genai-lite 0.8.2 → 0.9.0

package/dist/index.js

@@ -14,10 +14,14 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
     for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.silentLogger = exports.DEFAULT_LOG_LEVEL = exports.createDefaultLogger = exports.KNOWN_GGUF_MODELS = exports.detectGgufCapabilities = exports.createFallbackModelInfo = exports.parseTemplateWithMetadata = exports.extractInitialTaggedContent = exports.parseRoleTags = exports.parseStructuredContent = exports.extractRandomVariables = exports.getSmartPreview = exports.countTokens = exports.renderTemplate = exports.ImageService = exports.MistralClientAdapter = exports.OpenRouterClientAdapter = exports.LlamaCppServerClient = exports.LlamaCppClientAdapter = exports.fromEnvironment = exports.LLMService = void 0;
+exports.silentLogger = exports.DEFAULT_LOG_LEVEL = exports.createDefaultLogger = exports.KNOWN_GGUF_MODELS = exports.detectGgufCapabilities = exports.createFallbackModelInfo = exports.parseTemplateWithMetadata = exports.extractMarkerDelimitedContent = exports.extractInitialTaggedContent = exports.parseRoleTags = exports.parseStructuredContent = exports.extractRandomVariables = exports.getSmartPreview = exports.countTokens = exports.renderTemplate = exports.ImageService = exports.MistralClientAdapter = exports.OpenRouterClientAdapter = exports.LlamaCppServerClient = exports.LlamaCppClientAdapter = exports.fromEnvironment = exports.DEFAULT_RETRY_POLICY = exports.withRetry = exports.LLMService = void 0;
 // --- LLM Service ---
 var LLMService_1 = require("./llm/LLMService");
 Object.defineProperty(exports, "LLMService", { enumerable: true, get: function () { return LLMService_1.LLMService; } });
+// --- Retry Utilities ---
+var withRetry_1 = require("./shared/services/withRetry");
+Object.defineProperty(exports, "withRetry", { enumerable: true, get: function () { return withRetry_1.withRetry; } });
+Object.defineProperty(exports, "DEFAULT_RETRY_POLICY", { enumerable: true, get: function () { return withRetry_1.DEFAULT_RETRY_POLICY; } });
 // Export all core request/response/config types from the LLM module
 __exportStar(require("./llm/types"), exports);
 // Export all client adapter types
@@ -51,6 +55,7 @@ var parser_1 = require("./prompting/parser");
 Object.defineProperty(exports, "parseStructuredContent", { enumerable: true, get: function () { return parser_1.parseStructuredContent; } });
 Object.defineProperty(exports, "parseRoleTags", { enumerable: true, get: function () { return parser_1.parseRoleTags; } });
 Object.defineProperty(exports, "extractInitialTaggedContent", { enumerable: true, get: function () { return parser_1.extractInitialTaggedContent; } });
+Object.defineProperty(exports, "extractMarkerDelimitedContent", { enumerable: true, get: function () { return parser_1.extractMarkerDelimitedContent; } });
 Object.defineProperty(exports, "parseTemplateWithMetadata", { enumerable: true, get: function () { return parser_1.parseTemplateWithMetadata; } });
 var config_1 = require("./llm/config");
 Object.defineProperty(exports, "createFallbackModelInfo", { enumerable: true, get: function () { return config_1.createFallbackModelInfo; } });

package/dist/llm/LLMService.d.ts

@@ -2,6 +2,7 @@ import type { ApiKeyProvider, PresetMode } from '../types';
 import type { Logger, LogLevel } from '../logging/types';
 import type { LLMChatRequest, LLMChatRequestWithPreset, LLMResponse, LLMFailureResponse, ProviderInfo, ModelInfo, ApiProviderId, LLMSettings, ModelContext, LLMMessage } from "./types";
 import type { ModelPreset } from "../types/presets";
+import { type RetryPolicy } from "../shared/services/withRetry";
 export type { PresetMode };
 /**
  * Options for configuring the LLMService
@@ -15,6 +16,32 @@ export interface LLMServiceOptions {
     logLevel?: LogLevel;
     /** Custom logger implementation. If provided, logLevel is ignored. */
     logger?: Logger;
+    /**
+     * Retry policy for transient failures (rate limits, 5xx, network errors, timeouts).
+     * Defaults: maxRetries 2, initialDelayMs 500, maxDelayMs 10000, backoffFactor 2,
+     * retryOnTimeout true. Provider SDK-internal retries are disabled — this layer
+     * is the single owner of retry behavior. Set `maxRetries: 0` to disable.
+     */
+    retry?: Partial<RetryPolicy> & {
+        /** Whether REQUEST_TIMEOUT failures are retryable (default true) */
+        retryOnTimeout?: boolean;
+    };
+    /** Default per-request timeout in ms (overridable per call). SDK defaults apply when unset. */
+    timeoutMs?: number;
+}
+/**
+ * Per-call options for LLMService.sendMessage
+ */
+export interface SendMessageOptions {
+    /**
+     * Abort signal to cancel the request (client-side — the provider may still
+     * process and bill an already-dispatched request). Aborts are never retried.
+     */
+    signal?: AbortSignal;
+    /** Per-request timeout in ms (overrides the service-level timeoutMs) */
+    timeoutMs?: number;
+    /** Per-request retry cap (overrides the service-level retry.maxRetries) */
+    maxRetries?: number;
 }
 /**
  * Result from createMessages method
@@ -46,6 +73,8 @@ export declare class LLMService {
     private requestValidator;
     private settingsManager;
     private modelResolver;
+    private retryOptions;
+    private defaultTimeoutMs?;
     constructor(getApiKey: ApiKeyProvider, options?: LLMServiceOptions);
     /**
      * Gets list of supported LLM providers
@@ -66,7 +95,7 @@ export declare class LLMService {
      * @param request - The LLM chat request
      * @returns Promise resolving to either success or failure response
      */
-    sendMessage(request: LLMChatRequest | LLMChatRequestWithPreset): Promise<LLMResponse | LLMFailureResponse>;
+    sendMessage(request: LLMChatRequest | LLMChatRequestWithPreset, callOptions?: SendMessageOptions): Promise<LLMResponse | LLMFailureResponse>;
     /**
      * Gets all configured model presets
      *

package/dist/llm/LLMService.js

@@ -18,6 +18,8 @@ const AdapterRegistry_1 = require("../shared/services/AdapterRegistry");
 const RequestValidator_1 = require("./services/RequestValidator");
 const SettingsManager_1 = require("./services/SettingsManager");
 const ModelResolver_1 = require("./services/ModelResolver");
+const withRetry_1 = require("../shared/services/withRetry");
+const types_1 = require("./clients/types");
 /**
  * Main process service for LLM operations
  *
@@ -32,6 +34,8 @@ const ModelResolver_1 = require("./services/ModelResolver");
 class LLMService {
     constructor(getApiKey, options = {}) {
         this.getApiKey = getApiKey;
+        this.retryOptions = options.retry;
+        this.defaultTimeoutMs = options.timeoutMs;
         // Initialize logger - custom logger takes precedence over logLevel
         this.logger = options.logger ?? (0, defaultLogger_1.createDefaultLogger)(options.logLevel);
         // Initialize services with logger
@@ -42,7 +46,7 @@ class LLMService {
             adapterConstructors: config_1.ADAPTER_CONSTRUCTORS,
             adapterConfigs: config_1.ADAPTER_CONFIGS,
         }, this.logger);
-        this.requestValidator = new RequestValidator_1.RequestValidator();
+        this.requestValidator = new RequestValidator_1.RequestValidator(this.logger);
         this.settingsManager = new SettingsManager_1.SettingsManager(this.logger);
         this.modelResolver = new ModelResolver_1.ModelResolver(this.presetManager, this.adapterRegistry, this.logger);
     }
@@ -78,7 +82,7 @@ class LLMService {
      * @param request - The LLM chat request
      * @returns Promise resolving to either success or failure response
      */
-    async sendMessage(request) {
+    async sendMessage(request, callOptions) {
         this.logger.info(`LLMService.sendMessage called with presetId: ${request.presetId}, provider: ${request.providerId}, model: ${request.modelId}`);
         try {
             // Resolve model information from preset or direct IDs
@@ -107,7 +111,7 @@ class LLMService {
                 }
             }
             // Apply model-specific defaults and merge with user settings
-            const finalSettings = this.settingsManager.mergeSettingsForModel(modelId, providerId, combinedSettings);
+            const finalSettings = this.settingsManager.mergeSettingsForModel(modelId, providerId, combinedSettings, modelInfo);
             // Validate reasoning settings for model capabilities
             const reasoningValidation = this.requestValidator.validateReasoningSettings(modelInfo, finalSettings.reasoning, resolvedRequest);
             if (reasoningValidation) {
@@ -175,7 +179,41 @@ class LLMService {
                     };
                 }
                 this.logger.info(`Making LLM request with ${clientAdapter.constructor.name} for provider: ${providerId}`);
-                const result = await clientAdapter.sendMessage(internalRequest, apiKey);
+                // Unified retry layer: adapters never throw, so retry decisions are made on
+                // RETURNED failure responses. SDK-internal retries are disabled (maxRetries: 0
+                // at client construction) — this loop is the single owner of retry behavior.
+                const adapterOptions = {
+                    ...(callOptions?.signal && { signal: callOptions.signal }),
+                    ...((callOptions?.timeoutMs ?? this.defaultTimeoutMs) !== undefined && {
+                        timeoutMs: callOptions?.timeoutMs ?? this.defaultTimeoutMs,
+                    }),
+                };
+                const retryOnTimeout = this.retryOptions?.retryOnTimeout ?? true;
+                const retryableCodes = new Set([
+                    types_1.ADAPTER_ERROR_CODES.RATE_LIMIT_EXCEEDED,
+                    types_1.ADAPTER_ERROR_CODES.NETWORK_ERROR,
+                    ...(retryOnTimeout ? [types_1.ADAPTER_ERROR_CODES.REQUEST_TIMEOUT] : []),
+                ]);
+                const result = await (0, withRetry_1.withRetry)(() => clientAdapter.sendMessage(internalRequest, apiKey, adapterOptions), (res) => {
+                    if (res.object !== 'error') {
+                        return { retry: false };
+                    }
+                    const code = String(res.error.code);
+                    const status = res.error.status;
+                    const retryable = retryableCodes.has(code) ||
+                        (code === types_1.ADAPTER_ERROR_CODES.PROVIDER_ERROR &&
+                            typeof status === 'number' &&
+                            (status === 408 || status === 409 || status >= 500));
+                    return { retry: retryable, retryAfterMs: res.error.retryAfterMs };
+                }, {
+                    ...this.retryOptions,
+                    ...(callOptions?.maxRetries !== undefined && {
+                        maxRetries: callOptions.maxRetries,
+                    }),
+                    signal: callOptions?.signal,
+                    logger: this.logger,
+                    label: `${providerId}/${modelId}`,
+                });
                 // Post-process for thinking tag fallback
                 // This feature extracts reasoning from XML tags when native reasoning is not active.
                 // It's a fallback mechanism for models without native reasoning or when native is disabled.
@@ -375,7 +413,7 @@ class LLMService {
             else {
                 const { providerId, modelId, modelInfo, settings } = resolved;
                 // Merge settings with model defaults
-                const mergedSettings = this.settingsManager.mergeSettingsForModel(modelId, providerId, settings || {});
+                const mergedSettings = this.settingsManager.mergeSettingsForModel(modelId, providerId, settings || {}, modelInfo);
                 // Calculate native reasoning status
                 const nativeReasoningActive = !!(modelInfo.reasoning?.supported &&
                     (mergedSettings.reasoning?.enabled === true ||

package/dist/llm/clients/AnthropicClientAdapter.d.ts

@@ -1,5 +1,6 @@
 import type { LLMResponse, LLMFailureResponse } from "../types";
-import type { ILLMClientAdapter, InternalLLMChatRequest } from "./types";
+import type { ILLMClientAdapter, InternalLLMChatRequest, AdapterRequestOptions } from "./types";
+import type { Logger } from "../../logging/types";
 /**
  * Client adapter for Anthropic API integration
  *
@@ -12,14 +13,17 @@ import type { ILLMClientAdapter, InternalLLMChatRequest } from "./types";
  */
 export declare class AnthropicClientAdapter implements ILLMClientAdapter {
     private baseURL?;
+    private logger;
     /**
      * Creates a new Anthropic client adapter
      *
      * @param config Optional configuration for the adapter
      * @param config.baseURL Custom base URL for Anthropic-compatible APIs
+     * @param config.logger Custom logger instance
      */
     constructor(config?: {
         baseURL?: string;
+        logger?: Logger;
     });
     /**
      * Sends a chat message to Anthropic's API
@@ -28,7 +32,7 @@ export declare class AnthropicClientAdapter implements ILLMClientAdapter {
      * @param apiKey - The decrypted Anthropic API key
      * @returns Promise resolving to success or failure response
      */
-    sendMessage(request: InternalLLMChatRequest, apiKey: string): Promise<LLMResponse | LLMFailureResponse>;
+    sendMessage(request: InternalLLMChatRequest, apiKey: string, options?: AdapterRequestOptions): Promise<LLMResponse | LLMFailureResponse>;
     /**
      * Validates Anthropic API key format
      *

package/dist/llm/clients/AnthropicClientAdapter.js

@@ -11,7 +11,6 @@ const types_1 = require("./types");
 const errorUtils_1 = require("../../shared/adapters/errorUtils");
 const systemMessageUtils_1 = require("../../shared/adapters/systemMessageUtils");
 const defaultLogger_1 = require("../../logging/defaultLogger");
-const logger = (0, defaultLogger_1.createDefaultLogger)();
 /**
  * Client adapter for Anthropic API integration
  *
@@ -28,9 +27,11 @@ class AnthropicClientAdapter {
      *
      * @param config Optional configuration for the adapter
      * @param config.baseURL Custom base URL for Anthropic-compatible APIs
+     * @param config.logger Custom logger instance
      */
     constructor(config) {
         this.baseURL = config?.baseURL;
+        this.logger = config?.logger ?? (0, defaultLogger_1.createDefaultLogger)();
     }
     /**
      * Sends a chat message to Anthropic's API
@@ -39,7 +40,7 @@ class AnthropicClientAdapter {
      * @param apiKey - The decrypted Anthropic API key
      * @returns Promise resolving to success or failure response
      */
-    async sendMessage(request, apiKey) {
+    async sendMessage(request, apiKey, options) {
         try {
             // Check if structured output is requested - need beta API
             const useStructuredOutput = request.settings.structuredOutput?.schema &&
@@ -48,7 +49,13 @@ class AnthropicClientAdapter {
             const anthropic = new sdk_1.default({
                 apiKey,
                 ...(this.baseURL && { baseURL: this.baseURL }),
+                maxRetries: 0, // retries are owned by the unified LLMService retry layer
             });
+            // Per-request transport options (abort signal, timeout)
+            const requestTransportOptions = {
+                ...(options?.signal && { signal: options.signal }),
+                ...(options?.timeoutMs !== undefined && { timeout: options.timeoutMs }),
+            };
             // Format messages for Anthropic API (Claude has specific requirements)
             const { messages, systemMessage } = this.formatMessagesForAnthropic(request);
             // Prepare API call parameters
@@ -58,6 +65,9 @@ class AnthropicClientAdapter {
                 max_tokens: request.settings.maxTokens,
                 temperature: request.settings.temperature,
                 top_p: request.settings.topP,
+                ...(request.settings.topK !== undefined && {
+                    top_k: request.settings.topK,
+                }),
                 ...(systemMessage && { system: systemMessage }),
                 ...(request.settings.stopSequences.length > 0 && {
                     stop_sequences: request.settings.stopSequences,
@@ -115,8 +125,8 @@ class AnthropicClientAdapter {
                     };
                 }
             }
-            logger.info(`Making Anthropic API call for model: ${request.modelId}`);
-            logger.debug(`Anthropic API parameters:`, {
+            this.logger.info(`Making Anthropic API call for model: ${request.modelId}`);
+            this.logger.debug(`Anthropic API parameters:`, {
                 model: messageParams.model,
                 temperature: messageParams.temperature,
                 max_tokens: messageParams.max_tokens,
@@ -134,18 +144,22 @@ class AnthropicClientAdapter {
                     headers: {
                         'anthropic-beta': 'structured-outputs-2025-11-13',
                     },
+                    ...requestTransportOptions,
                 });
             }
             else {
-                completion = await anthropic.messages.create(messageParams);
+                completion =
+                    Object.keys(requestTransportOptions).length > 0
+                        ? await anthropic.messages.create(messageParams, requestTransportOptions)
+                        : await anthropic.messages.create(messageParams);
             }
-            logger.info(`Anthropic API call successful, response ID: ${completion.id}`);
+            this.logger.info(`Anthropic API call successful, response ID: ${completion.id}`);
             // Convert to standardized response format
             // Cast to any to handle beta response type differences
             return this.createSuccessResponse(completion, request);
         }
         catch (error) {
-            logger.error("Anthropic API error:", error);
+            this.logger.error("Anthropic API error:", error);
             return this.createErrorResponse(error, request);
         }
     }
@@ -245,7 +259,7 @@ class AnthropicClientAdapter {
                 const modifiedIndex = (0, systemMessageUtils_1.prependSystemToFirstUserMessage)(simpleMessages, combinedSystemContent, request.settings.systemMessageFallback);
                 if (modifiedIndex !== -1) {
                     messages[modifiedIndex].content = simpleMessages[modifiedIndex].content;
-                    logger.debug(`Model ${request.modelId} doesn't support system messages - prepended to first user message`);
+                    this.logger.debug(`Model ${request.modelId} doesn't support system messages - prepended to first user message`);
                 }
                 // Don't set systemMessage - it stays undefined
             }
@@ -253,7 +267,7 @@ class AnthropicClientAdapter {
         // Anthropic requires messages to start with 'user' role
         // If the first message is not from user, we need to handle this
         if (messages.length > 0 && messages[0].role !== "user") {
-            logger.warn("Anthropic API requires first message to be from user. Adjusting message order.");
+            this.logger.warn("Anthropic API requires first message to be from user. Adjusting message order.");
             // Find the first user message and move it to the front, or create a default one
             const firstUserIndex = messages.findIndex((msg) => msg.role === "user");
             if (firstUserIndex > 0) {
@@ -295,7 +309,7 @@ class AnthropicClientAdapter {
                 // If roles don't alternate properly, we might need to combine messages
                 // or insert a placeholder. For now, we'll skip non-alternating messages
                 // and log a warning.
-                logger.warn(`Skipping message with unexpected role: expected ${expectedRole}, got ${message.role}`);
+                this.logger.warn(`Skipping message with unexpected role: expected ${expectedRole}, got ${message.role}`);
             }
         }
         return cleanedMessages;
@@ -386,7 +400,7 @@ class AnthropicClientAdapter {
     createErrorResponse(error, request) {
         // Use shared error mapping utility for common error patterns
         const initialProviderMessage = error instanceof sdk_1.default.APIError ? error.message : undefined;
-        let { errorCode, errorMessage, errorType, status } = (0, errorUtils_1.getCommonMappedErrorDetails)(error, initialProviderMessage);
+        let { errorCode, errorMessage, errorType, status, retryAfterMs } = (0, errorUtils_1.getCommonMappedErrorDetails)(error, initialProviderMessage);
         // Apply Anthropic-specific refinements for 400 errors based on message content
         if (error instanceof sdk_1.default.APIError && status === 400) {
             if (error.message.toLowerCase().includes("context length") ||
@@ -408,6 +422,7 @@ class AnthropicClientAdapter {
                 code: errorCode,
                 type: errorType,
                 ...(status && { status }),
+                ...(retryAfterMs !== undefined && { retryAfterMs }),
                 providerError: error,
             },
             object: "error",

package/dist/llm/clients/GeminiClientAdapter.d.ts

@@ -1,5 +1,6 @@
 import type { LLMResponse, LLMFailureResponse } from "../types";
-import type { ILLMClientAdapter, InternalLLMChatRequest } from "./types";
+import type { ILLMClientAdapter, InternalLLMChatRequest, AdapterRequestOptions } from "./types";
+import type { Logger } from "../../logging/types";
 /**
  * Client adapter for Google Gemini API integration
  *
@@ -12,14 +13,17 @@ import type { ILLMClientAdapter, InternalLLMChatRequest } from "./types";
  */
 export declare class GeminiClientAdapter implements ILLMClientAdapter {
     private baseURL?;
+    private logger;
     /**
      * Creates a new Gemini client adapter
      *
      * @param config Optional configuration for the adapter
      * @param config.baseURL Custom base URL (unused for Gemini but kept for consistency)
+     * @param config.logger Custom logger instance
      */
     constructor(config?: {
         baseURL?: string;
+        logger?: Logger;
     });
     /**
      * Sends a chat message to Gemini's API
@@ -28,7 +32,7 @@ export declare class GeminiClientAdapter implements ILLMClientAdapter {
      * @param apiKey - The decrypted Gemini API key
      * @returns Promise resolving to success or failure response
      */
-    sendMessage(request: InternalLLMChatRequest, apiKey: string): Promise<LLMResponse | LLMFailureResponse>;
+    sendMessage(request: InternalLLMChatRequest, apiKey: string, options?: AdapterRequestOptions): Promise<LLMResponse | LLMFailureResponse>;
     /**
      * Validates Gemini API key format
      *

package/dist/llm/clients/GeminiClientAdapter.js

@@ -8,7 +8,6 @@ const types_1 = require("./types");
 const errorUtils_1 = require("../../shared/adapters/errorUtils");
 const systemMessageUtils_1 = require("../../shared/adapters/systemMessageUtils");
 const defaultLogger_1 = require("../../logging/defaultLogger");
-const logger = (0, defaultLogger_1.createDefaultLogger)();
 /**
  * Client adapter for Google Gemini API integration
  *
@@ -25,9 +24,11 @@ class GeminiClientAdapter {
      *
      * @param config Optional configuration for the adapter
      * @param config.baseURL Custom base URL (unused for Gemini but kept for consistency)
+     * @param config.logger Custom logger instance
      */
     constructor(config) {
         this.baseURL = config?.baseURL;
+        this.logger = config?.logger ?? (0, defaultLogger_1.createDefaultLogger)();
     }
     /**
      * Sends a chat message to Gemini's API
@@ -36,14 +37,14 @@ class GeminiClientAdapter {
      * @param apiKey - The decrypted Gemini API key
      * @returns Promise resolving to success or failure response
      */
-    async sendMessage(request, apiKey) {
+    async sendMessage(request, apiKey, options) {
         try {
-            // Initialize Gemini client
+            // Initialize Gemini client (note: @google/genai performs no automatic retries)
             const genAI = new genai_1.GoogleGenAI({ apiKey });
             // Format the request for Gemini API
             const { contents, generationConfig, safetySettings, systemInstruction } = this.formatInternalRequestToGemini(request);
-            logger.info(`Making Gemini API call for model: ${request.modelId}`);
-            logger.debug(`Gemini API parameters:`, {
+            this.logger.info(`Making Gemini API call for model: ${request.modelId}`);
+            this.logger.debug(`Gemini API parameters:`, {
                 model: request.modelId,
                 temperature: generationConfig.temperature,
                 maxOutputTokens: generationConfig.maxOutputTokens,
@@ -59,14 +60,18 @@ class GeminiClientAdapter {
                     ...generationConfig,
                     safetySettings: safetySettings,
                     ...(systemInstruction && { systemInstruction: systemInstruction }),
+                    ...(options?.signal && { abortSignal: options.signal }),
+                    ...(options?.timeoutMs !== undefined && {
+                        httpOptions: { timeout: options.timeoutMs },
+                    }),
                 },
             });
-            logger.info(`Gemini API call successful, processing response`);
+            this.logger.info(`Gemini API call successful, processing response`);
             // Convert to standardized response format
             return this.createSuccessResponse(result, request);
         }
         catch (error) {
-            logger.error("Gemini API error:", error);
+            this.logger.error("Gemini API error:", error);
             return this.createErrorResponse(error, request);
         }
     }
@@ -142,7 +147,7 @@ class GeminiClientAdapter {
                 if (modifiedIndex !== -1) {
                     // Update the actual contents array
                     contents[modifiedIndex].parts[0].text = simpleContents[modifiedIndex].content;
-                    logger.debug(`Model ${request.modelId} doesn't support system instructions - prepended to first user message`);
+                    this.logger.debug(`Model ${request.modelId} doesn't support system instructions - prepended to first user message`);
                 }
                 // Don't set systemInstruction - it stays undefined
             }
@@ -152,6 +157,12 @@ class GeminiClientAdapter {
             maxOutputTokens: request.settings.maxTokens,
             temperature: request.settings.temperature,
             ...(request.settings.topP && { topP: request.settings.topP }),
+            ...(request.settings.topK !== undefined && {
+                topK: request.settings.topK,
+            }),
+            ...(request.settings.seed !== undefined && {
+                seed: request.settings.seed,
+            }),
             ...(request.settings.stopSequences &&
                 request.settings.stopSequences.length > 0 && {
                 stopSequences: request.settings.stopSequences,
@@ -306,7 +317,7 @@ class GeminiClientAdapter {
     createErrorResponse(error, request) {
         // Use shared error mapping utility for common error patterns
         const initialProviderMessage = error?.message;
-        let { errorCode, errorMessage, errorType, status } = (0, errorUtils_1.getCommonMappedErrorDetails)(error, initialProviderMessage);
+        let { errorCode, errorMessage, errorType, status, retryAfterMs } = (0, errorUtils_1.getCommonMappedErrorDetails)(error, initialProviderMessage);
         // Apply Gemini-specific refinements for certain error types
         if (error && error.message) {
             const message = error.message.toLowerCase();
@@ -336,6 +347,7 @@ class GeminiClientAdapter {
                 code: errorCode,
                 type: errorType,
                 ...(status && { status }),
+                ...(retryAfterMs !== undefined && { retryAfterMs }),
                 providerError: error,
             },
             object: "error",

package/dist/llm/clients/LlamaCppClientAdapter.d.ts

@@ -1,14 +1,17 @@
 import type { LLMResponse, LLMFailureResponse, ModelInfo } from "../types";
-import type { ILLMClientAdapter, InternalLLMChatRequest } from "./types";
+import type { ILLMClientAdapter, InternalLLMChatRequest, AdapterRequestOptions } from "./types";
 import { LlamaCppServerClient } from "./LlamaCppServerClient";
+import type { Logger } from "../../logging/types";
 /**
  * Configuration options for LlamaCppClientAdapter
  */
 export interface LlamaCppClientConfig {
-    /** Base URL of the llama.cpp server (default: http://localhost:8080) */
+    /** Base URL of the llama.cpp server (default: http://127.0.0.1:8080) */
     baseURL?: string;
     /** Whether to check server health before sending requests (default: false) */
     checkHealth?: boolean;
+    /** Logger instance for adapter logging */
+    logger?: Logger;
 }
 /**
  * Client adapter for llama.cpp server integration
@@ -30,7 +33,7 @@ export interface LlamaCppClientConfig {
  * ```typescript
  * // Create adapter for local server
  * const adapter = new LlamaCppClientAdapter({
- *   baseURL: 'http://localhost:8080',
+ *   baseURL: 'http://127.0.0.1:8080',
  *   checkHealth: true
  * });
  *
@@ -51,6 +54,7 @@ export declare class LlamaCppClientAdapter implements ILLMClientAdapter {
     private serverClient;
     private cachedModelCapabilities;
     private detectionAttempted;
+    private logger;
     /**
      * Creates a new llama.cpp client adapter
      *
@@ -80,7 +84,7 @@ export declare class LlamaCppClientAdapter implements ILLMClientAdapter {
      * @param apiKey - Not used for llama.cpp (local server), but kept for interface compatibility
      * @returns Promise resolving to success or failure response
      */
-    sendMessage(request: InternalLLMChatRequest, apiKey: string): Promise<LLMResponse | LLMFailureResponse>;
+    sendMessage(request: InternalLLMChatRequest, apiKey: string, options?: AdapterRequestOptions): Promise<LLMResponse | LLMFailureResponse>;
     /**
      * Validates API key format
      *