@code-yeongyu/senpi-ai 2026.8.18 → 2026.8.19

package/README.md

@@ -801,7 +801,7 @@ Many models support thinking/reasoning capabilities where they can show their in
 const model = models.getModel('anthropic', 'claude-sonnet-4-5')!;
 // or models.getModel('openai', 'gpt-5-mini');
 // or models.getModel('google', 'gemini-2.5-flash');
-// or models.getModel('xai', 'grok-4.5');
+// or models.getModel('xai', 'grok-4.6');
 
 // Check if model supports reasoning
 if (model.reasoning) {
@@ -1191,8 +1191,10 @@ interface OpenAICompletionsCompat {
   requiresThinkingAsText?: boolean;  // Whether thinking blocks must be converted to text (default: false)
   requiresReasoningContentOnAssistantMessages?: boolean; // Whether all replayed assistant messages must include empty reasoning_content when reasoning is enabled (default: auto-detected for DeepSeek)
   thinkingFormat?: 'openai' | 'openrouter' | 'deepseek' | 'together' | 'baseten' | 'zai' | 'qwen' | 'chat-template' | 'qwen-chat-template' | 'string-thinking' | 'ant-ling'; // Format for reasoning param: 'openai' uses reasoning_effort, 'openrouter' uses reasoning: { effort }, 'deepseek' uses thinking: { type } plus reasoning_effort when supported, 'together' uses reasoning: { enabled } plus reasoning_effort when supported, 'baseten' uses configurable chat_template_args plus reasoning_effort when supported, 'zai' uses thinking: { type }, 'qwen' uses enable_thinking, 'chat-template' uses configurable chat_template_kwargs, 'qwen-chat-template' uses chat_template_kwargs.enable_thinking and preserve_thinking, 'string-thinking' uses top-level thinking, 'ant-ling' uses reasoning: { effort } only for mapped efforts (default: openai)
-  chatTemplateKwargs?: Record<string, string | number | boolean | null | { '$var': 'thinking.enabled' | 'thinking.effort'; omitWhenOff?: boolean }>; // chat_template_kwargs values; use $var for pi-controlled thinking values
-  chatTemplateArgs?: Record<string, string | number | boolean | null | { '$var': 'thinking.enabled' | 'thinking.effort'; omitWhenOff?: boolean }>; // chat_template_args values for thinkingFormat: 'baseten'; use $var for pi-controlled thinking values
+  chatTemplateKwargs?: Record<string, string | number | boolean | null | { '$var': 'thinking.enabled' | 'thinking.effort' | 'thinking.budget'; omitWhenOff?: boolean }>; // chat_template_kwargs values; use $var for pi-controlled thinking values
+  chatTemplateArgs?: Record<string, string | number | boolean | null | { '$var': 'thinking.enabled' | 'thinking.effort' | 'thinking.budget'; omitWhenOff?: boolean }>; // chat_template_args values for thinkingFormat: 'baseten'; use $var for pi-controlled thinking values
+  thinkingTokenBudgetField?: 'thinking_token_budget' | 'thinking_budget' | 'thinking_budget_tokens'; // Top-level field that caps reasoning tokens from thinkingBudgets (vLLM / Qwen / llama.cpp). Off by default.
+  supportsThinkingTokenBudget?: boolean; // Alias for thinkingTokenBudgetField: 'thinking_token_budget' (vLLM). Prefer thinkingTokenBudgetField. Default: false.
   cacheControlFormat?: 'anthropic';  // Anthropic-style cache_control on system prompt, last tool, and last user/assistant text content
   openRouterRouting?: OpenRouterRouting; // OpenRouter routing preferences (default: {})
   vercelGatewayRouting?: VercelGatewayRouting; // Vercel AI Gateway routing preferences (default: {})

package/dist/api/anthropic-messages.d.ts

@@ -1,6 +1,6 @@
 import Anthropic from "@anthropic-ai/sdk";
 import type { MessageCreateParamsNonStreaming } from "@anthropic-ai/sdk/resources/messages.js";
-import type { Context, Model, SimpleStreamOptions, StreamFunction, StreamOptions } from "../types.ts";
+import type { AnthropicRefusalFallback, Context, Model, SimpleStreamOptions, StreamFunction, StreamOptions } from "../types.ts";
 export { getAnthropicCompat } from "../utils/prompt-cache-ttl.ts";
 export type AnthropicEffort = "low" | "medium" | "high" | "xhigh" | "max";
 export type AnthropicThinkingDisplay = "summarized" | "omitted";
@@ -52,6 +52,12 @@ export interface AnthropicOptions extends StreamOptions {
      * Default: true.
      */
     interleavedThinking?: boolean;
+    /**
+     * Anthropic refusal fallback. When set, the request includes the server-side
+     * fallback beta and Anthropic retries eligible refusals on the configured
+     * fallback target before returning a final response.
+     */
+    refusalFallbacks?: AnthropicRefusalFallback;
     /**
      * Anthropic tool choice behavior. String values map to Anthropic's built-in
      * choices; `{ type: "tool", name }` forces a specific tool.

package/dist/api/anthropic-messages.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"anthropic-messages.d.ts","sourceRoot":"","sources":["../../src/api/anthropic-messages.ts"],"names":[],"mappings":"AAAA,OAAO,SAAS,MAAM,mBAAmB,CAAC;AAC1C,OAAO,KAAK,EAGX,+BAA+B,EAK/B,MAAM,yCAAyC,CAAC;AAEjD,OAAO,KAAK,EAKX,OAAO,EAGP,KAAK,EAGL,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAOb,MAAM,aAAa,CAAC;AAmCrB,OAAO,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AA6JlE,MAAM,MAAM,eAAe,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;AAE1E,MAAM,MAAM,wBAAwB,GAAG,YAAY,GAAG,SAAS,CAAC;AAoEhE,MAAM,WAAW,gBAAiB,SAAQ,aAAa;IACtD;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B;;;;OAIG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;;;;;;;;OAWG;IACH,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB;;;;;;;;;;;OAWG;IACH,eAAe,CAAC,EAAE,wBAAwB,CAAC;IAC3C;;;;;OAKG;IACH,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IACtE;;;;OAIG;IACH,MAAM,CAAC,EAAE,SAAS,CAAC;CACnB;AA+1BD,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,oBAAoB,EAAE,gBAAgB,CA2ZzE,CAAC;AA0FF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,oBAAoB,EAAE,mBAAmB,CAyClF,CAAC;AAyIF,wBAAgB,mCAAmC,CAClD,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,EAClC,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,gBAAgB,GACxB,+BAA+B,CAkBjC"}
+{"version":3,"file":"anthropic-messages.d.ts","sourceRoot":"","sources":["../../src/api/anthropic-messages.ts"],"names":[],"mappings":"AAAA,OAAO,SAAS,MAAM,mBAAmB,CAAC;AAC1C,OAAO,KAAK,EAGX,+BAA+B,EAK/B,MAAM,yCAAyC,CAAC;AAEjD,OAAO,KAAK,EACX,wBAAwB,EAKxB,OAAO,EAGP,KAAK,EAGL,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAOb,MAAM,aAAa,CAAC;AAmCrB,OAAO,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AA6JlE,MAAM,MAAM,eAAe,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;AAE1E,MAAM,MAAM,wBAAwB,GAAG,YAAY,GAAG,SAAS,CAAC;AAyEhE,MAAM,WAAW,gBAAiB,SAAQ,aAAa;IACtD;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B;;;;OAIG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;;;;;;;;OAWG;IACH,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB;;;;;;;;;;;OAWG;IACH,eAAe,CAAC,EAAE,wBAAwB,CAAC;IAC3C;;;;;OAKG;IACH,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;;OAIG;IACH,gBAAgB,CAAC,EAAE,wBAAwB,CAAC;IAC5C;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IACtE;;;;OAIG;IACH,MAAM,CAAC,EAAE,SAAS,CAAC;CACnB;AA+1BD,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,oBAAoB,EAAE,gBAAgB,CA6ZzE,CAAC;AA0FF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,oBAAoB,EAAE,mBAAmB,CAkDlF,CAAC;AA6IF,wBAAgB,mCAAmC,CAClD,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,EAClC,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,gBAAgB,GACxB,+BAA+B,CAkBjC"}

package/dist/api/anthropic-messages.js

@@ -175,6 +175,7 @@ const UNSUPPORTED_NATIVE_COMPUTER_TOOL_MODEL_MARKERS = [
     "opus-4-8",
     "opus-4.8",
 ];
+const SERVER_SIDE_FALLBACK_BETA = "server-side-fallback-2026-07-01";
 // A provider can reject its own empty signatures. Learn that capability for one
 // conversation without changing the shared model definition used by other sessions.
 const unsignedThinkingTextReplayFallbacks = new Set();
@@ -941,7 +942,7 @@ export const stream = (model, context, options) => {
                 }
                 const cacheRetention = resolveCacheRetention(options?.cacheRetention ?? model.cacheRetention, options?.env, "short");
                 const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
-                const created = createClient(model, apiKey, options?.interleavedThinking ?? true, shouldUseFineGrainedToolStreamingBeta(model, context), optionsHeaders, options?.fetch, copilotDynamicHeaders, cacheSessionId, options?.env);
+                const created = createClient(model, apiKey, options?.interleavedThinking ?? true, shouldUseFineGrainedToolStreamingBeta(model, context), options?.refusalFallbacks !== undefined, optionsHeaders, options?.fetch, copilotDynamicHeaders, cacheSessionId, options?.env);
                 client = created.client;
                 isOAuth = created.isOAuthToken;
             }
@@ -1005,6 +1006,7 @@ export const stream = (model, context, options) => {
             for await (const event of iterateAnthropicEvents(response, requestSignal)) {
                 if (event.type === "message_start") {
                     output.responseId = event.message.id;
+                    output.model = event.message.model;
                     // Capture initial token usage from message_start event
                     // This ensures we have input token counts even if the stream is aborted early
                     output.usage.input = event.message.usage.input_tokens || 0;
@@ -1357,9 +1359,16 @@ function mapThinkingLevelToEffort(model, level) {
 export const streamSimple = (model, context, options) => {
     const apiKey = options?.apiKey;
     assertRequestAuth(model.provider, apiKey, providerHeadersToRecord(options?.headers));
-    const base = buildBaseOptions(model, context, options, options?.apiKey);
+    const base = {
+        ...buildBaseOptions(model, context, options, options?.apiKey),
+        toolChoice: options?.toolChoice,
+    };
     if (!options?.reasoning) {
-        return stream(model, context, { ...base, thinkingEnabled: false });
+        return stream(model, context, {
+            ...base,
+            refusalFallbacks: options?.refusalFallbacks,
+            thinkingEnabled: false,
+        });
     }
     // For models with adaptive thinking: use an effort level.
     // For older models: use budget-based thinking.
@@ -1367,6 +1376,7 @@ export const streamSimple = (model, context, options) => {
         const effort = mapThinkingLevelToEffort(model, options.reasoning);
         return stream(model, context, {
             ...base,
+            refusalFallbacks: options?.refusalFallbacks,
             thinkingEnabled: true,
             effort,
         });
@@ -1377,6 +1387,7 @@ export const streamSimple = (model, context, options) => {
     const maxTokens = clampMaxTokensToContext(model, context, adjusted.maxTokens);
     return stream(model, context, {
         ...base,
+        refusalFallbacks: options?.refusalFallbacks,
         maxTokens,
         thinkingEnabled: true,
         thinkingBudgetTokens: Math.min(adjusted.thinkingBudget, Math.max(0, maxTokens - 1024)),
@@ -1385,7 +1396,7 @@ export const streamSimple = (model, context, options) => {
 function isOAuthToken(apiKey) {
     return apiKey.includes("sk-ant-oat");
 }
-function createClient(model, apiKey, interleavedThinking, useFineGrainedToolStreamingBeta, optionsHeaders, fetch, dynamicHeaders, sessionId, env) {
+function createClient(model, apiKey, interleavedThinking, useFineGrainedToolStreamingBeta, useServerSideFallbackBeta, optionsHeaders, fetch, dynamicHeaders, sessionId, env) {
     // Adaptive thinking models have interleaved thinking built in, so skip the beta header.
     const needsInterleavedBeta = interleavedThinking && !supportsAdaptiveThinking(model);
     const betaFeatures = [];
@@ -1395,6 +1406,9 @@ function createClient(model, apiKey, interleavedThinking, useFineGrainedToolStre
     if (needsInterleavedBeta) {
         betaFeatures.push(INTERLEAVED_THINKING_BETA);
     }
+    if (useServerSideFallbackBeta) {
+        betaFeatures.push(SERVER_SIDE_FALLBACK_BETA);
+    }
     if (model.provider === "cloudflare-ai-gateway") {
         const client = new Anthropic({
             apiKey: null,
@@ -1599,6 +1613,9 @@ function buildParams(model, context, isOAuthToken, options, unsignedThinkingRepl
         }
     }
     applyExtraBodyToAnthropicParams(params, options?.extraBody);
+    if (options?.refusalFallbacks !== undefined) {
+        params.fallbacks = options.refusalFallbacks;
+    }
     return params;
 }
 function applyExtraBodyToAnthropicParams(params, extraBody) {