kimiflare 0.36.2 → 0.38.0

package/dist/index.js

@@ -486,6 +486,7 @@ async function* runKimi(opts2) {
   if (opts2.cloudMode && !opts2.cloudToken) {
     throw new KimiApiError("kimiflare: cloud mode requires a cloud token. Run `kimiflare auth cloud` to authenticate.", void 0, 401);
   }
+  const requestId = opts2.requestId ?? crypto.randomUUID();
   const { url, headers: gatewayHeaders } = buildKimiRequestTarget(opts2);
   const body = {
     messages: sanitizeMessagesForApi(opts2.messages),
@@ -510,6 +511,7 @@ async function* runKimi(opts2) {
         headers["X-Session-ID"] = opts2.sessionId;
         headers["x-session-affinity"] = opts2.sessionId;
       }
+      headers["X-Request-ID"] = requestId;
       res = await fetch(url, {
         method: "POST",
         headers,
@@ -547,7 +549,31 @@ async function* runKimi(opts2) {
     if (!res.body) throw new KimiApiError("kimiflare: empty response body", void 0, res.status);
     const meta = readGatewayMeta(res.headers);
     if (meta) yield { type: "gateway_meta", meta };
-    yield* parseStream(res.body, opts2.signal);
+    let lastUsage = null;
+    for await (const ev of parseStream(res.body, opts2.signal)) {
+      if (ev.type === "usage") lastUsage = ev.usage;
+      yield ev;
+    }
+    if (opts2.cloudMode && lastUsage && opts2.cloudToken) {
+      const reportUrl = "https://api.kimiflare.com/v1/usage/report";
+      const reportHeaders = {
+        Authorization: `Bearer ${opts2.cloudToken}`,
+        "Content-Type": "application/json"
+      };
+      if (opts2.cloudDeviceId) reportHeaders["X-Device-ID"] = opts2.cloudDeviceId;
+      if (opts2.sessionId) reportHeaders["X-Session-ID"] = opts2.sessionId;
+      fetch(reportUrl, {
+        method: "POST",
+        headers: reportHeaders,
+        body: JSON.stringify({
+          request_id: requestId,
+          prompt_tokens: lastUsage.prompt_tokens,
+          completion_tokens: lastUsage.completion_tokens,
+          cached_tokens: lastUsage.prompt_tokens_details?.cached_tokens ?? 0
+        })
+      }).catch(() => {
+      });
+    }
     return;
   }
 }
@@ -6878,9 +6904,9 @@ function buildRightParts(usage, contextLimit, sessionUsage, gatewayMeta, cloudMo
     parts.push(`out ${sessionUsage.completionTokens}`);
     parts.push(`ctx ${pct}%`);
     if (cloudMode) {
-      parts.push(`\x1B[9m${sessionUsage.cost.toFixed(5)}\x1B[29m`);
+      parts.push(`\x1B[9m$${sessionUsage.cost.toFixed(5)}\x1B[29m`);
     } else {
-      parts.push(`${sessionUsage.cost.toFixed(5)}`);
+      parts.push(`$${sessionUsage.cost.toFixed(5)}`);
     }
   } else {
     const cached = usage.prompt_tokens_details?.cached_tokens ?? 0;
@@ -6889,9 +6915,9 @@ function buildRightParts(usage, contextLimit, sessionUsage, gatewayMeta, cloudMo
     parts.push(`out ${usage.completion_tokens}`);
     parts.push(`ctx ${pct}%`);
     if (cloudMode) {
-      parts.push(`\x1B[9m${cost.total.toFixed(5)}\x1B[29m`);
+      parts.push(`\x1B[9m$${cost.total.toFixed(5)}\x1B[29m`);
     } else {
-      parts.push(`${cost.total.toFixed(5)}`);
+      parts.push(`$${cost.total.toFixed(5)}`);
     }
   }
   if (cloudMode && cloudBudget) {
@@ -6902,6 +6928,7 @@ function buildRightParts(usage, contextLimit, sessionUsage, gatewayMeta, cloudMo
   return parts;
 }
 function formatTokens(n) {
+  if (n >= 1e9) return `${(n / 1e9).toFixed(1)}B`;
   if (n >= 1e6) return `${(n / 1e6).toFixed(1)}M`;
   if (n >= 1e3) return `${(n / 1e3).toFixed(1)}K`;
   return String(n);
@@ -14047,6 +14074,17 @@ function App({
             const sid = ensureSessionId();
             void recordUsage(sid, u, gatewayUsageLookupFromConfig(cfg, meta ?? gatewayMetaRef.current));
             void getCostReport(sid).then((report) => setSessionUsage(report.session));
+            if (cfg?.cloudMode && (cloudToken ?? initialCloudToken)) {
+              const token = cloudToken ?? initialCloudToken;
+              const did = cloudDeviceId ?? initialCloudDeviceId;
+              void (async () => {
+                const { fetchCloudUsage: fetchCloudUsage2 } = await Promise.resolve().then(() => (init_auth(), auth_exports));
+                const usage2 = await fetchCloudUsage2(token, did);
+                if (usage2) {
+                  setCloudBudget({ remaining: usage2.remaining, limit: usage2.input_token_limit });
+                }
+              })();
+            }
           },
           onGatewayMeta: updateGatewayMeta,
           askPermission: (req) => new Promise((resolve2) => {
@@ -15126,6 +15164,17 @@ ${lines.join("\n")}` }]);
           const sid = ensureSessionId();
           void recordUsage(sid, u, gatewayUsageLookupFromConfig(cfg, meta ?? gatewayMetaRef.current));
           void getCostReport(sid).then((report) => setSessionUsage(report.session));
+          if (cfg?.cloudMode && (cloudToken ?? initialCloudToken)) {
+            const token = cloudToken ?? initialCloudToken;
+            const did = cloudDeviceId ?? initialCloudDeviceId;
+            void (async () => {
+              const { fetchCloudUsage: fetchCloudUsage2 } = await Promise.resolve().then(() => (init_auth(), auth_exports));
+              const usage2 = await fetchCloudUsage2(token, did);
+              if (usage2) {
+                setCloudBudget({ remaining: usage2.remaining, limit: usage2.input_token_limit });
+              }
+            })();
+          }
         },
         onGatewayMeta: updateGatewayMeta,
         onTasks: (nextTasks) => {