aieo 0.1.34 → 0.1.35

package/dist/index.cjs

@@ -58,6 +58,7 @@ __export(index_exports, {
   getTokenPricing: () => getTokenPricing,
   hasApiKeyForProvider: () => hasApiKeyForProvider,
   loadModelPricing: () => loadModelPricing,
+  normalizeApiKey: () => normalizeApiKey,
   normalizeCallerBaseURL: () => normalizeCallerBaseURL,
   normalizeUsage: () => normalizeUsage,
   resolveLLMConfig: () => resolveLLMConfig,
@@ -193,7 +194,7 @@ var ConversationStorage = class {
   }
 };
 
-// ../../node_modules/@ai-sdk/provider/dist/index.mjs
+// ../../node_modules/@ai-sdk/anthropic/node_modules/@ai-sdk/provider/dist/index.mjs
 var marker = "vercel.ai.error";
 var symbol = Symbol.for(marker);
 var _a;
@@ -539,7 +540,7 @@ var UnsupportedFunctionalityError = class extends (_b14 = AISDKError, _a14 = sym
   }
 };
 
-// ../../node_modules/@ai-sdk/provider-utils/dist/index.mjs
+// ../../node_modules/@ai-sdk/anthropic/node_modules/@ai-sdk/provider-utils/dist/index.mjs
 var z4 = __toESM(require("zod/v4"), 1);
 var import_v3 = require("zod/v3");
 var import_v32 = require("zod/v3");
@@ -734,7 +735,7 @@ var EventSourceParserStream = class extends TransformStream {
   }
 };
 
-// ../../node_modules/@ai-sdk/provider-utils/dist/index.mjs
+// ../../node_modules/@ai-sdk/anthropic/node_modules/@ai-sdk/provider-utils/dist/index.mjs
 function combineHeaders(...headers) {
   return headers.reduce(
     (combinedHeaders, currentHeaders) => ({
@@ -23104,13 +23105,15 @@ function createOpenAI(options = {}) {
 var openai = createOpenAI();
 
 // src/provider.ts
+var import_xai = require("@ai-sdk/xai");
 var import_ai_sdk_provider = require("@openrouter/ai-sdk-provider");
 var LLM_GATEWAY_URL = process.env.LLM_GATEWAY_URL?.replace(/\/$/, "");
 var GATEWAY_PATHS = {
   anthropic: "/anthropic/v1",
   google: "/genai/v1beta",
   openai: "/openai/v1",
-  openrouter: "/openai/v1"
+  openrouter: "/openai/v1",
+  xai: "/openai/v1"
 };
 function getGatewayBaseURL(provider) {
   return LLM_GATEWAY_URL ? `${LLM_GATEWAY_URL}${GATEWAY_PATHS[provider]}` : void 0;
@@ -23141,7 +23144,8 @@ var PROVIDERS = [
   "anthropic",
   "google",
   "openai",
-  "openrouter"
+  "openrouter",
+  "xai"
 ];
 var MODELS = {
   anthropic: {
@@ -23157,19 +23161,24 @@ var MODELS = {
   },
   openrouter: {
     kimi: "moonshotai/kimi-k2.6"
+  },
+  xai: {
+    grok: "grok-4"
   }
 };
 var DEFAULT_MODELS = {
   anthropic: MODELS.anthropic.sonnet,
   google: MODELS.google.gemini,
   openai: MODELS.openai.gpt,
-  openrouter: MODELS.openrouter.kimi
+  openrouter: MODELS.openrouter.kimi,
+  xai: MODELS.xai.grok
 };
 var LIGHT_MODELS = {
   anthropic: "claude-haiku-4-5",
   google: "gemini-2.0-flash",
   openai: "gpt-4.1-mini",
-  openrouter: "moonshotai/kimi-k2.6"
+  openrouter: "moonshotai/kimi-k2.6",
+  xai: "grok-4-fast-non-reasoning"
 };
 function getLightModelForProvider(provider) {
   return LIGHT_MODELS[provider];
@@ -23204,6 +23213,8 @@ function getProviderForModel(modelName) {
       return "google";
     case "gpt":
       return "openai";
+    case "grok":
+      return "xai";
     // Full model IDs
     case "claude-sonnet-5":
     case "claude-opus-4-6":
@@ -23216,24 +23227,34 @@ function getProviderForModel(modelName) {
     case "gpt-4.1-mini":
       return "openai";
     default:
+      if (typeof modelName === "string" && modelName.toLowerCase().startsWith("grok")) {
+        return "xai";
+      }
       if (process.env.LLM_PROVIDER && PROVIDERS.includes(process.env.LLM_PROVIDER)) {
         return process.env.LLM_PROVIDER;
       }
       return "anthropic";
   }
 }
+function normalizeApiKey(key) {
+  if (typeof key !== "string") return void 0;
+  const trimmed = key.trim();
+  return trimmed.length > 0 ? trimmed : void 0;
+}
 function lookupApiKeyForProvider(provider) {
   switch (provider) {
     case "anthropic":
-      return process.env.ANTHROPIC_API_KEY;
+      return normalizeApiKey(process.env.ANTHROPIC_API_KEY);
     case "google":
-      return process.env.GOOGLE_API_KEY;
+      return normalizeApiKey(process.env.GOOGLE_API_KEY);
     case "openai":
-      return process.env.OPENAI_API_KEY;
+      return normalizeApiKey(process.env.OPENAI_API_KEY);
     case "openrouter":
-      return process.env.OPENROUTER_API_KEY;
+      return normalizeApiKey(process.env.OPENROUTER_API_KEY);
+    case "xai":
+      return normalizeApiKey(process.env.XAI_API_KEY);
     case "claude_code":
-      return process.env.CLAUDE_CODE_API_KEY;
+      return normalizeApiKey(process.env.CLAUDE_CODE_API_KEY);
     default:
       return void 0;
   }
@@ -23248,6 +23269,73 @@ function getApiKeyForProvider(provider) {
   }
   return apiKey;
 }
+var DEFAULT_LLM_HTTP_TIMEOUT_MS = parseInt(process.env.LLM_HTTP_TIMEOUT_MS || "", 10) || 3e5;
+var LLM_HTTP_TIMEOUT_RETRIES = (() => {
+  const parsed = parseInt(process.env.LLM_HTTP_TIMEOUT_RETRIES || "", 10);
+  return Number.isFinite(parsed) && parsed >= 0 ? parsed : 2;
+})();
+function combineSignals(signals) {
+  const live = signals.filter((s) => !!s);
+  if (live.length === 0) return void 0;
+  if (live.length === 1) return live[0];
+  if (typeof AbortSignal.any === "function") {
+    return AbortSignal.any(live);
+  }
+  const ctrl = new AbortController();
+  if (live.some((s) => s.aborted)) {
+    ctrl.abort();
+    return ctrl.signal;
+  }
+  const abort = () => {
+    try {
+      ctrl.abort();
+    } catch (_) {
+    }
+  };
+  for (const s of live) s.addEventListener("abort", abort, { once: true });
+  return ctrl.signal;
+}
+function buildTimeoutFetch(timeoutMs, runSignal) {
+  return async (input, init) => {
+    const url = typeof input === "string" ? input : input.url;
+    const sdkSignal = init?.signal;
+    const canResend = (typeof input === "string" || input instanceof URL) && (init?.body == null || typeof init.body === "string");
+    const maxAttempts = canResend ? 1 + LLM_HTTP_TIMEOUT_RETRIES : 1;
+    for (let attempt = 1; ; attempt++) {
+      const startMs = Date.now();
+      const timeoutCtrl = new AbortController();
+      let timer = setTimeout(() => {
+        timeoutCtrl.abort();
+      }, timeoutMs);
+      const clearTimer = () => {
+        if (timer !== void 0) {
+          clearTimeout(timer);
+          timer = void 0;
+        }
+      };
+      const combinedSignal = combineSignals([
+        sdkSignal,
+        runSignal,
+        timeoutCtrl.signal
+      ]);
+      try {
+        const response = await fetch(input, { ...init, signal: combinedSignal });
+        clearTimer();
+        return response;
+      } catch (err) {
+        clearTimer();
+        const elapsedMs = Date.now() - startMs;
+        const isTimeout = timeoutCtrl.signal.aborted && !runSignal?.aborted && !sdkSignal?.aborted;
+        const reason = isTimeout ? "timeout" : "aborted";
+        console.warn(
+          `[llm-fetch] ${reason} after ${elapsedMs}ms (attempt ${attempt}/${maxAttempts}): ${url} \u2014 ${err?.message ?? String(err)}`
+        );
+        if (isTimeout && attempt < maxAttempts) continue;
+        throw err;
+      }
+    }
+  };
+}
 function getModelForProvider(provider, modelName) {
   const model = MODELS[provider][modelName];
   if (!model) {
@@ -23257,22 +23345,27 @@ function getModelForProvider(provider, modelName) {
   }
   return model;
 }
-function getModelDetails(modelName, apiKeyIn, baseUrl, headers) {
+function getModelDetails(modelName, apiKeyIn, baseUrl, headers, abortSignal, timeoutMs) {
   const provider = getProviderForModel(modelName);
-  const apiKey = apiKeyIn || getApiKeyForProvider(provider);
+  const callerKey = normalizeApiKey(apiKeyIn);
+  const apiKey = callerKey || getApiKeyForProvider(provider);
   console.log("===> getModelDetails", {
     provider,
     modelName: modelName || "(default)",
-    keySource: apiKeyIn ? "request body" : "env var",
+    keySource: callerKey ? "request body" : "env var",
     apiKeyPrefix: apiKey ? apiKey.slice(0, 12) + "..." : "(missing)",
     baseUrl: baseUrl || "(default)",
-    headerKeys: headers ? Object.keys(headers) : []
+    headerKeys: headers ? Object.keys(headers) : [],
+    hasAbortSignal: !!abortSignal,
+    timeoutMs: timeoutMs ?? DEFAULT_LLM_HTTP_TIMEOUT_MS
   });
   const model = getModel(provider, {
     modelName,
     apiKey,
     baseUrl,
-    headers
+    headers,
+    abortSignal,
+    timeoutMs
   });
   let modelId;
   if (modelName && modelName.includes("/")) {
@@ -23295,7 +23388,7 @@ function getModel(provider, opts) {
   if (typeof opts === "string") {
     opts = { apiKey: opts };
   }
-  const apiKey = opts?.apiKey || getApiKeyForProvider(provider);
+  const apiKey = normalizeApiKey(opts?.apiKey) || getApiKeyForProvider(provider);
   let modelId;
   if (opts?.modelName && opts.modelName.includes("/")) {
     const parts = opts.modelName.split("/");
@@ -23311,7 +23404,8 @@ function getModel(provider, opts) {
       "haiku",
       "gemini",
       "gpt",
-      "kimi"
+      "kimi",
+      "grok"
     ];
     if (knownShortcuts.includes(opts.modelName)) {
       modelId = getModelForProvider(provider, opts.modelName);
@@ -23335,12 +23429,17 @@ function getModel(provider, opts) {
   if (extraHeaders) {
     console.log(`[headers] attaching ${Object.keys(extraHeaders).length} custom header(s) to ${provider} client`);
   }
+  const timeoutFetch = buildTimeoutFetch(
+    opts?.timeoutMs ?? DEFAULT_LLM_HTTP_TIMEOUT_MS,
+    opts?.abortSignal
+  );
   switch (provider) {
     case "anthropic":
       const anthropic2 = createAnthropic({
         apiKey,
         ...baseURL && { baseURL },
-        ...extraHeaders && { headers: extraHeaders }
+        ...extraHeaders && { headers: extraHeaders },
+        fetch: timeoutFetch
       });
       return anthropic2(modelId);
     case "google": {
@@ -23353,13 +23452,15 @@ function getModel(provider, opts) {
         const googleCompat = createOpenAI({
           apiKey,
           baseURL: openaiCompatBase,
-          ...extraHeaders && { headers: extraHeaders }
+          ...extraHeaders && { headers: extraHeaders },
+          fetch: timeoutFetch
         });
         return googleCompat.chat(`gemini/${modelId}`);
       }
       const google2 = createGoogleGenerativeAI({
         apiKey,
-        ...extraHeaders && { headers: extraHeaders }
+        ...extraHeaders && { headers: extraHeaders },
+        fetch: timeoutFetch
       });
       return google2(modelId);
     }
@@ -23367,14 +23468,16 @@ function getModel(provider, opts) {
       const openai2 = createOpenAI({
         apiKey,
         ...baseURL && { baseURL },
-        ...extraHeaders && { headers: extraHeaders }
+        ...extraHeaders && { headers: extraHeaders },
+        fetch: timeoutFetch
       });
       return openai2(modelId);
     case "openrouter": {
       const openrouter = (0, import_ai_sdk_provider.createOpenRouter)({
         apiKey,
         ...baseURL && { baseURL },
-        ...extraHeaders && { headers: extraHeaders }
+        ...extraHeaders && { headers: extraHeaders },
+        fetch: timeoutFetch
       });
       const isMoonshot = modelId.startsWith("moonshotai/") || modelId.toLowerCase().includes("kimi");
       const settings = {
@@ -23383,6 +23486,23 @@ function getModel(provider, opts) {
       };
       return openrouter(modelId, settings);
     }
+    case "xai": {
+      if (baseURL) {
+        const xaiCompat = createOpenAI({
+          apiKey,
+          baseURL,
+          ...extraHeaders && { headers: extraHeaders },
+          fetch: timeoutFetch
+        });
+        return xaiCompat.chat(`xai/${modelId}`);
+      }
+      const xai = (0, import_xai.createXai)({
+        apiKey,
+        ...extraHeaders && { headers: extraHeaders },
+        fetch: timeoutFetch
+      });
+      return xai(modelId);
+    }
     // case "claude_code":
     //   try {
     //     const customProvider = createClaudeCode({
@@ -23417,22 +23537,46 @@ var MODEL_CONTEXT_LIMITS = {
   "gemini-3-pro-preview": 1e6,
   "gemini-2.0-flash": 1e6,
   // OpenAI
-  "gpt-5": 128e3,
+  "gpt-5": 4e5,
+  "gpt-5.5": 105e4,
+  "gpt-5.6-terra": 105e4,
+  "gpt-5.6-luna": 105e4,
+  "gpt-5.6-sol": 105e4,
   "gpt-4.1-mini": 1e6,
   // OpenRouter — values from the OpenRouter model catalog
   // (https://openrouter.ai/api/v1/models, context_length).
+  "stealth/ox-alpha": 1048576,
+  "openai/gpt-5": 4e5,
+  "openai/gpt-5.5": 105e4,
+  "openai/gpt-5.6-terra": 105e4,
+  "openai/gpt-5.6-luna": 105e4,
+  "openai/gpt-5.6-sol": 105e4,
   "moonshotai/kimi-k3": 1048576,
   "moonshotai/kimi-k2.7-code": 262144,
   "moonshotai/kimi-k2.6": 262144,
   "moonshotai/kimi-k2.5": 262144,
   "moonshotai/kimi-k2-thinking": 262144,
-  "moonshotai/kimi-k2": 131072
+  "moonshotai/kimi-k2": 131072,
+  // xAI (and the same models via OpenRouter as x-ai/...). Getting these
+  // right matters beyond display: contextLimit drives truncateOldToolResults,
+  // and premature truncation rewrites old messages — which invalidates Grok's
+  // automatic prefix cache on every subsequent step.
+  "grok-4": 256e3,
+  "grok-4-fast": 2e6,
+  "grok-4-fast-reasoning": 2e6,
+  "grok-4-fast-non-reasoning": 2e6,
+  "grok-code-fast-1": 256e3,
+  "x-ai/grok-4": 256e3,
+  "x-ai/grok-4-fast": 2e6,
+  "x-ai/grok-code-fast-1": 256e3
 };
 var DEFAULT_CONTEXT_LIMITS = {
   anthropic: 1e6,
   google: 1e6,
-  openai: 128e3,
-  openrouter: 128e3
+  // gpt-5.x models are all 400k+; only legacy chat models are smaller
+  openai: 4e5,
+  openrouter: 256e3,
+  xai: 256e3
 };
 var FALLBACK_CONTEXT_LIMIT = 128e3;
 function getContextLimit(modelId, provider) {
@@ -23456,6 +23600,15 @@ var TOKEN_PRICING = {
   openrouter: {
     inputTokenPrice: 0.6,
     outputTokenPrice: 3
+  },
+  // grok-4 rates. cacheReadPrice matters: computeSessionCost bills cache
+  // reads at full input price when it's absent, and Grok discounts cached
+  // prompt tokens 4x. (Grok has no cache-write charge — writes are billed
+  // as ordinary input, so no cacheWritePrice here.)
+  xai: {
+    inputTokenPrice: 3,
+    outputTokenPrice: 15,
+    cacheReadPrice: 0.75
   }
 };
 function getTokenPricing(provider) {
@@ -23542,6 +23695,10 @@ function getProviderOptions(provider, thinkingSpeed, modelName) {
       return {
         openrouter: { usage: { include: true } }
       };
+    case "xai":
+      return {
+        xai: {}
+      };
     default:
       throw new Error(`Unsupported provider: ${provider}`);
   }
@@ -23554,10 +23711,11 @@ function resolveLLMConfig(opts) {
     providerHint = void 0;
   }
   const provider = modelName ? getProviderForModel(modelName) : providerHint || process.env.LLM_PROVIDER || getProviderForModel();
+  const callerKey = normalizeApiKey(opts?.apiKey);
+  const apiKey = callerKey || getApiKeyForProvider(provider);
   console.log(
-    `[resolveLLMConfig] provider=${provider} modelName=${modelName || "(default)"} light=${!!opts?.light}`
+    `[resolveLLMConfig] provider=${provider} modelName=${modelName || "(default)"} light=${!!opts?.light} keySource=${callerKey ? "request body" : "env var"} apiKeyPrefix=${apiKey.slice(0, 8)}...`
   );
-  const apiKey = opts?.apiKey || getApiKeyForProvider(provider);
   let effectiveModelName = modelName;
   if (!effectiveModelName && opts?.light) {
     effectiveModelName = getLightModelForProvider(provider);
@@ -23596,7 +23754,7 @@ function withLegacyUsage(usage) {
 }
 function normalizeUsage(raw) {
   const inputTokens = tokenCount(raw?.inputTokens);
-  const cache_read = raw?.cache_read !== void 0 ? tokenCount(raw.cache_read) : tokenCount(raw?.inputTokenDetails?.cacheReadTokens);
+  const cache_read = raw?.cache_read !== void 0 ? tokenCount(raw.cache_read) : tokenCount(raw?.inputTokenDetails?.cacheReadTokens) || tokenCount(raw?.cachedInputTokens);
   const cache_write = raw?.cache_write !== void 0 ? tokenCount(raw.cache_write) : tokenCount(raw?.inputTokenDetails?.cacheWriteTokens);
   const input = raw?.input !== void 0 ? tokenCount(raw.input) : raw?.inputTokenDetails?.noCacheTokens !== void 0 ? tokenCount(raw.inputTokenDetails.noCacheTokens) : Math.max(inputTokens - cache_read - cache_write, 0);
   const output = raw?.output !== void 0 ? tokenCount(raw.output) : tokenCount(raw?.outputTokens);
@@ -23817,6 +23975,7 @@ function consoleLogger(name19) {
   getTokenPricing,
   hasApiKeyForProvider,
   loadModelPricing,
+  normalizeApiKey,
   normalizeCallerBaseURL,
   normalizeUsage,
   resolveLLMConfig,