@code-yeongyu/senpi-ai 2026.10.11 → 2026.10.12

package/dist/providers/xai.js

@@ -17,7 +17,13 @@ export function xaiProvider() {
                 load: loadXaiOAuth,
             }),
         },
-        models: Object.values(XAI_MODELS),
+        // xAI's Responses endpoint executes web_search in the same request.
+        // https://docs.x.ai/developers/tools/web-search
+        // Tool shape only: enable with supportsWebSearchPreview after verifying the route.
+        models: Object.values(XAI_MODELS).map((model) => ({
+            ...model,
+            compat: { ...model.compat, webSearchToolType: "web_search" },
+        })),
         api: openAIResponsesApi(),
     });
 }

package/dist/types.d.ts

@@ -19,6 +19,13 @@ import type { AssistantMessageEventStream } from "./utils/event-stream.ts";
 export type { Model } from "./model.ts";
 export type { SessionAffinityFormat } from "./openai-responses-compat.ts";
 export type OpenAIResponsesCompat = BaseOpenAIResponsesCompat & {
+    /**
+     * Exact native search type for a compatible Responses endpoint.
+     * OpenAI supports both names; xAI documents only web_search.
+     * https://developers.openai.com/api/docs/guides/tools-web-search
+     * https://docs.x.ai/developers/tools/web-search
+     */
+    webSearchToolType?: "web_search" | "web_search_preview";
     /** Whether the model supports message-anchored `additional_tools` input items. Default: false. */
     supportsAdditionalTools?: boolean;
     /** Whether the exact model accepts developer or system messages after the conversation has started. When false, later system messages are folded into the leading system message. Default: false; the generated model catalog enables it for verified models. */
@@ -524,6 +531,11 @@ export interface AssistantMessage {
     stopDetails?: AssistantStopDetails;
     deferred?: DeferredHandle;
     errorMessage?: string;
+    /** Redaction-safe cause retained when a session projects a provider limit as user-facing advice. */
+    errorCause?: {
+        type: "usage_limit";
+        scope: "account" | "model";
+    };
     /** Structured provider failure family, minted only by a provider adapter from structured error metadata. */
     providerDiagnostic?: ProviderDiagnostic;
     /** Explicit owner for an abort initiated by the provider retry watchdog. */
@@ -786,6 +798,16 @@ export type AssistantMessageEvent = {
  * Use this to override URL-based auto-detection for custom providers.
  */
 export interface OpenAICompletionsCompat {
+    /**
+     * In-turn web search support on this exact endpoint. OpenRouter and Moonshot
+     * have documented native defaults; Z.AI PAYG chat requires explicit opt-in
+     * because its coding-plan endpoint documents remote MCP search instead.
+     * https://openrouter.ai/docs/guides/features/server-tools/web-search
+     * https://platform.kimi.ai/docs/guide/use-web-search
+     * https://docs.z.ai/guides/tools/web-search
+     * https://docs.z.ai/devpack/mcp/search-mcp-server
+     */
+    supportsWebSearch?: boolean;
     /** Whether the provider supports the `store` field. Default: auto-detected from URL. */
     supportsStore?: boolean;
     /** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */

package/dist/utils/prompt-cache-ttl.d.ts

@@ -24,7 +24,8 @@ export type PromptCacheLifetime = {
 };
 export declare function isAnthropicApiBaseUrl(baseUrl: string): boolean;
 export declare function getAnthropicCompat(model: Model<"anthropic-messages">): Required<Omit<AnthropicMessagesCompat, "forceAdaptiveThinking" | "supportsMidConvoEffort" | "sessionAffinityFormat">> & Pick<AnthropicMessagesCompat, "sessionAffinityFormat">;
-export type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "vllmPriority" | "cacheControlFormat" | "toolCallFormat" | "deferredToolsMode" | "toolSchemaFlavor" | "supportsPromptCacheKey" | "chatTemplateArgs" | "supportsThinkingTokenBudget" | "thinkingTokenBudgetField" | "veniceParameters" | "supportsForcedToolChoice"> & {
+export type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "vllmPriority" | "cacheControlFormat" | "toolCallFormat" | "deferredToolsMode" | "toolSchemaFlavor" | "supportsPromptCacheKey" | "chatTemplateArgs" | "supportsThinkingTokenBudget" | "thinkingTokenBudgetField" | "veniceParameters" | "supportsForcedToolChoice" | "supportsWebSearch"> & {
+    supportsWebSearch?: OpenAICompletionsCompat["supportsWebSearch"];
     /** Declared forced tool_choice support; absent means supported. */
     supportsForcedToolChoice?: OpenAICompletionsCompat["supportsForcedToolChoice"];
     cacheControlFormat?: OpenAICompletionsCompat["cacheControlFormat"];

package/dist/utils/prompt-cache-ttl.js

@@ -188,6 +188,7 @@ export function getOpenAICompletionsCompat(model) {
     if (!model.compat)
         return detected;
     return {
+        supportsWebSearch: model.compat.supportsWebSearch,
         supportsStore: model.compat.supportsStore ?? detected.supportsStore,
         supportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,
         supportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,

package/dist/utils/provider-failure-description.d.ts

@@ -15,7 +15,9 @@ export declare function stripTurnRetrySuppressionPrefix(message: string): string
  * interruptions; returns undefined for failures that have no wording here so
  * the caller can fall back to the (marker-stripped) raw text.
  */
-export declare function describeProviderFailureForUser(errorMessage: string | undefined, options?: ProviderStallDescriptionOptions): string | undefined;
+export declare function describeProviderFailureForUser(errorMessage: string | undefined, options?: ProviderStallDescriptionOptions & {
+    usageLimit?: "account" | "model";
+}): string | undefined;
 /**
  * The turn-ending message for a provider module that can no longer be imported because the
  * installed package was replaced under the running process (#2358), or undefined for any other

package/dist/utils/provider-failure-description.js

@@ -44,6 +44,14 @@ export function describeProviderFailureForUser(errorMessage, options = {}) {
     const message = stripTurnRetrySuppressionPrefix(errorMessage).trim();
     if (!message)
         return undefined;
+    if (options.usageLimit === "account") {
+        const provider = options.model?.split("/")[0];
+        const reset = /\bresets?\s+([^"}\n]+)/i
+            .exec(message)?.[1]
+            ?.trim()
+            .replace(/[.;]+$/, "");
+        return `The ${provider ? `${provider} ` : ""}account usage limit has been reached${reset ? ` (resets ${reset})` : ""}. Switch to another provider with /model, or configure another provider in retry.fallbackChains.`;
+    }
     const stall = describeProviderStallForUser(message, options);
     if (stall !== undefined)
         return stall;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@code-yeongyu/senpi-ai",
-	"version": "2026.10.11",
+	"version": "2026.10.12",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -80,7 +80,7 @@
 		"check:model-data": "tsx scripts/check-model-data.ts",
 		"build": "tsgo -p tsconfig.build.json && shx chmod +x dist/cli.js && shx rm -rf dist/providers/data && shx cp -r src/providers/data dist/providers/data",
 		"build:offline": "npm run check:model-data && tsc -p tsconfig.build.json && shx chmod +x dist/cli.js && shx rm -rf dist/providers/data && shx cp -r src/providers/data dist/providers/data",
-		"test": "vitest --run",
+		"test": "node ../../scripts/run-tests.mjs vitest --run",
 		"prepublishOnly": "shx rm -rf dist && tsx scripts/generate-models.ts && tsc -p tsconfig.build.json && shx chmod +x dist/cli.js",
 		"dev": "tsc -p tsconfig.build.json --watch --preserveWatchOutput",
 		"dev:tsc": "tsc -p tsconfig.build.json --watch --preserveWatchOutput"
@@ -88,7 +88,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "0.127.0",
 		"@aws-sdk/client-bedrock-runtime": "3.1136.0",
-		"@earendil-works/pi-telemetry": "npm:@code-yeongyu/senpi-telemetry@2026.10.11",
+		"@earendil-works/pi-telemetry": "npm:@code-yeongyu/senpi-telemetry@2026.10.12",
 		"@google/genai": "2.23.0",
 		"@smithy/node-http-handler": "4.12.1",
 		"http-proxy-agent": "9.1.0",