@kognai/clawrouter-x402 0.1.1 → 0.1.3

package/dist/clawrouter-v2.js

@@ -288,11 +288,24 @@ async function callAnthropicDirect(model, messages, maxTokens = 4096) {
         throw new Error(`Anthropic direct API returned ${res.status}: ${res.data.slice(0, 300)}`);
     }
     const json = JSON.parse(res.data);
+    const inTok = json.usage?.input_tokens || 0;
+    const outTok = json.usage?.output_tokens || 0;
+    // TICKET-350 cost visibility: the direct path used to hardcode cost_usd=0, so
+    // the honest-routing haiku calls logged $0 and the ledger under-counted real
+    // spend. Compute the actual cost from token usage × per-million USD rates.
+    const DIRECT_RATES = {
+        'claude-haiku-4-5': { in: 0.80, out: 4.00 },
+        'claude-sonnet-4': { in: 3.00, out: 15.00 },
+        'claude-opus-4': { in: 15.00, out: 75.00 },
+    };
+    const rateKey = Object.keys(DIRECT_RATES).find(k => model.includes(k));
+    const rate = rateKey ? DIRECT_RATES[rateKey] : { in: 0.80, out: 4.00 }; // default ≈ haiku
+    const cost_usd = (inTok * rate.in + outTok * rate.out) / 1_000_000;
     return {
         content: json.content?.[0]?.text || '',
-        input_tokens: json.usage?.input_tokens || 0,
-        output_tokens: json.usage?.output_tokens || 0,
-        cost_usd: 0, // billing handled outside (no gateway metering on direct path)
+        input_tokens: inTok,
+        output_tokens: outTok,
+        cost_usd,
     };
 }
 // ── x402 Payment Signing (EIP-3009 TransferWithAuthorization) ─────────────────
@@ -723,6 +736,32 @@ async function routeCall(req) {
         output_tokens = result.output_tokens;
         cost_usd = 0; // local = free
     }
+    else if (route.tier === 'T2.5') {
+        // TICKET-350 honest routing: routine cloud work (the 'exec' tier) runs a
+        // CHEAP+FAST model DIRECTLY — bypassing the OpenClaw gateway, which flattened
+        // EVERY model to openclaw/main=opus AND had a cold ~12min first call. So
+        // routine code is now genuinely cheap + fast (and the cost_usd recorded is
+        // the REAL model price, not the gateway's). Hard/escalated tasks (T3/apex)
+        // still go through the gateway → opus. Falls back to the gateway if the
+        // direct call errors (e.g. missing key) so routing never hard-fails.
+        const cheapModel = process.env.KOGNAI_EXEC_MODEL || 'claude-haiku-4-5-20251001';
+        try {
+            const result = await callAnthropicDirect(cheapModel, messages, payload.max_tokens || 4096);
+            content = result.content;
+            input_tokens = result.input_tokens;
+            output_tokens = result.output_tokens;
+            cost_usd = result.cost_usd;
+            route = { tier: 'T2.5-EXEC', model: cheapModel, local: false };
+        }
+        catch (e) {
+            console.warn(`[ClawRouter] direct exec call failed (${e?.message || e}) — falling back to gateway`);
+            const result = await callCloudGateway(route.model, messages, payload.max_tokens || 4096);
+            content = result.content;
+            input_tokens = result.input_tokens;
+            output_tokens = result.output_tokens;
+            cost_usd = result.cost_usd;
+        }
+    }
     else {
         const result = await callCloudGateway(route.model, messages, payload.max_tokens || 4096);
         content = result.content;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kognai/clawrouter-x402",
-  "version": "0.1.1",
+  "version": "0.1.3",
   "private": false,
   "description": "ClawRouter v2 — Kognai's viem-backed x402 ModelRouter implementation. Injected into @kognai/orchestrator-core's router seam at startup, so core never depends on viem; products that don't route on-chain simply don't install this package.",
   "main": "dist/index.js",