileName : "auto";
+        const scoring = classifyByRules(
+          debugPrompt,
+          systemPrompt,
+          estimatedTokens,
+          DEFAULT_ROUTING_CONFIG.scoring
+        );
+        const debugRouting = route(debugPrompt, systemPrompt, maxTokens, {
+          ...routerOpts,
+          routingProfile: debugProfile
+        });
+        const dimLines = (scoring.dimensions ?? []).map((d) => {
+          const nameStr = (d.name + ":").padEnd(24);
+          const scoreStr = d.score.toFixed(2).padStart(6);
+          const sigStr = d.signal ? `  [${d.signal}]` : "";
+          return `  ${nameStr}${scoreStr}${sigStr}`;
+        }).join("\n");
+        const sess = sessionId ? sessionStore.getSession(sessionId) : void 0;
+        const sessLine = sess ? `Session: ${sessionId.slice(0, 8)}... \u2192 pinned: ${sess.model} (${sess.requestCount} requests)` : sessionId ? `Session: ${sessionId.slice(0, 8)}... \u2192 no pinned model` : "Session: none";
+        const { simpleMedium, mediumComplex, complexReasoning } = DEFAULT_ROUTING_CONFIG.scoring.tierBoundaries;
+        const debugText = [
+          "ClawRouter Debug",
+          "",
+          `Profile: ${debugProfile} | Tier: ${debugRouting.tier} | Model: ${debugRouting.model}`,
+          `Confidence: ${debugRouting.confidence.toFixed(2)} | Cost: $${debugRouting.costEstimate.toFixed(4)} | Savings: ${(debugRouting.savings * 100).toFixed(0)}%`,
+          `Reasoning: ${debugRouting.reasoning}`,
+          "",
+          `Scoring (weighted: ${scoring.score.toFixed(3)})`,
+          dimLines,
+          "",
+          `Tier Boundaries: SIMPLE <${simpleMedium.toFixed(2)} | MEDIUM <${mediumComplex.toFixed(2)} | COMPLEX <${complexReasoning.toFixed(2)} | REASONING >=${complexReasoning.toFixed(2)}`,
+          "",
+          sessLine
+        ].join("\n");
+        const completionId = `chatcmpl-debug-${Date.now()}`;
+        const timestamp = Math.floor(Date.now() / 1e3);
+        const syntheticResponse = {
+          id: completionId,
+          object: "chat.completion",
+          created: timestamp,
+          model: "clawrouter/debug",
+          choices: [
+            {
+              index: 0,
+              message: { role: "assistant", content: debugText },
+              finish_reason: "stop"
+            }
+          ],
+          usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 }
+        };
+        if (isStreaming) {
+          res.writeHead(200, {
+            "Content-Type": "text/event-stream",
+            "Cache-Control": "no-cache",
+            Connection: "keep-alive"
+          });
+          const sseChunk = {
+            id: completionId,
+            object: "chat.completion.chunk",
+            created: timestamp,
+            model: "clawrouter/debug",
+            choices: [
+              {
+                index: 0,
+                delta: { role: "assistant", content: debugText },
+                finish_reason: null
+              }
+            ]
+          };
+          const sseDone = {
+            id: completionId,
+            object: "chat.completion.chunk",
+            created: timestamp,
+            model: "clawrouter/debug",
+            choices: [{ index: 0, delta: {}, finish_reason: "stop" }]
+          };
+          res.write(`data: ${JSON.stringify(sseChunk)}
+
+`);
+          res.write(`data: ${JSON.stringify(sseDone)}
+
+`);
+          res.write("data: [DONE]\n\n");
+          res.end();
+        } else {
+          res.writeHead(200, { "Content-Type": "application/json" });
+          res.end(JSON.stringify(syntheticResponse));
+        }
+        console.log(`[ClawRouter] /debug command \u2192 ${debugRouting.tier} | ${debugRouting.model}`);
+        return;
+      }
+      if (lastContent.startsWith("/imagegen")) {
+        const imageArgs = lastContent.slice("/imagegen".length).trim();
+        let imageModel = "google/nano-banana";
+        let imageSize = "1024x1024";
+        let imagePrompt = imageArgs;
+        const modelMatch = imageArgs.match(/--model\s+(\S+)/);
+        if (modelMatch) {
+          const raw = modelMatch[1];
+          const IMAGE_MODEL_ALIASES = {
+            "dall-e-3": "openai/dall-e-3",
+            dalle3: "openai/dall-e-3",
+            dalle: "openai/dall-e-3",
+            "gpt-image": "openai/gpt-image-1",
+            "gpt-image-1": "openai/gpt-image-1",
+            flux: "black-forest/flux-1.1-pro",
+            "flux-pro": "black-forest/flux-1.1-pro",
+            banana: "google/nano-banana",
+            "nano-banana": "google/nano-banana",
+            "banana-pro": "google/nano-banana-pro",
+            "nano-banana-pro": "google/nano-banana-pro"
+          };
+          imageModel = IMAGE_MODEL_ALIASES[raw] ?? raw;
+          imagePrompt = imagePrompt.replace(/--model\s+\S+/, "").trim();
+        }
+        const sizeMatch = imageArgs.match(/--size\s+(\d+x\d+)/);
+        if (sizeMatch) {
+          imageSize = sizeMatch[1];
+          imagePrompt = imagePrompt.replace(/--size\s+\d+x\d+/, "").trim();
+        }
+        if (!imagePrompt) {
+          const errorText = [
+            "Usage: /imagegen <prompt>",
+            "",
+            "Options:",
+            "  --model <model>  Model to use (default: nano-banana)",
+            "  --size <WxH>     Image size (default: 1024x1024)",
+            "",
+            "Models:",
+            "  nano-banana       Google Gemini Flash \u2014 $0.05/image",
+            "  banana-pro        Google Gemini Pro \u2014 $0.10/image (up to 4K)",
+            "  dall-e-3          OpenAI DALL-E 3 \u2014 $0.04/image",
+            "  gpt-image         OpenAI GPT Image 1 \u2014 $0.02/image",
+            "  flux              Black Forest Flux 1.1 Pro \u2014 $0.04/image",
+            "",
+            "Examples:",
+            "  /imagegen a cat wearing sunglasses",
+            "  /imagegen --model dall-e-3 a futuristic city at sunset",
+            "  /imagegen --model banana-pro --size 2048x2048 mountain landscape"
+          ].join("\n");
+          const completionId = `chatcmpl-image-${Date.now()}`;
+          const timestamp = Math.floor(Date.now() / 1e3);
+          if (isStreaming) {
+            res.writeHead(200, {
+              "Content-Type": "text/event-stream",
+              "Cache-Control": "no-cache",
+              Connection: "keep-alive"
+            });
+            res.write(
+              `data: ${JSON.stringify({ id: completionId, object: "chat.completion.chunk", created: timestamp, model: "clawrouter/image", choices: [{ index: 0, delta: { role: "assistant", content: errorText }, finish_reason: null }] })}
+
+`
+            );
+            res.write(
+              `data: ${JSON.stringify({ id: completionId, object: "chat.completion.chunk", created: timestamp, model: "clawrouter/image", choices: [{ index: 0, delta: {}, finish_reason: "stop" }] })}
+
+`
+            );
+            res.write("data: [DONE]\n\n");
+            res.end();
+          } else {
+            res.writeHead(200, { "Content-Type": "application/json" });
+            res.end(
+              JSON.stringify({
+                id: completionId,
+                object: "chat.completion",
+                created: timestamp,
+                model: "clawrouter/image",
+                choices: [
+                  {
+                    index: 0,
+                    message: { role: "assistant", content: errorText },
+                    finish_reason: "stop"
+                  }
+                ],
+                usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 }
+              })
+            );
+          }
+          console.log(`[ClawRouter] /imagegen command \u2192 showing usage help`);
+          return;
+        }
+        console.log(
+          `[ClawRouter] /imagegen command \u2192 ${imageModel} (${imageSize}): ${imagePrompt.slice(0, 80)}...`
+        );
+        try {
+          const imageUpstreamUrl = `${apiBase}/v1/images/generations`;
+          const imageBody = JSON.stringify({
+            model: imageModel,
+            prompt: imagePrompt,
+            size: imageSize,
+            n: 1
+          });
+          const imageResponse = await payFetch(imageUpstreamUrl, {
+            method: "POST",
+            headers: { "content-type": "application/json", "user-agent": USER_AGENT },
+            body: imageBody
+          });
+          const imageResult = await imageResponse.json();
+          let responseText;
+          if (!imageResponse.ok || imageResult.error) {
+            const errMsg = typeof imageResult.error === "string" ? imageResult.error : imageResult.error?.message ?? `HTTP ${imageResponse.status}`;
+            responseText = `Image generation failed: ${errMsg}`;
+            console.log(`[ClawRouter] /imagegen error: ${errMsg}`);
+          } else {
+            const images = imageResult.data ?? [];
+            if (images.length === 0) {
+              responseText = "Image generation returned no results.";
+            } else {
+              const lines = [];
+              for (const img of images) {
+                if (img.url) {
+                  if (img.url.startsWith("data:")) {
+                    try {
+                      const hostedUrl = await uploadDataUriToHost(img.url);
+                      lines.push(hostedUrl);
+                    } catch (uploadErr) {
+                      console.error(
+                        `[ClawRouter] /imagegen: failed to upload data URI: ${uploadErr instanceof Error ? uploadErr.message : String(uploadErr)}`
+                      );
+                      lines.push(
+                        "Image generated but upload failed. Try again or use --model dall-e-3."
+                      );
+                    }
+                  } else {
+                    lines.push(img.url);
+                  }
+                }
+                if (img.revised_prompt) lines.push(`Revised prompt: ${img.revised_prompt}`);
+              }
+              lines.push("", `Model: ${imageModel} | Size: ${imageSize}`);
+              responseText = lines.join("\n");
+            }
+            console.log(`[ClawRouter] /imagegen success: ${images.length} image(s) generated`);
+          }
+          const completionId = `chatcmpl-image-${Date.now()}`;
+          const timestamp = Math.floor(Date.now() / 1e3);
+          if (isStreaming) {
+            res.writeHead(200, {
+              "Content-Type": "text/event-stream",
+              "Cache-Control": "no-cache",
+              Connection: "keep-alive"
+            });
+            res.write(
+              `data: ${JSON.stringify({ id: completionId, object: "chat.completion.chunk", created: timestamp, model: "clawrouter/image", choices: [{ index: 0, delta: { role: "assistant", content: responseText }, finish_reason: null }] })}
+
+`
+            );
+            res.write(
+              `data: ${JSON.stringify({ id: completionId, object: "chat.completion.chunk", created: timestamp, model: "clawrouter/image", choices: [{ index: 0, delta: {}, finish_reason: "stop" }] })}
+
+`
+            );
+            res.write("data: [DONE]\n\n");
+            res.end();
+          } else {
+            res.writeHead(200, { "Content-Type": "application/json" });
+            res.end(
+              JSON.stringify({
+                id: completionId,
+                object: "chat.completion",
+                created: timestamp,
+                model: "clawrouter/image",
+                choices: [
+                  {
+                    index: 0,
+                    message: { role: "assistant", content: responseText },
+                    finish_reason: "stop"
+                  }
+                ],
+                usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 }
+              })
+            );
+          }
+        } catch (err) {
+          const errMsg = err instanceof Error ? err.message : String(err);
+          console.error(`[ClawRouter] /imagegen error: ${errMsg}`);
+          if (!res.headersSent) {
+            res.writeHead(500, { "Content-Type": "application/json" });
+            res.end(
+              JSON.stringify({
+                error: { message: `Image generation failed: ${errMsg}`, type: "image_error" }
+              })
+            );
+          }
+        }
+        return;
+      }
       if (parsed.stream === true) {
         parsed.stream = false;
         bodyModified = true;
@@ -5154,54 +5708,118 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
             latencyMs: 0
           });
         } else {
-          const sessionId2 = getSessionId(
-            req.headers
-          );
-          const existingSession = sessionId2 ? sessionStore.getSession(sessionId2) : void 0;
+          effectiveSessionId = getSessionId(req.headers) ?? deriveSessionId(parsedMessages);
+          const existingSession = effectiveSessionId ? sessionStore.getSession(effectiveSessionId) : void 0;
+          const rawPrompt = lastUserMsg?.content;
+          const prompt = typeof rawPrompt === "string" ? rawPrompt : Array.isArray(rawPrompt) ? rawPrompt.filter((b) => b.type === "text").map((b) => b.text ?? "").join(" ") : "";
+          const systemMsg = parsedMessages.find((m) => m.role === "system");
+          const systemPrompt = typeof systemMsg?.content === "string" ? systemMsg.content : void 0;
+          const tools = parsed.tools;
+          hasTools = Array.isArray(tools) && tools.length > 0;
+          if (hasTools && tools) {
+            console.log(`[ClawRouter] Tools detected (${tools.length}), agentic mode via keywords`);
+          }
+          hasVision = parsedMessages.some((m) => {
+            if (Array.isArray(m.content)) {
+              return m.content.some((p) => p.type === "image_url");
+            }
+            return false;
+          });
+          if (hasVision) {
+            console.log(`[ClawRouter] Vision content detected, filtering to vision-capable models`);
+          }
+          routingDecision = route(prompt, systemPrompt, maxTokens, {
+            ...routerOpts,
+            routingProfile: routingProfile ?? void 0
+          });
           if (existingSession) {
-            console.log(
-              `[ClawRouter] Session ${sessionId2?.slice(0, 8)}... using pinned model: ${existingSession.model}`
-            );
-            parsed.model = existingSession.model;
-            modelId = existingSession.model;
-            bodyModified = true;
-            sessionStore.touchSession(sessionId2);
-          } else {
-            const messages = parsed.messages;
-            let lastUserMsg;
-            if (messages) {
-              for (let i = messages.length - 1; i >= 0; i--) {
-                if (messages[i].role === "user") {
-                  lastUserMsg = messages[i];
-                  break;
-                }
+            const tierRank = {
+              SIMPLE: 0,
+              MEDIUM: 1,
+              COMPLEX: 2,
+              REASONING: 3
+            };
+            const existingRank = tierRank[existingSession.tier] ?? 0;
+            const newRank = tierRank[routingDecision.tier] ?? 0;
+            if (newRank > existingRank) {
+              console.log(
+                `[ClawRouter] Session ${effectiveSessionId?.slice(0, 8)}... upgrading: ${existingSession.tier} \u2192 ${routingDecision.tier} (${routingDecision.model})`
+              );
+              parsed.model = routingDecision.model;
+              modelId = routingDecision.model;
+              bodyModified = true;
+              if (effectiveSessionId) {
+                sessionStore.setSession(
+                  effectiveSessionId,
+                  routingDecision.model,
+                  routingDecision.tier
+                );
               }
-            }
-            const systemMsg = messages?.find((m) => m.role === "system");
-            const prompt = typeof lastUserMsg?.content === "string" ? lastUserMsg.content : "";
-            const systemPrompt = typeof systemMsg?.content === "string" ? systemMsg.content : void 0;
-            const tools = parsed.tools;
-            const hasTools = Array.isArray(tools) && tools.length > 0;
-            if (hasTools && tools) {
+            } else {
               console.log(
-                `[ClawRouter] Tools detected (${tools.length}), agentic mode via keywords`
+                `[ClawRouter] Session ${effectiveSessionId?.slice(0, 8)}... keeping pinned model: ${existingSession.model} (${existingSession.tier} >= ${routingDecision.tier})`
               );
+              parsed.model = existingSession.model;
+              modelId = existingSession.model;
+              bodyModified = true;
+              sessionStore.touchSession(effectiveSessionId);
+              routingDecision = {
+                ...routingDecision,
+                model: existingSession.model,
+                tier: existingSession.tier
+              };
             }
-            routingDecision = route(prompt, systemPrompt, maxTokens, {
-              ...routerOpts,
-              routingProfile: routingProfile ?? void 0
-            });
+            const lastAssistantMsg = [...parsedMessages].reverse().find((m) => m.role === "assistant");
+            const assistantToolCalls = lastAssistantMsg?.tool_calls;
+            const toolCallNames = Array.isArray(assistantToolCalls) ? assistantToolCalls.map((tc) => tc.function?.name).filter((n) => Boolean(n)) : void 0;
+            const contentHash = hashRequestContent(prompt, toolCallNames);
+            const shouldEscalate = sessionStore.recordRequestHash(effectiveSessionId, contentHash);
+            if (shouldEscalate) {
+              const activeTierConfigs = (() => {
+                if (routingDecision.reasoning?.includes("agentic") && routerOpts.config.agenticTiers) {
+                  return routerOpts.config.agenticTiers;
+                }
+                if (routingProfile === "eco" && routerOpts.config.ecoTiers) {
+                  return routerOpts.config.ecoTiers;
+                }
+                if (routingProfile === "premium" && routerOpts.config.premiumTiers) {
+                  return routerOpts.config.premiumTiers;
+                }
+                return routerOpts.config.tiers;
+              })();
+              const escalation = sessionStore.escalateSession(
+                effectiveSessionId,
+                activeTierConfigs
+              );
+              if (escalation) {
+                console.log(
+                  `[ClawRouter] \u26A1 3-strike escalation: ${existingSession.model} \u2192 ${escalation.model} (${existingSession.tier} \u2192 ${escalation.tier})`
+                );
+                parsed.model = escalation.model;
+                modelId = escalation.model;
+                routingDecision = {
+                  ...routingDecision,
+                  model: escalation.model,
+                  tier: escalation.tier
+                };
+              }
+            }
+          } else {
             parsed.model = routingDecision.model;
             modelId = routingDecision.model;
             bodyModified = true;
-            if (sessionId2) {
-              sessionStore.setSession(sessionId2, routingDecision.model, routingDecision.tier);
+            if (effectiveSessionId) {
+              sessionStore.setSession(
+                effectiveSessionId,
+                routingDecision.model,
+                routingDecision.tier
+              );
               console.log(
-                `[ClawRouter] Session ${sessionId2.slice(0, 8)}... pinned to model: ${routingDecision.model}`
+                `[ClawRouter] Session ${effectiveSessionId.slice(0, 8)}... pinned to model: ${routingDecision.model}`
               );
             }
-            options.onRouted?.(routingDecision);
           }
+          options.onRouted?.(routingDecision);
         }
       }
       if (bodyModified) {
@@ -5294,6 +5912,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
   }
   deduplicator.markInflight(dedupKey);
   let estimatedCostMicros;
+  let balanceFallbackNotice;
   const isFreeModel = modelId === FREE_MODEL;
   if (modelId && !options.skipBalanceCheck && !isFreeModel) {
     const estimated = estimateAmount(modelId, body.length, maxTokens);
@@ -5304,12 +5923,17 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
       if (sufficiency.info.isEmpty || !sufficiency.sufficient) {
         const originalModel = modelId;
         console.log(
-          `[ClawRouter] Wallet ${sufficiency.info.isEmpty ? "empty" : "insufficient"} ($${sufficiency.info.balanceUSD}), falling back to free model: ${FREE_MODEL} (requested: ${originalModel})`
+          `[ClawRouter] Wallet ${sufficiency.info.isEmpty ? "empty" : "insufficient"} (${sufficiency.info.balanceUSD}), falling back to free model: ${FREE_MODEL} (requested: ${originalModel})`
         );
         modelId = FREE_MODEL;
         const parsed = JSON.parse(body.toString());
         parsed.model = FREE_MODEL;
         body = Buffer.from(JSON.stringify(parsed));
+        balanceFallbackNotice = sufficiency.info.isEmpty ? `> **\u26A0\uFE0F Wallet empty** \u2014 using free model. Fund your wallet to use ${originalModel}.
+
+` : `> **\u26A0\uFE0F Insufficient balance** (${sufficiency.info.balanceUSD}) \u2014 using free model instead of ${originalModel}.
+
+`;
         options.onLowBalance?.({
           balanceUSD: sufficiency.info.balanceUSD,
           walletAddress: sufficiency.info.walletAddress
@@ -5373,8 +5997,18 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
     if (routingDecision) {
       const estimatedInputTokens = Math.ceil(body.length / 4);
       const estimatedTotalTokens = estimatedInputTokens + maxTokens;
-      const useAgenticTiers = routingDecision.reasoning?.includes("agentic") && routerOpts.config.agenticTiers;
-      const tierConfigs = useAgenticTiers ? routerOpts.config.agenticTiers : routerOpts.config.tiers;
+      const tierConfigs = (() => {
+        if (routingDecision.reasoning?.includes("agentic") && routerOpts.config.agenticTiers) {
+          return routerOpts.config.agenticTiers;
+        }
+        if (routingProfile === "eco" && routerOpts.config.ecoTiers) {
+          return routerOpts.config.ecoTiers;
+        }
+        if (routingProfile === "premium" && routerOpts.config.premiumTiers) {
+          return routerOpts.config.premiumTiers;
+        }
+        return routerOpts.config.tiers;
+      })();
       const fullChain = getFallbackChain(routingDecision.tier, tierConfigs);
       const contextFiltered = getFallbackChainFiltered(
         routingDecision.tier,
@@ -5388,14 +6022,27 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
           `[ClawRouter] Context filter (~${estimatedTotalTokens} tokens): excluded ${contextExcluded.join(", ")}`
         );
       }
-      modelsToTry = contextFiltered.slice(0, MAX_FALLBACK_ATTEMPTS);
+      const toolFiltered = filterByToolCalling(contextFiltered, hasTools, supportsToolCalling);
+      const toolExcluded = contextFiltered.filter((m) => !toolFiltered.includes(m));
+      if (toolExcluded.length > 0) {
+        console.log(
+          `[ClawRouter] Tool-calling filter: excluded ${toolExcluded.join(", ")} (no structured function call support)`
+        );
+      }
+      const visionFiltered = filterByVision(toolFiltered, hasVision, supportsVision);
+      const visionExcluded = toolFiltered.filter((m) => !visionFiltered.includes(m));
+      if (visionExcluded.length > 0) {
+        console.log(
+          `[ClawRouter] Vision filter: excluded ${visionExcluded.join(", ")} (no vision support)`
+        );
+      }
+      modelsToTry = visionFiltered.slice(0, MAX_FALLBACK_ATTEMPTS);
       modelsToTry = prioritizeNonRateLimited(modelsToTry);
     } else {
-      if (modelId && modelId !== FREE_MODEL) {
-        modelsToTry = [modelId, FREE_MODEL];
-      } else {
-        modelsToTry = modelId ? [modelId] : [];
-      }
+      modelsToTry = modelId ? [modelId] : [];
+    }
+    if (!modelsToTry.includes(FREE_MODEL)) {
+      modelsToTry.push(FREE_MODEL);
     }
     let upstream;
     let lastError;
@@ -5429,6 +6076,17 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         if (result.errorStatus === 429) {
           markRateLimited(tryModel);
         }
+        const isPaymentErr = /payment.*verification.*failed|insufficient.*funds/i.test(
+          result.errorBody || ""
+        );
+        if (isPaymentErr && tryModel !== FREE_MODEL) {
+          const freeIdx = modelsToTry.indexOf(FREE_MODEL);
+          if (freeIdx > i + 1) {
+            console.log(`[ClawRouter] Payment error \u2014 skipping to free model: ${FREE_MODEL}`);
+            i = freeIdx - 1;
+            continue;
+          }
+        }
         console.log(
           `[ClawRouter] Provider error from ${tryModel}, trying fallback: ${result.errorBody?.slice(0, 100)}`
         );
@@ -5446,6 +6104,12 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
       clearInterval(heartbeatInterval);
       heartbeatInterval = void 0;
     }
+    if (debugMode && headersSentEarly && routingDecision) {
+      const debugComment = `: x-clawrouter-debug profile=${routingProfile ?? "auto"} tier=${routingDecision.tier} model=${actualModelUsed} agentic=${routingDecision.agenticScore?.toFixed(2) ?? "n/a"} confidence=${routingDecision.confidence.toFixed(2)} reasoning=${routingDecision.reasoning}
+
+`;
+      safeWrite(res, debugComment);
+    }
     if (routingDecision && actualModelUsed !== routingDecision.model) {
       const estimatedInputTokens = Math.ceil(body.length / 4);
       const newCosts = calculateModelCost(
@@ -5464,6 +6128,12 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         savings: newCosts.savings
       };
       options.onRouted?.(routingDecision);
+      if (effectiveSessionId) {
+        sessionStore.setSession(effectiveSessionId, actualModelUsed, routingDecision.tier);
+        console.log(
+          `[ClawRouter] Session ${effectiveSessionId.slice(0, 8)}... updated pin to fallback: ${actualModelUsed}`
+        );
+      }
     }
     if (!upstream) {
       const rawErrBody = lastError?.body || "All models in fallback chain failed";
@@ -5526,6 +6196,10 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         const jsonStr = jsonBody.toString();
         try {
           const rsp = JSON.parse(jsonStr);
+          if (rsp.usage && typeof rsp.usage === "object") {
+            const u = rsp.usage;
+            if (typeof u.prompt_tokens === "number") responseInputTokens = u.prompt_tokens;
+          }
           const baseChunk = {
             id: rsp.id ?? `chatcmpl-${Date.now()}`,
             object: "chat.completion.chunk",
@@ -5551,6 +6225,18 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
 `;
               safeWrite(res, roleData);
               responseChunks.push(Buffer.from(roleData));
+              if (balanceFallbackNotice) {
+                const noticeChunk = {
+                  ...baseChunk,
+                  choices: [{ index, delta: { content: balanceFallbackNotice }, logprobs: null, finish_reason: null }]
+                };
+                const noticeData = `data: ${JSON.stringify(noticeChunk)}
+
+`;
+                safeWrite(res, noticeData);
+                responseChunks.push(Buffer.from(noticeData));
+                balanceFallbackNotice = void 0;
+              }
               if (content) {
                 const contentChunk = {
                   ...baseChunk,
@@ -5625,23 +6311,46 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
       });
       responseHeaders["x-context-used-kb"] = String(originalContextSizeKB);
       responseHeaders["x-context-limit-kb"] = String(CONTEXT_LIMIT_KB);
-      res.writeHead(upstream.status, responseHeaders);
+      if (debugMode && routingDecision) {
+        responseHeaders["x-clawrouter-profile"] = routingProfile ?? "auto";
+        responseHeaders["x-clawrouter-tier"] = routingDecision.tier;
+        responseHeaders["x-clawrouter-model"] = actualModelUsed;
+        responseHeaders["x-clawrouter-confidence"] = routingDecision.confidence.toFixed(2);
+        responseHeaders["x-clawrouter-reasoning"] = routingDecision.reasoning;
+        if (routingDecision.agenticScore !== void 0) {
+          responseHeaders["x-clawrouter-agentic-score"] = routingDecision.agenticScore.toFixed(2);
+        }
+      }
+      const bodyParts = [];
       if (upstream.body) {
         const reader = upstream.body.getReader();
         try {
           while (true) {
             const { done, value } = await reader.read();
             if (done) break;
-            const chunk = Buffer.from(value);
-            safeWrite(res, chunk);
-            responseChunks.push(chunk);
+            bodyParts.push(Buffer.from(value));
           }
         } finally {
           reader.releaseLock();
         }
       }
+      let responseBody = Buffer.concat(bodyParts);
+      if (balanceFallbackNotice && responseBody.length > 0) {
+        try {
+          const parsed = JSON.parse(responseBody.toString());
+          if (parsed.choices?.[0]?.message?.content !== void 0) {
+            parsed.choices[0].message.content = balanceFallbackNotice + parsed.choices[0].message.content;
+            responseBody = Buffer.from(JSON.stringify(parsed));
+          }
+        } catch {
+        }
+        balanceFallbackNotice = void 0;
+      }
+      responseHeaders["content-length"] = String(responseBody.length);
+      res.writeHead(upstream.status, responseHeaders);
+      safeWrite(res, responseBody);
+      responseChunks.push(responseBody);
       res.end();
-      const responseBody = Buffer.concat(responseChunks);
       deduplicator.complete(dedupKey, {
         status: upstream.status,
         headers: responseHeaders,
@@ -5664,6 +6373,10 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         if (rspJson.choices?.[0]?.message?.content) {
           accumulatedContent = rspJson.choices[0].message.content;
         }
+        if (rspJson.usage && typeof rspJson.usage === "object") {
+          if (typeof rspJson.usage.prompt_tokens === "number")
+            responseInputTokens = rspJson.usage.prompt_tokens;
+        }
       } catch {
       }
     }
@@ -5689,7 +6402,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
     deduplicator.removeInflight(dedupKey);
     balanceMonitor.invalidate();
     if (err instanceof Error && err.name === "AbortError") {
-      throw new Error(`Request timed out after ${timeoutMs}ms`);
+      throw new Error(`Request timed out after ${timeoutMs}ms`, { cause: err });
     }
     throw err;
   }
@@ -5712,13 +6425,53 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
       cost: costWithBuffer,
       baselineCost: baselineWithBuffer,
       savings: accurateCosts.savings,
-      latencyMs: Date.now() - startTime
+      latencyMs: Date.now() - startTime,
+      ...responseInputTokens !== void 0 && { inputTokens: responseInputTokens }
     };
     logUsage(entry).catch(() => {
     });
   }
 }
 
+// src/report.ts
+async function generateReport(period, json = false) {
+  const days = period === "daily" ? 1 : period === "weekly" ? 7 : 30;
+  const stats = await getStats(days);
+  if (json) {
+    return JSON.stringify(stats, null, 2);
+  }
+  return formatMarkdownReport(period, days, stats);
+}
+function formatMarkdownReport(period, days, stats) {
+  const lines = [];
+  lines.push(`# ClawRouter ${capitalize(period)} Report`);
+  lines.push(`**Period:** Last ${days} day${days > 1 ? "s" : ""}`);
+  lines.push(`**Generated:** ${(/* @__PURE__ */ new Date()).toISOString()}`);
+  lines.push("");
+  lines.push("## \u{1F4CA} Usage Summary");
+  lines.push("");
+  lines.push(`| Metric | Value |`);
+  lines.push(`|--------|-------|`);
+  lines.push(`| Total Requests | ${stats.totalRequests} |`);
+  lines.push(`| Total Cost | $${stats.totalCost.toFixed(4)} |`);
+  lines.push(`| Baseline Cost | $${stats.totalBaselineCost.toFixed(4)} |`);
+  lines.push(`| **Savings** | **$${stats.totalSavings.toFixed(4)}** |`);
+  lines.push(`| Savings % | ${stats.savingsPercentage.toFixed(1)}% |`);
+  lines.push(`| Avg Latency | ${stats.avgLatencyMs.toFixed(0)}ms |`);
+  lines.push("");
+  lines.push("## \u{1F916} Model Distribution");
+  lines.push("");
+  const sortedModels = Object.entries(stats.byModel).sort((a, b) => b[1].count - a[1].count).slice(0, 10);
+  for (const [model, data] of sortedModels) {
+    lines.push(`- ${model}: ${data.count} reqs, $${data.cost.toFixed(4)}`);
+  }
+  lines.push("");
+  return lines.join("\n");
+}
+function capitalize(str) {
+  return str.charAt(0).toUpperCase() + str.slice(1);
+}
+
 // src/doctor.ts
 import { platform, arch, freemem, totalmem } from "os";
 import { createPublicClient as createPublicClient3, http as http3 } from "viem";
@@ -5811,7 +6564,6 @@ async function collectNetworkInfo() {
     blockrunLatency = Date.now() - start;
     blockrunReachable = response.ok || response.status === 402;
   } catch {
-    blockrunReachable = false;
   }
   let proxyRunning = false;
   try {
@@ -5821,7 +6573,6 @@ async function collectNetworkInfo() {
     });
     proxyRunning = response.ok;
   } catch {
-    proxyRunning = false;
   }
   return {
     blockrunApi: { reachable: blockrunReachable, latencyMs: blockrunLatency },
@@ -6065,6 +6816,7 @@ Usage:
   clawrouter [options]
   clawrouter doctor [opus] [question]
   clawrouter partners [test]
+  clawrouter report [daily|weekly|monthly] [--json]
 
 Options:
   --version, -v     Show version number
@@ -6107,6 +6859,9 @@ function parseArgs(args) {
     doctor: false,
     partners: false,
     partnersTest: false,
+    report: false,
+    reportPeriod: "daily",
+    reportJson: false,
     port: void 0
   };
   for (let i = 0; i < args.length; i++) {
@@ -6123,6 +6878,20 @@ function parseArgs(args) {
         result.partnersTest = true;
         i++;
       }
+    } else if (arg === "report") {
+      result.report = true;
+      const next = args[i + 1];
+      if (next && ["daily", "weekly", "monthly"].includes(next)) {
+        result.reportPeriod = next;
+        i++;
+        if (args[i + 1] === "--json") {
+          result.reportJson = true;
+          i++;
+        }
+      } else if (next === "--json") {
+        result.reportJson = true;
+        i++;
+      }
     } else if (arg === "--port" && args[i + 1]) {
       result.port = parseInt(args[i + 1], 10);
       i++;
@@ -6170,7 +6939,9 @@ ClawRouter Partner APIs (v${VERSION})
       console.log(`    ${svc.description}`);
       console.log(`    Tool:    blockrun_${svc.id}`);
       console.log(`    Method:  ${svc.method} /v1${svc.proxyPath}`);
-      console.log(`    Pricing: ${svc.pricing.perUnit} per ${svc.pricing.unit} (min ${svc.pricing.minimum}, max ${svc.pricing.maximum})`);
+      console.log(
+        `    Pricing: ${svc.pricing.perUnit} per ${svc.pricing.unit} (min ${svc.pricing.minimum}, max ${svc.pricing.maximum})`
+      );
       console.log();
     }
     if (args.partnersTest) {
@@ -6191,6 +6962,11 @@ ClawRouter Partner APIs (v${VERSION})
     }
     process.exit(0);
   }
+  if (args.report) {
+    const report = await generateReport(args.reportPeriod, args.reportJson);
+    console.log(report);
+    process.exit(0);
+  }
   const wallet = await resolveOrGenerateWalletKey();
   if (wallet.source === "generated") {
     console.log(`[ClawRouter] Generated new wallet: ${wallet.address}`);