kimiflare 0.73.1 → 0.73.2

package/dist/index.js

@@ -1211,27 +1211,24 @@ async function* runKimi(opts2) {
   }
   const requestId = opts2.requestId ?? crypto.randomUUID();
   const { url, headers: gatewayHeaders } = buildKimiRequestTarget(opts2);
-  const isUniversalEndpoint = url.includes("/compat/chat/completions");
+  const isCloudEndpoint = url.startsWith("https://api.kimiflare.com");
   const entry = getModelOrInfer(opts2.model);
   const supportsTemperature = entry.supports.temperature !== false;
   const supportsReasoning = entry.supports.reasoning === true;
+  const compatModel = entry.provider === "workers-ai" ? `workers-ai/${opts2.model}` : opts2.model;
   const body = {
     messages: sanitizeMessagesForApi(opts2.messages),
     ...opts2.tools && opts2.tools.length ? { tools: opts2.tools, tool_choice: "auto", parallel_tool_calls: true } : {},
     stream: true,
     ...supportsTemperature ? { temperature: opts2.temperature ?? 0.2 } : {},
     max_completion_tokens: opts2.maxCompletionTokens ?? 16384,
-    // Universal Endpoint routes by the `model` body field (e.g. "anthropic/claude-sonnet-4-6").
-    // Workers AI endpoints route by URL path and ignore body.model.
-    ...isUniversalEndpoint ? { model: opts2.model } : {},
+    ...isCloudEndpoint ? {} : { model: compatModel },
     // OpenAI's streaming API omits `usage` by default — you have to explicitly
     // opt in via stream_options. Without this, the status bar's token /
-    // context-% / cost columns stay blank for every model that routes through
-    // the Universal Endpoint (verified empirically by curling /compat with and
-    // without this flag — see PR #468 discussion). CF docs don't mention
+    // context-% / cost columns stay blank. CF docs don't mention
     // stream_options but accept it transparently and forward it upstream;
     // providers that don't recognize the field ignore it.
-    ...isUniversalEndpoint ? { stream_options: { include_usage: true } } : {}
+    ...isCloudEndpoint ? {} : { stream_options: { include_usage: true } }
   };
   if (opts2.reasoningEffort && supportsReasoning) {
     body.reasoning_effort = opts2.reasoningEffort;
@@ -1381,25 +1378,24 @@ function gatewayHeadersFor(opts2) {
 function buildKimiRequestTarget(opts2) {
   validateModelId(opts2.model);
   if (opts2.cloudMode) {
-    const headers = opts2.cloudToken ? { Authorization: `Bearer ${opts2.cloudToken}` } : {};
-    if (opts2.cloudDeviceId) headers["X-Device-ID"] = opts2.cloudDeviceId;
-    return { url: "https://api.kimiflare.com/v1/chat", headers };
+    const headers2 = opts2.cloudToken ? { Authorization: `Bearer ${opts2.cloudToken}` } : {};
+    if (opts2.cloudDeviceId) headers2["X-Device-ID"] = opts2.cloudDeviceId;
+    return { url: "https://api.kimiflare.com/v1/chat", headers: headers2 };
+  }
+  if (!opts2.gateway?.id) {
+    throw new KimiApiError(
+      [
+        `kimiflare: ${opts2.model} routes through Cloudflare AI Gateway, but no gateway is configured.`,
+        ``,
+        `To fix: run  /gateway <your-gateway-id>  (create one at https://dash.cloudflare.com/?to=/:account/ai-gateway).`
+      ].join("\n"),
+      void 0,
+      400
+    );
   }
   const entry = getModelOrInfer(opts2.model);
+  const headers = gatewayHeadersFor(opts2);
   if (entry.provider !== "workers-ai") {
-    if (!opts2.gateway?.id) {
-      throw new KimiApiError(
-        [
-          `kimiflare: ${opts2.model} (${entry.provider}) is routed through Cloudflare AI Gateway, but no gateway is configured.`,
-          ``,
-          `To fix: run  /gateway <your-gateway-id>  (create one at https://dash.cloudflare.com/?to=/:account/ai-gateway),`,
-          `or switch back to a Workers AI model:  /model @cf/moonshotai/kimi-k2.6`
-        ].join("\n"),
-        void 0,
-        400
-      );
-    }
-    const headers = gatewayHeadersFor(opts2);
     const useUnified = !!opts2.unifiedBilling;
     const alias = opts2.providerKeyAliases?.[entry.provider];
     const providerKey = opts2.providerKeys?.[entry.provider];
@@ -1415,24 +1411,12 @@ function buildKimiRequestTarget(opts2) {
         401
       );
     }
-    return {
-      url: `https://gateway.ai.cloudflare.com/v1/${encodeURIComponent(opts2.accountId)}/${encodeURIComponent(
-        opts2.gateway.id
-      )}/compat/chat/completions`,
-      headers
-    };
-  }
-  if (!opts2.gateway?.id) {
-    return {
-      url: `https://api.cloudflare.com/client/v4/accounts/${encodeURIComponent(opts2.accountId)}/ai/run/${opts2.model}`,
-      headers: {}
-    };
   }
   return {
     url: `https://gateway.ai.cloudflare.com/v1/${encodeURIComponent(opts2.accountId)}/${encodeURIComponent(
       opts2.gateway.id
-    )}/workers-ai/${opts2.model}`,
-    headers: gatewayHeadersFor(opts2)
+    )}/compat/chat/completions`,
+    headers
   };
 }
 function readGatewayMeta(headers) {
@@ -3206,20 +3190,23 @@ async function fetchEmbeddings(opts2) {
     if (opts2.cloudToken) headers.Authorization = `Bearer ${opts2.cloudToken}`;
     if (opts2.cloudDeviceId) headers["X-Device-ID"] = opts2.cloudDeviceId;
   } else {
-    url = opts2.gateway ? `https://gateway.ai.cloudflare.com/v1/${opts2.accountId}/${opts2.gateway.id}/workers-ai/${model}` : `https://api.cloudflare.com/client/v4/accounts/${opts2.accountId}/ai/run/${model}`;
-    headers.Authorization = `Bearer ${opts2.apiToken}`;
-    if (opts2.gateway) {
-      const merged = {
-        ...opts2.gateway.metadata ?? {},
-        feature: "embedding"
-      };
-      const entries = Object.entries(merged).slice(0, 5);
-      headers["cf-aig-metadata"] = JSON.stringify(Object.fromEntries(entries));
+    if (!opts2.gateway) {
+      throw new Error(
+        "embeddings require an AI Gateway to be configured (run /gateway <id> or set aiGatewayId in config)"
+      );
     }
-    if (opts2.gateway?.cacheTtl !== void 0) {
+    url = `https://gateway.ai.cloudflare.com/v1/${opts2.accountId}/${opts2.gateway.id}/workers-ai/${model}`;
+    headers.Authorization = `Bearer ${opts2.apiToken}`;
+    const merged = {
+      ...opts2.gateway.metadata ?? {},
+      feature: "embedding"
+    };
+    const entries = Object.entries(merged).slice(0, 5);
+    headers["cf-aig-metadata"] = JSON.stringify(Object.fromEntries(entries));
+    if (opts2.gateway.cacheTtl !== void 0) {
       headers["cf-aig-cache-ttl"] = String(opts2.gateway.cacheTtl);
     }
-    if (opts2.gateway?.skipCache !== void 0) {
+    if (opts2.gateway.skipCache !== void 0) {
       headers["cf-aig-skip-cache"] = String(opts2.gateway.skipCache);
     }
   }