@blockrun/clawrouter 0.12.39 → 0.12.40

package/dist/index.js

@@ -2022,6 +2022,99 @@ function getFallbackChainFiltered(tier, tierConfigs, estimatedTotalTokens, getCo
   return filtered;
 }
 
+// src/router/strategy.ts
+var RulesStrategy = class {
+  name = "rules";
+  route(prompt, systemPrompt, maxOutputTokens, options) {
+    const { config, modelPricing } = options;
+    const fullText = `${systemPrompt ?? ""} ${prompt}`;
+    const estimatedTokens = Math.ceil(fullText.length / 4);
+    const ruleResult = classifyByRules(prompt, systemPrompt, estimatedTokens, config.scoring);
+    const { routingProfile } = options;
+    let tierConfigs;
+    let profileSuffix;
+    let profile;
+    if (routingProfile === "eco" && config.ecoTiers) {
+      tierConfigs = config.ecoTiers;
+      profileSuffix = " | eco";
+      profile = "eco";
+    } else if (routingProfile === "premium" && config.premiumTiers) {
+      tierConfigs = config.premiumTiers;
+      profileSuffix = " | premium";
+      profile = "premium";
+    } else {
+      const agenticScore = ruleResult.agenticScore ?? 0;
+      const isAutoAgentic = agenticScore >= 0.5;
+      const isExplicitAgentic = config.overrides.agenticMode ?? false;
+      const hasToolsInRequest = options.hasTools ?? false;
+      const useAgenticTiers = (hasToolsInRequest || isAutoAgentic || isExplicitAgentic) && config.agenticTiers != null;
+      tierConfigs = useAgenticTiers ? config.agenticTiers : config.tiers;
+      profileSuffix = useAgenticTiers ? ` | agentic${hasToolsInRequest ? " (tools)" : ""}` : "";
+      profile = useAgenticTiers ? "agentic" : "auto";
+    }
+    const agenticScoreValue = ruleResult.agenticScore;
+    if (estimatedTokens > config.overrides.maxTokensForceComplex) {
+      const decision2 = selectModel(
+        "COMPLEX",
+        0.95,
+        "rules",
+        `Input exceeds ${config.overrides.maxTokensForceComplex} tokens${profileSuffix}`,
+        tierConfigs,
+        modelPricing,
+        estimatedTokens,
+        maxOutputTokens,
+        routingProfile,
+        agenticScoreValue
+      );
+      return { ...decision2, tierConfigs, profile };
+    }
+    const hasStructuredOutput = systemPrompt ? /json|structured|schema/i.test(systemPrompt) : false;
+    let tier;
+    let confidence;
+    const method = "rules";
+    let reasoning = `score=${ruleResult.score.toFixed(2)} | ${ruleResult.signals.join(", ")}`;
+    if (ruleResult.tier !== null) {
+      tier = ruleResult.tier;
+      confidence = ruleResult.confidence;
+    } else {
+      tier = config.overrides.ambiguousDefaultTier;
+      confidence = 0.5;
+      reasoning += ` | ambiguous -> default: ${tier}`;
+    }
+    if (hasStructuredOutput) {
+      const tierRank = { SIMPLE: 0, MEDIUM: 1, COMPLEX: 2, REASONING: 3 };
+      const minTier = config.overrides.structuredOutputMinTier;
+      if (tierRank[tier] < tierRank[minTier]) {
+        reasoning += ` | upgraded to ${minTier} (structured output)`;
+        tier = minTier;
+      }
+    }
+    reasoning += profileSuffix;
+    const decision = selectModel(
+      tier,
+      confidence,
+      method,
+      reasoning,
+      tierConfigs,
+      modelPricing,
+      estimatedTokens,
+      maxOutputTokens,
+      routingProfile,
+      agenticScoreValue
+    );
+    return { ...decision, tierConfigs, profile };
+  }
+};
+var registry = /* @__PURE__ */ new Map();
+registry.set("rules", new RulesStrategy());
+function getStrategy(name) {
+  const strategy = registry.get(name);
+  if (!strategy) {
+    throw new Error(`Unknown routing strategy: ${name}`);
+  }
+  return strategy;
+}
+
 // src/router/config.ts
 var DEFAULT_ROUTING_CONFIG = {
   version: "2.0",
@@ -3114,7 +3207,11 @@ var DEFAULT_ROUTING_CONFIG = {
     SIMPLE: {
       primary: "nvidia/gpt-oss-120b",
       // FREE! $0.00/$0.00
-      fallback: ["google/gemini-2.5-flash-lite", "google/gemini-2.5-flash", "deepseek/deepseek-chat"]
+      fallback: [
+        "google/gemini-2.5-flash-lite",
+        "google/gemini-2.5-flash",
+        "deepseek/deepseek-chat"
+      ]
     },
     MEDIUM: {
       primary: "google/gemini-2.5-flash-lite",
@@ -3239,77 +3336,8 @@ var DEFAULT_ROUTING_CONFIG = {
 
 // src/router/index.ts
 function route(prompt, systemPrompt, maxOutputTokens, options) {
-  const { config, modelPricing } = options;
-  const fullText = `${systemPrompt ?? ""} ${prompt}`;
-  const estimatedTokens = Math.ceil(fullText.length / 4);
-  const ruleResult = classifyByRules(prompt, systemPrompt, estimatedTokens, config.scoring);
-  const { routingProfile } = options;
-  let tierConfigs;
-  let profileSuffix;
-  if (routingProfile === "eco" && config.ecoTiers) {
-    tierConfigs = config.ecoTiers;
-    profileSuffix = " | eco";
-  } else if (routingProfile === "premium" && config.premiumTiers) {
-    tierConfigs = config.premiumTiers;
-    profileSuffix = " | premium";
-  } else {
-    const agenticScore = ruleResult.agenticScore ?? 0;
-    const isAutoAgentic = agenticScore >= 0.5;
-    const isExplicitAgentic = config.overrides.agenticMode ?? false;
-    const hasToolsInRequest = options.hasTools ?? false;
-    const useAgenticTiers = (hasToolsInRequest || isAutoAgentic || isExplicitAgentic) && config.agenticTiers != null;
-    tierConfigs = useAgenticTiers ? config.agenticTiers : config.tiers;
-    profileSuffix = useAgenticTiers ? ` | agentic${hasToolsInRequest ? " (tools)" : ""}` : "";
-  }
-  const agenticScoreValue = ruleResult.agenticScore;
-  if (estimatedTokens > config.overrides.maxTokensForceComplex) {
-    return selectModel(
-      "COMPLEX",
-      0.95,
-      "rules",
-      `Input exceeds ${config.overrides.maxTokensForceComplex} tokens${profileSuffix}`,
-      tierConfigs,
-      modelPricing,
-      estimatedTokens,
-      maxOutputTokens,
-      routingProfile,
-      agenticScoreValue
-    );
-  }
-  const hasStructuredOutput = systemPrompt ? /json|structured|schema/i.test(systemPrompt) : false;
-  let tier;
-  let confidence;
-  const method = "rules";
-  let reasoning = `score=${ruleResult.score.toFixed(2)} | ${ruleResult.signals.join(", ")}`;
-  if (ruleResult.tier !== null) {
-    tier = ruleResult.tier;
-    confidence = ruleResult.confidence;
-  } else {
-    tier = config.overrides.ambiguousDefaultTier;
-    confidence = 0.5;
-    reasoning += ` | ambiguous -> default: ${tier}`;
-  }
-  if (hasStructuredOutput) {
-    const tierRank = { SIMPLE: 0, MEDIUM: 1, COMPLEX: 2, REASONING: 3 };
-    const minTier = config.overrides.structuredOutputMinTier;
-    if (tierRank[tier] < tierRank[minTier]) {
-      reasoning += ` | upgraded to ${minTier} (structured output)`;
-      tier = minTier;
-    }
-  }
-  reasoning += profileSuffix;
-  return selectModel(
-    tier,
-    confidence,
-    method,
-    reasoning,
-    tierConfigs,
-    modelPricing,
-    estimatedTokens,
-    maxOutputTokens,
-    routingProfile,
-    agenticScoreValue
-  );
+  const strategy = getStrategy("rules");
+  return strategy.route(prompt, systemPrompt, maxOutputTokens, options);
 }
 
 // src/logger.ts
@@ -5454,6 +5482,12 @@ var ROUTING_PROFILES = /* @__PURE__ */ new Set([
   "premium"
 ]);
 var FREE_MODEL = "nvidia/gpt-oss-120b";
+var FREE_TIER_CONFIGS = {
+  SIMPLE: { primary: FREE_MODEL, fallback: [] },
+  MEDIUM: { primary: FREE_MODEL, fallback: [] },
+  COMPLEX: { primary: FREE_MODEL, fallback: [] },
+  REASONING: { primary: FREE_MODEL, fallback: [] }
+};
 var freeRequestCount = 0;
 var MAX_MESSAGES = 200;
 var CONTEXT_LIMIT_KB = 5120;
@@ -7328,7 +7362,17 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
 
 `;
           }
-          routingDecision = { model: freeModel, tier: "SIMPLE", confidence: 1, method: "rules", reasoning: "free profile" };
+          routingDecision = {
+            model: freeModel,
+            tier: "SIMPLE",
+            confidence: 1,
+            method: "rules",
+            reasoning: "free profile",
+            costEstimate: 0,
+            baselineCost: 0,
+            savings: 1,
+            tierConfigs: FREE_TIER_CONFIGS
+          };
         } else {
           effectiveSessionId = getSessionId(req.headers) ?? deriveSessionId(parsedMessages);
           const existingSession = effectiveSessionId ? sessionStore.getSession(effectiveSessionId) : void 0;
@@ -7419,18 +7463,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
             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 activeTierConfigs = routingDecision.tierConfigs ?? routerOpts.config.tiers;
               const escalation = sessionStore.escalateSession(
                 effectiveSessionId,
                 activeTierConfigs
@@ -7646,18 +7679,7 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
     if (routingDecision) {
       const estimatedInputTokens = Math.ceil(body.length / 4);
       const estimatedTotalTokens = estimatedInputTokens + maxTokens;
-      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 tierConfigs = routingDecision.tierConfigs ?? routerOpts.config.tiers;
       const fullChain = getFallbackChain(routingDecision.tier, tierConfigs);
       const contextFiltered = getFallbackChainFiltered(
         routingDecision.tier,
@@ -7737,6 +7759,14 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         status: result.errorStatus || 500
       };
       if (result.isProviderError && !isLastAttempt) {
+        const isExplicitModelError = !routingDecision;
+        const isUnknownExplicitModel = isExplicitModelError && /unknown.*model|invalid.*model/i.test(result.errorBody || "");
+        if (isUnknownExplicitModel) {
+          console.log(
+            `[ClawRouter] Explicit model error from ${tryModel}, not falling back: ${result.errorBody?.slice(0, 100)}`
+          );
+          break;
+        }
         if (result.errorStatus === 429) {
           markRateLimited(tryModel);
           try {