@kenkaiiii/gg-ai 5.81.0 → 5.82.0

package/dist/index.cjs

@@ -31,6 +31,7 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 var index_exports = {};
 __export(index_exports, {
   EventStream: () => EventStream,
+  FAST_MODE_SUFFIX: () => FAST_MODE_SUFFIX,
   GGAIError: () => GGAIError,
   ProviderError: () => ProviderError,
   REDACTION_MARKER: () => REDACTED,
@@ -63,6 +64,7 @@ __export(index_exports, {
   redactValue: () => redactValue,
   registerPalsuProvider: () => registerPalsuProvider,
   rerank: () => rerank,
+  resolveFastModeModel: () => resolveFastModeModel,
   resolveToolSchema: () => resolveToolSchema,
   sanitizeMessagesForWire: () => sanitizeMessagesForWire,
   setProviderDiagnostic: () => setProviderDiagnostic,
@@ -156,7 +158,8 @@ function isUsageLimitError(err) {
 }
 function isHardBillingMessage(message) {
   const lower = message.toLowerCase();
-  return lower.includes("insufficient balance") || lower.includes("insufficient credits") || lower.includes("more credits") || lower.includes("insufficient_quota") || lower.includes("exceeded your current quota") || lower.includes("quota exceeded") || lower.includes("no resource package") || lower.includes("recharge") || lower.includes("balance is too low") || lower.includes("out of credits") || lower.includes("arrears") || lower.includes("arrearage") || lower.includes("token quota") || lower.includes("exceeded_current_quota_error") || lower.includes("check your account balance") || lower.includes("does not yet include access") || lower.includes("subscription plan") || lower.includes("billing");
+  return lower.includes("insufficient balance") || lower.includes("insufficient credits") || lower.includes("more credits") || lower.includes("insufficient_quota") || lower.includes("exceeded your current quota") || lower.includes("quota exceeded") || lower.includes("no resource package") || lower.includes("recharge") || lower.includes("balance is too low") || lower.includes("out of credits") || // Anthropic fast mode on a subscription without usage credits turned on.
+  lower.includes("usage credits are required") || lower.includes("arrears") || lower.includes("arrearage") || lower.includes("token quota") || lower.includes("exceeded_current_quota_error") || lower.includes("check your account balance") || lower.includes("does not yet include access") || lower.includes("subscription plan") || lower.includes("billing");
 }
 function formatResetTime(resetsAt) {
   const when = new Date(resetsAt * 1e3);
@@ -1653,6 +1656,7 @@ function fineGrainedToolStreamingEnabled() {
   return v === "1" || v === "true" || v === "yes" || v === "on";
 }
 var SERVER_FALLBACK_BETA = "server-side-fallback-2026-07-01";
+var FAST_MODE_BETA = "fast-mode-2026-02-01";
 var anthropicServerFallback = {
   disabled: /* @__PURE__ */ new Set(),
   reset() {
@@ -1795,6 +1799,7 @@ async function* runStream(options) {
   const { system: rawSystem, messages } = toAnthropicMessages(downgradedMessages, cacheControl, {
     fallbackBlocks: useServerFallback
   });
+  const useFastMode = options.speed === "fast" && options.provider === "anthropic" && isDirectAnthropicApi(options.baseUrl);
   const system = isOAuth ? [
     {
       type: "text",
@@ -1865,6 +1870,7 @@ async function* runStream(options) {
       return contextEdits.length ? { context_management: { edits: contextEdits } } : {};
     })(),
     ...useServerFallback ? { fallbacks: "default" } : {},
+    ...useFastMode ? { speed: "fast" } : {},
     stream: useStreaming
   };
   const params = mergeProviderOptions(baseParams, options.providerOptions);
@@ -1883,7 +1889,8 @@ async function* runStream(options) {
     // so a pre-warmed cache expires before the user's first turn. cacheControl.ttl
     // is only "1h" on the first-party endpoint (see toAnthropicCacheControl).
     ...cacheControl?.ttl === "1h" ? ["extended-cache-ttl-2025-04-11"] : [],
-    ...useServerFallback ? [SERVER_FALLBACK_BETA] : []
+    ...useServerFallback ? [SERVER_FALLBACK_BETA] : [],
+    ...useFastMode ? [FAST_MODE_BETA] : []
   ];
   const toRequestOptions = (betas) => ({
     signal: options.signal ?? void 0,
@@ -2268,6 +2275,9 @@ function messageToResponse(message) {
     }
   };
 }
+function isFastModeCreditsMessage(message) {
+  return /usage credits are required for fast mode/i.test(message);
+}
 function readUnifiedRateLimit(headers) {
   const status = readHeader(headers, "anthropic-ratelimit-unified-status");
   const resetRaw = readHeader(
@@ -2291,6 +2301,14 @@ function toError(err) {
     const fallbackMessage = isRawJsonErrorEcho(err.message) ? emptyProviderErrorMessage(err.status) : err.message;
     const messageCandidate = bodyMessage ?? err.message;
     const message = isRawHtmlErrorEcho(messageCandidate) ? providerHtmlErrorMessage(err.status) : bodyType && bodyMessage ? `${bodyType}: ${bodyMessage}` : bodyMessage ?? fallbackMessage;
+    if (isFastModeCreditsMessage(message)) {
+      return new ProviderError("anthropic", message, {
+        statusCode: err.status,
+        ...requestId ? { requestId } : {},
+        hint: "Fast mode bills Claude usage credits, not your plan. Turn them on at claude.ai/settings/usage, or switch to Claude Opus 5.5 (standard speed) in the model selector.",
+        cause: err
+      });
+    }
     if (err.status === 429) {
       const limit = readUnifiedRateLimit(err.headers);
       const farOff = limit.resetsAt != null && limit.resetsAt * 1e3 - Date.now() > 6e4;
@@ -2546,8 +2564,9 @@ async function* runStream2(options) {
   if (options.provider === "sakana" && options.model === "fugu-ultra" && (options.thinking === "max" || options.thinking === "ultra")) {
     params.reasoning_effort = "max";
   }
-  if (options.provider === "openai" && options.serviceTier) {
-    params.service_tier = options.serviceTier;
+  const serviceTier = options.serviceTier ?? (options.speed === "fast" ? "priority" : void 0);
+  if (options.provider === "openai" && serviceTier) {
+    params.service_tier = serviceTier;
   }
   if (isKimiK3) {
     const paramsAny = params;
@@ -3057,6 +3076,9 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
   if (options.temperature != null && !options.thinking) {
     body.temperature = options.temperature;
   }
+  if (options.speed === "fast") {
+    body.service_tier = "priority";
+  }
   body.reasoning = {
     // GPT-5.6/6 require at least low; older models still support thinking off.
     // Apply the floor here for every caller, including one-off prompt rewrites.
@@ -4792,6 +4814,18 @@ function observePreparedContext(before, after, tools) {
   };
 }
 
+// src/fast-mode.ts
+var FAST_MODE_SUFFIX = "-fast";
+var FAST_MODE_BASE_MODELS = {
+  anthropic: ["claude-opus-5-5"],
+  openai: ["gpt-6-astra", "gpt-6.1-sol", "gpt-6-luna"]
+};
+function resolveFastModeModel(provider, model) {
+  if (!model.endsWith(FAST_MODE_SUFFIX)) return { model, fast: false };
+  const base = model.slice(0, -FAST_MODE_SUFFIX.length);
+  return FAST_MODE_BASE_MODELS[provider]?.includes(base) ? { model: base, fast: true } : { model, fast: false };
+}
+
 // src/stream.ts
 var OPENAI_EMBED_BATCH_SIZE = 2048;
 var OPENROUTER_EMBED_BATCH_SIZE = 64;
@@ -4966,7 +5000,9 @@ function requireLocalBaseUrl(baseUrl) {
   }
   return baseUrl;
 }
-function stream(options) {
+function stream(requested) {
+  const fastMode = resolveFastModeModel(requested.provider, requested.model);
+  const options = fastMode.fast ? { ...requested, model: fastMode.model, speed: "fast" } : requested;
   const entry = providerRegistry.get(options.provider);
   if (!entry) {
     throw new GGAIError(
@@ -5598,6 +5634,7 @@ function palsuRerankScore(query, document) {
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   EventStream,
+  FAST_MODE_SUFFIX,
   GGAIError,
   ProviderError,
   REDACTION_MARKER,
@@ -5630,6 +5667,7 @@ function palsuRerankScore(query, document) {
   redactValue,
   registerPalsuProvider,
   rerank,
+  resolveFastModeModel,
   resolveToolSchema,
   sanitizeMessagesForWire,
   setProviderDiagnostic,