@juspay/neurolink 9.26.2 → 9.28.0

package/dist/providers/litellm.d.ts

@@ -29,6 +29,16 @@ export declare class LiteLLMProvider extends BaseProvider {
      * Note: This is only used when tools are disabled
      */
     protected executeStream(options: StreamOptions, analysisSchema?: ZodType<unknown, ZodTypeDef, unknown> | Schema<unknown>): Promise<StreamResult>;
+    /**
+     * Generate an embedding for a single text input
+     * Uses the LiteLLM proxy with OpenAI-compatible embedding API
+     */
+    embed(text: string, modelName?: string): Promise<number[]>;
+    /**
+     * Generate embeddings for multiple text inputs
+     * Uses the LiteLLM proxy with OpenAI-compatible embedding API
+     */
+    embedMany(texts: string[], modelName?: string): Promise<number[][]>;
     /**
      * Get available models from LiteLLM proxy server
      * Dynamically fetches from /v1/models endpoint with caching and fallback

package/dist/providers/litellm.js

@@ -1,4 +1,5 @@
 import { createOpenAI } from "@ai-sdk/openai";
+import { SpanKind, SpanStatusCode, trace } from "@opentelemetry/api";
 import { Output, streamText, } from "ai";
 import { BaseProvider } from "../core/baseProvider.js";
 import { DEFAULT_MAX_STEPS } from "../core/constants.js";
@@ -7,8 +8,10 @@ import { createProxyFetch } from "../proxy/proxyFetch.js";
 import { AuthenticationError, InvalidModelError, NetworkError, ProviderError, RateLimitError, } from "../types/errors.js";
 import { isAbortError } from "../utils/errorHandling.js";
 import { logger } from "../utils/logger.js";
+import { calculateCost } from "../utils/pricing.js";
 import { getProviderModel } from "../utils/providerConfig.js";
 import { composeAbortSignals, createTimeoutController, TimeoutError, } from "../utils/timeout.js";
+const streamTracer = trace.getTracer("neurolink.provider.litellm");
 // Configuration helpers
 const getLiteLLMConfig = () => {
     return {
@@ -55,7 +58,9 @@ export class LiteLLMProvider extends BaseProvider {
             apiKey: config.apiKey,
             fetch: createProxyFetch(),
         });
-        this.model = customOpenAI(this.modelName || getDefaultLiteLLMModel());
+        this.model = customOpenAI(this.modelName || getDefaultLiteLLMModel(), {
+            structuredOutputs: false,
+        });
         logger.debug("LiteLLM Provider initialized", {
             modelName: this.modelName,
             provider: this.providerName,
@@ -206,7 +211,64 @@ export class LiteLLMProvider extends BaseProvider {
                     });
                 }
             }
-            const result = await streamText(streamOptions);
+            // Wrap streamText in an OTel span to capture provider-level latency, token usage, and cost
+            const streamSpan = streamTracer.startSpan("neurolink.provider.streamText", {
+                kind: SpanKind.CLIENT,
+                attributes: {
+                    "gen_ai.system": "litellm",
+                    "gen_ai.request.model": model.modelId || this.modelName || "unknown",
+                },
+            });
+            let result;
+            try {
+                result = streamText(streamOptions);
+            }
+            catch (streamError) {
+                streamSpan.setStatus({
+                    code: SpanStatusCode.ERROR,
+                    message: streamError instanceof Error
+                        ? streamError.message
+                        : String(streamError),
+                });
+                streamSpan.end();
+                throw streamError;
+            }
+            // Collect token usage, cost, and finish reason asynchronously when the stream completes,
+            // then end the span. This avoids blocking the stream consumer.
+            result.usage
+                .then((usage) => {
+                streamSpan.setAttribute("gen_ai.usage.input_tokens", usage.promptTokens || 0);
+                streamSpan.setAttribute("gen_ai.usage.output_tokens", usage.completionTokens || 0);
+                const cost = calculateCost(this.providerName, this.modelName, {
+                    input: usage.promptTokens || 0,
+                    output: usage.completionTokens || 0,
+                    total: (usage.promptTokens || 0) + (usage.completionTokens || 0),
+                });
+                if (cost && cost > 0) {
+                    streamSpan.setAttribute("neurolink.cost", cost);
+                }
+            })
+                .catch(() => {
+                // Usage may not be available if the stream is aborted
+            });
+            result.finishReason
+                .then((reason) => {
+                streamSpan.setAttribute("gen_ai.response.finish_reason", reason || "unknown");
+            })
+                .catch(() => {
+                // Finish reason may not be available if the stream is aborted
+            });
+            result.text
+                .then(() => {
+                streamSpan.end();
+            })
+                .catch((err) => {
+                streamSpan.setStatus({
+                    code: SpanStatusCode.ERROR,
+                    message: err instanceof Error ? err.message : String(err),
+                });
+                streamSpan.end();
+            });
             timeoutController?.cleanup();
             // Transform stream to content object stream using fullStream (handles both text and tool calls)
             // Note: fullStream includes tool results, textStream only has text
@@ -269,6 +331,46 @@ export class LiteLLMProvider extends BaseProvider {
             throw this.handleProviderError(error);
         }
     }
+    /**
+     * Generate an embedding for a single text input
+     * Uses the LiteLLM proxy with OpenAI-compatible embedding API
+     */
+    async embed(text, modelName) {
+        const { embed: aiEmbed } = await import("ai");
+        const { createOpenAI } = await import("@ai-sdk/openai");
+        const config = getLiteLLMConfig();
+        const embeddingModelName = modelName ||
+            process.env.LITELLM_EMBEDDING_MODEL ||
+            "gemini-embedding-001";
+        const customOpenAI = createOpenAI({
+            baseURL: config.baseURL,
+            apiKey: config.apiKey,
+            fetch: createProxyFetch(),
+        });
+        const embeddingModel = customOpenAI.textEmbeddingModel(embeddingModelName);
+        const result = await aiEmbed({ model: embeddingModel, value: text });
+        return result.embedding;
+    }
+    /**
+     * Generate embeddings for multiple text inputs
+     * Uses the LiteLLM proxy with OpenAI-compatible embedding API
+     */
+    async embedMany(texts, modelName) {
+        const { embedMany: aiEmbedMany } = await import("ai");
+        const { createOpenAI } = await import("@ai-sdk/openai");
+        const config = getLiteLLMConfig();
+        const embeddingModelName = modelName ||
+            process.env.LITELLM_EMBEDDING_MODEL ||
+            "gemini-embedding-001";
+        const customOpenAI = createOpenAI({
+            baseURL: config.baseURL,
+            apiKey: config.apiKey,
+            fetch: createProxyFetch(),
+        });
+        const embeddingModel = customOpenAI.textEmbeddingModel(embeddingModelName);
+        const result = await aiEmbedMany({ model: embeddingModel, values: texts });
+        return result.embeddings;
+    }
     /**
      * Get available models from LiteLLM proxy server
      * Dynamically fetches from /v1/models endpoint with caching and fallback

package/dist/types/configTypes.d.ts

@@ -6,6 +6,9 @@ import { MCPToolRegistry } from "../mcp/toolRegistry.js";
 import type { HITLConfig } from "../types/hitlTypes.js";
 import type { ConversationMemoryConfig } from "./conversation.js";
 import type { ObservabilityConfig } from "./observability.js";
+import type { RoutingStrategy } from "../mcp/routing/index.js";
+import type { CacheStrategy } from "../mcp/caching/index.js";
+import type { ToolMiddleware } from "../mcp/toolIntegration.js";
 /**
  * Main NeuroLink configuration type
  */
@@ -28,6 +31,59 @@ export type NeurolinkConstructorConfig = {
     toolRegistry?: MCPToolRegistry;
     observability?: ObservabilityConfig;
     modelAliasConfig?: import("./generateTypes.js").ModelAliasConfig;
+    /** MCP enhancement modules configuration (cache, router, batcher, annotations, middleware) */
+    mcp?: MCPEnhancementsConfig;
+};
+/**
+ * Configuration for MCP enhancement modules wired into generate()/stream() paths.
+ *
+ * These modules are automatically applied during tool execution when configured:
+ * - cache: Tool result caching (disabled by default)
+ * - annotations: Auto-infer tool safety metadata (enabled by default)
+ * - router: Multi-server tool routing (auto-activates with 2+ servers)
+ * - batcher: Batch programmatic tool calls (disabled by default)
+ * - discovery: Enhanced tool search and filtering (enabled by default)
+ * - middleware: Global tool execution middleware chain (empty by default)
+ */
+export type MCPEnhancementsConfig = {
+    /** Tool result caching. Default: disabled. Enable to cache read-only tool results. */
+    cache?: {
+        enabled?: boolean;
+        /** Cache TTL in milliseconds. Default: 300000 (5 min) */
+        ttl?: number;
+        /** Maximum cache entries. Default: 500 */
+        maxSize?: number;
+        /** Eviction strategy. Default: 'lru' */
+        strategy?: CacheStrategy;
+    };
+    /** Tool annotation auto-inference. Default: enabled. */
+    annotations?: {
+        enabled?: boolean;
+        /** Auto-infer annotations from tool name/description. Default: true */
+        autoInfer?: boolean;
+    };
+    /** Tool routing for multi-server environments. Auto-activates when 2+ external servers exist. */
+    router?: {
+        enabled?: boolean;
+        /** Routing strategy. Default: 'least-loaded' */
+        strategy?: RoutingStrategy;
+        /** Enable session affinity. Default: false */
+        enableAffinity?: boolean;
+    };
+    /** Request batching for programmatic executeTool() calls. Default: disabled. */
+    batcher?: {
+        enabled?: boolean;
+        /** Max requests per batch. Default: 10 */
+        maxBatchSize?: number;
+        /** Max wait before flushing batch in ms. Default: 100 */
+        maxWaitMs?: number;
+    };
+    /** Enhanced tool discovery. Default: enabled. */
+    discovery?: {
+        enabled?: boolean;
+    };
+    /** Global tool middleware applied to every tool execution. Default: empty. */
+    middleware?: ToolMiddleware[];
 };
 /**
  * Provider-specific configuration

package/dist/types/conversation.d.ts

@@ -33,8 +33,8 @@
  * - Current time (ISO): `new Date().toISOString()`
  */
 import type { Mem0Config } from "../memory/mem0Initializer.js";
-import type { Memory } from "../memory/hippocampusInitializer.js";
-export type { Memory };
+import type { Memory, CustomStorageConfig } from "../memory/hippocampusInitializer.js";
+export type { Memory, CustomStorageConfig };
 /**
  * Configuration for conversation memory feature
  */

package/dist/types/generateTypes.d.ts

@@ -259,6 +259,8 @@ export type GenerateOptions = {
      * Default: false (backward compatible — tool schemas are injected into system prompt).
      */
     skipToolPromptInjection?: boolean;
+    /** Disable tool result caching for this request (overrides global mcp.cache.enabled) */
+    disableToolCache?: boolean;
     /** Maximum number of tool execution steps (default: 200) */
     maxSteps?: number;
     /**
@@ -662,6 +664,8 @@ export type TextGenerationOptions = {
     toolFilter?: string[];
     /** Exclude these tools by name (blacklist). Applied after toolFilter. */
     excludeTools?: string[];
+    /** Disable tool result caching for this request (overrides global mcp.cache.enabled) */
+    disableToolCache?: boolean;
     /**
      * Tool choice configuration for the generation.
      * Controls whether and which tools the model must call.

package/dist/types/index.d.ts

@@ -4,7 +4,7 @@
 export { AIProviderName } from "../constants/enums.js";
 export * from "./cli.js";
 export * from "./common.js";
-export type { AnalyticsConfig, BackupInfo, BackupMetadata, CacheConfig, ConfigUpdateOptions, ConfigValidationResult, FallbackConfig, NeuroLinkConfig, PerformanceConfig, RetryConfig, ToolConfig, } from "./configTypes.js";
+export type { AnalyticsConfig, BackupInfo, BackupMetadata, CacheConfig, ConfigUpdateOptions, ConfigValidationResult, FallbackConfig, MCPEnhancementsConfig, NeuroLinkConfig, PerformanceConfig, RetryConfig, ToolConfig, } from "./configTypes.js";
 export type { ExternalMCPConfigValidation, ExternalMCPManagerConfig, ExternalMCPOperationResult, ExternalMCPServerEvents, ExternalMCPServerHealth, ExternalMCPServerInstance, ExternalMCPServerStatus, ExternalMCPToolContext, ExternalMCPToolInfo, ExternalMCPToolResult, } from "./externalMcp.js";
 export type { AuthorizationUrlResult, CircuitBreakerConfig, CircuitBreakerEvents, CircuitBreakerState, CircuitBreakerStats, DiscoveredMcp, ExternalToolExecutionOptions, FlexibleValidationResult, HTTPRetryConfig, MCPClientResult, MCPConnectedServer, MCPDiscoveredServer, MCPExecutableTool, MCPOAuthConfig, MCPServerCategory, MCPServerConfig, MCPServerConnectionStatus, MCPServerMetadata, MCPServerRegistryEntry, MCPServerStatus, MCPToolInfo, MCPToolMetadata, MCPTransportType, McpMetadata, McpRegistry, NeuroLinkExecutionContext, NeuroLinkMCPServer, NeuroLinkMCPTool, OAuthClientInformation, OAuthTokens as McpOAuthTokens, RateLimitConfig, TokenBucketRateLimitConfig, TokenExchangeRequest, TokenStorage, ToolDiscoveryResult, ToolRegistryEvents, ToolValidationResult, } from "./mcpTypes.js";
 export type { ModelCapability, ModelFilter, ModelPricing, ModelResolutionContext, ModelStats, ModelUseCase, } from "./providers.js";
@@ -38,4 +38,5 @@ export * from "./contextTypes.js";
 export * from "./fileReferenceTypes.js";
 export * from "./ragTypes.js";
 export * from "./conversationMemoryInterface.js";
+export type { CustomStorageConfig } from "./conversation.js";
 export * from "./subscriptionTypes.js";

package/dist/types/streamTypes.d.ts

@@ -329,6 +329,8 @@ export type StreamOptions = {
     toolFilter?: string[];
     /** Exclude these tools by name (blacklist). Applied after toolFilter. */
     excludeTools?: string[];
+    /** Disable tool result caching for this request (overrides global mcp.cache.enabled) */
+    disableToolCache?: boolean;
     /**
      * Skip injecting tool schemas into the system prompt.
      * When true, tools are ONLY passed natively via the provider's `tools` parameter,

package/dist/types/tools.d.ts

@@ -75,6 +75,8 @@ export type ToolInfo = {
     serverId?: string;
     inputSchema?: StandardRecord;
     outputSchema?: StandardRecord;
+    /** MCP tool annotations (safety hints, metadata). Auto-inferred when mcp.annotations.autoInfer is enabled. */
+    annotations?: import("../mcp/toolAnnotations.js").MCPToolAnnotations;
     [key: string]: unknown;
 };
 /**

package/dist/utils/pricing.js

@@ -9,69 +9,198 @@
  * (Bedrock, Azure, Mistral, etc.) will return 0 from calculateCost().
  */
 const PRICING = {
-    // Anthropic (direct API)
+    // Anthropic (direct API) — updated March 2026
     anthropic: {
+        // Claude 4.6 family
+        "claude-opus-4-6": {
+            input: 5.0 / 1_000_000,
+            output: 25.0 / 1_000_000,
+            cacheRead: 0.5 / 1_000_000,
+            cacheCreation: 6.25 / 1_000_000,
+        },
+        "claude-sonnet-4-6": {
+            input: 3.0 / 1_000_000,
+            output: 15.0 / 1_000_000,
+            cacheRead: 0.3 / 1_000_000,
+            cacheCreation: 3.75 / 1_000_000,
+        },
+        // Claude 4.5 family
         "claude-sonnet-4-5-20250929": {
             input: 3.0 / 1_000_000,
             output: 15.0 / 1_000_000,
             cacheRead: 0.3 / 1_000_000,
             cacheCreation: 3.75 / 1_000_000,
         },
-        "claude-opus-4-6": {
+        "claude-opus-4-5": {
+            input: 5.0 / 1_000_000,
+            output: 25.0 / 1_000_000,
+            cacheRead: 0.5 / 1_000_000,
+            cacheCreation: 6.25 / 1_000_000,
+        },
+        "claude-haiku-4-5-20251001": {
+            input: 1.0 / 1_000_000,
+            output: 5.0 / 1_000_000,
+            cacheRead: 0.1 / 1_000_000,
+            cacheCreation: 1.25 / 1_000_000,
+        },
+        // Claude 4.0/4.1 family
+        "claude-opus-4-1": {
             input: 15.0 / 1_000_000,
             output: 75.0 / 1_000_000,
             cacheRead: 1.5 / 1_000_000,
             cacheCreation: 18.75 / 1_000_000,
         },
-        "claude-haiku-4-5-20251001": {
+        "claude-opus-4": {
+            input: 15.0 / 1_000_000,
+            output: 75.0 / 1_000_000,
+            cacheRead: 1.5 / 1_000_000,
+            cacheCreation: 18.75 / 1_000_000,
+        },
+        "claude-sonnet-4": {
+            input: 3.0 / 1_000_000,
+            output: 15.0 / 1_000_000,
+            cacheRead: 0.3 / 1_000_000,
+            cacheCreation: 3.75 / 1_000_000,
+        },
+        // Claude 3.x family
+        "claude-3-7-sonnet": {
+            input: 3.0 / 1_000_000,
+            output: 15.0 / 1_000_000,
+            cacheRead: 0.3 / 1_000_000,
+            cacheCreation: 3.75 / 1_000_000,
+        },
+        "claude-3-5-sonnet": {
+            input: 3.0 / 1_000_000,
+            output: 15.0 / 1_000_000,
+            cacheRead: 0.3 / 1_000_000,
+            cacheCreation: 3.75 / 1_000_000,
+        },
+        "claude-3-5-haiku": {
             input: 0.8 / 1_000_000,
             output: 4.0 / 1_000_000,
             cacheRead: 0.08 / 1_000_000,
             cacheCreation: 1.0 / 1_000_000,
         },
+        "claude-3-opus": {
+            input: 15.0 / 1_000_000,
+            output: 75.0 / 1_000_000,
+            cacheRead: 1.5 / 1_000_000,
+            cacheCreation: 18.75 / 1_000_000,
+        },
+        "claude-3-sonnet": { input: 3.0 / 1_000_000, output: 15.0 / 1_000_000 },
+        "claude-3-haiku": { input: 0.25 / 1_000_000, output: 1.25 / 1_000_000 },
     },
-    // Google Vertex AI (same models, same pricing)
+    // Google Vertex AI — Claude models on Vertex (same pricing, @ date suffix)
     vertex: {
-        "claude-sonnet-4-5@20250929": {
+        "claude-sonnet-4-6": {
             input: 3.0 / 1_000_000,
             output: 15.0 / 1_000_000,
             cacheRead: 0.3 / 1_000_000,
             cacheCreation: 3.75 / 1_000_000,
         },
         "claude-opus-4-6": {
-            input: 15.0 / 1_000_000,
-            output: 75.0 / 1_000_000,
-            cacheRead: 1.5 / 1_000_000,
-            cacheCreation: 18.75 / 1_000_000,
+            input: 5.0 / 1_000_000,
+            output: 25.0 / 1_000_000,
+            cacheRead: 0.5 / 1_000_000,
+            cacheCreation: 6.25 / 1_000_000,
         },
-        "claude-haiku-4-5@20251001": {
+        "claude-sonnet-4-5": {
+            input: 3.0 / 1_000_000,
+            output: 15.0 / 1_000_000,
+            cacheRead: 0.3 / 1_000_000,
+            cacheCreation: 3.75 / 1_000_000,
+        },
+        "claude-opus-4-5": {
+            input: 5.0 / 1_000_000,
+            output: 25.0 / 1_000_000,
+            cacheRead: 0.5 / 1_000_000,
+            cacheCreation: 6.25 / 1_000_000,
+        },
+        "claude-haiku-4-5": {
+            input: 1.0 / 1_000_000,
+            output: 5.0 / 1_000_000,
+            cacheRead: 0.1 / 1_000_000,
+            cacheCreation: 1.25 / 1_000_000,
+        },
+        "claude-3-5-haiku": {
             input: 0.8 / 1_000_000,
             output: 4.0 / 1_000_000,
             cacheRead: 0.08 / 1_000_000,
             cacheCreation: 1.0 / 1_000_000,
         },
+        "claude-3-5-sonnet": {
+            input: 3.0 / 1_000_000,
+            output: 15.0 / 1_000_000,
+            cacheRead: 0.3 / 1_000_000,
+            cacheCreation: 3.75 / 1_000_000,
+        },
     },
-    // OpenAI
+    // OpenAI — updated March 2026
     openai: {
+        // GPT-5.x family
+        "gpt-5.4": { input: 2.5 / 1_000_000, output: 15.0 / 1_000_000 },
+        "gpt-5.2": { input: 1.75 / 1_000_000, output: 14.0 / 1_000_000 },
+        "gpt-5.1": { input: 0.625 / 1_000_000, output: 5.0 / 1_000_000 },
+        "gpt-5.1-codex": { input: 1.25 / 1_000_000, output: 10.0 / 1_000_000 },
+        "gpt-5": { input: 1.25 / 1_000_000, output: 10.0 / 1_000_000 },
+        "gpt-5-mini": { input: 0.25 / 1_000_000, output: 2.0 / 1_000_000 },
+        "gpt-5-nano": { input: 0.05 / 1_000_000, output: 0.4 / 1_000_000 },
+        // GPT-4.1 family
+        "gpt-4.1": { input: 2.0 / 1_000_000, output: 8.0 / 1_000_000 },
+        "gpt-4.1-mini": { input: 0.4 / 1_000_000, output: 1.6 / 1_000_000 },
+        "gpt-4.1-nano": { input: 0.1 / 1_000_000, output: 0.4 / 1_000_000 },
+        // GPT-4o family
         "gpt-4o": { input: 2.5 / 1_000_000, output: 10.0 / 1_000_000 },
         "gpt-4o-mini": { input: 0.15 / 1_000_000, output: 0.6 / 1_000_000 },
-        "gpt-4-turbo": { input: 10.0 / 1_000_000, output: 30.0 / 1_000_000 },
+        // o-series reasoning
+        o3: { input: 2.0 / 1_000_000, output: 8.0 / 1_000_000 },
+        "o3-mini": { input: 1.1 / 1_000_000, output: 4.4 / 1_000_000 },
+        "o4-mini": { input: 1.1 / 1_000_000, output: 4.4 / 1_000_000 },
         o1: { input: 15.0 / 1_000_000, output: 60.0 / 1_000_000 },
-        "o1-mini": { input: 1.1 / 1_000_000, output: 4.4 / 1_000_000 },
+        "o1-mini": { input: 0.55 / 1_000_000, output: 2.2 / 1_000_000 },
+        // Legacy
+        "gpt-4-turbo": { input: 10.0 / 1_000_000, output: 30.0 / 1_000_000 },
+        "gpt-4": { input: 30.0 / 1_000_000, output: 60.0 / 1_000_000 },
+        "gpt-3.5-turbo": { input: 0.5 / 1_000_000, output: 1.0 / 1_000_000 },
     },
-    // Google (Gemini)
+    // Google (Gemini) — updated March 2026
     google: {
-        "gemini-2.5-flash": { input: 0.15 / 1_000_000, output: 0.6 / 1_000_000 },
+        // Gemini 3.x family
+        "gemini-3.1-pro": { input: 2.0 / 1_000_000, output: 12.0 / 1_000_000 },
+        "gemini-3-pro": { input: 2.0 / 1_000_000, output: 12.0 / 1_000_000 },
+        "gemini-3-flash": { input: 0.5 / 1_000_000, output: 3.0 / 1_000_000 },
+        "gemini-3.1-flash-lite": {
+            input: 0.25 / 1_000_000,
+            output: 1.5 / 1_000_000,
+        },
+        // Gemini 2.5 family
+        "gemini-2.5-flash": { input: 0.3 / 1_000_000, output: 2.5 / 1_000_000 },
         "gemini-2.5-pro": { input: 1.25 / 1_000_000, output: 10.0 / 1_000_000 },
         "gemini-2.5-flash-lite": {
+            input: 0.1 / 1_000_000,
+            output: 0.4 / 1_000_000,
+        },
+        // Gemini 2.0 family (deprecated June 2026)
+        "gemini-2.0-flash": { input: 0.15 / 1_000_000, output: 0.6 / 1_000_000 },
+        "gemini-2.0-flash-lite": {
             input: 0.075 / 1_000_000,
             output: 0.3 / 1_000_000,
         },
-        "gemini-2.0-flash": { input: 0.1 / 1_000_000, output: 0.4 / 1_000_000 },
-        "gemini-2.0-pro": { input: 1.25 / 1_000_000, output: 10.0 / 1_000_000 },
+        // Gemini 1.5 family
         "gemini-1.5-pro": { input: 1.25 / 1_000_000, output: 5.0 / 1_000_000 },
         "gemini-1.5-flash": { input: 0.075 / 1_000_000, output: 0.3 / 1_000_000 },
     },
+    // Mistral AI
+    mistral: {
+        "mistral-large": { input: 2.0 / 1_000_000, output: 6.0 / 1_000_000 },
+        "mistral-medium": { input: 2.7 / 1_000_000, output: 8.1 / 1_000_000 },
+        "mistral-small": { input: 0.2 / 1_000_000, output: 0.6 / 1_000_000 },
+        codestral: { input: 0.3 / 1_000_000, output: 0.9 / 1_000_000 },
+        "open-mistral-nemo": {
+            input: 0.15 / 1_000_000,
+            output: 0.15 / 1_000_000,
+        },
+    },
 };
 /**
  * Map of normalized provider aliases to canonical PRICING keys.
@@ -80,10 +209,20 @@ const PRICING = {
 const PROVIDER_ALIASES = {
     googleai: "google",
     googleaistudio: "google",
+    googlevertex: "vertex",
     anthropic: "anthropic",
     openai: "openai",
     vertex: "vertex",
     google: "google",
+    mistral: "mistral",
+    mistralai: "mistral",
+    azure: "openai",
+    azureopenai: "openai",
+    bedrock: "anthropic",
+    amazonbedrock: "anthropic",
+    litellm: "__cross_provider__",
+    openrouter: "__cross_provider__",
+    openaicompatible: "__cross_provider__",
 };
 /**
  * Look up per-token rates for a provider/model combination.
@@ -96,6 +235,27 @@ const PROVIDER_ALIASES = {
 function findRates(provider, model) {
     const stripped = provider.toLowerCase().replace(/[^a-z]/g, "");
     const normalizedProvider = PROVIDER_ALIASES[stripped] ?? stripped;
+    // Proxy providers (LiteLLM, OpenRouter): search all known providers for a model match
+    if (normalizedProvider === "__cross_provider__") {
+        for (const providerPricing of Object.values(PRICING)) {
+            // Exact match
+            if (providerPricing[model]) {
+                return providerPricing[model];
+            }
+            const sortedKeys = Object.keys(providerPricing).sort((a, b) => b.length - a.length);
+            // model is a prefix of a known key (e.g. "claude-sonnet-4-5" matches "claude-sonnet-4-5-20250929")
+            const reverseKey = sortedKeys.find((k) => k.startsWith(model));
+            if (reverseKey) {
+                return providerPricing[reverseKey];
+            }
+            // Known key is a prefix of model (e.g. "gpt-4o" matches "gpt-4o-2024-08-06")
+            const forwardKey = sortedKeys.find((k) => model.startsWith(k));
+            if (forwardKey) {
+                return providerPricing[forwardKey];
+            }
+        }
+        return undefined;
+    }
     const providerPricing = PRICING[normalizedProvider] || PRICING[provider];
     if (!providerPricing) {
         return undefined;

package/dist/utils/schemaConversion.d.ts

@@ -12,7 +12,12 @@ import type { ZodUnknownSchema } from "../types/tools.js";
  */
 export declare function inlineJsonSchema(schema: Record<string, unknown>, definitions?: Record<string, Record<string, unknown>>, visited?: Set<string>): Record<string, unknown>;
 /**
- * Convert Zod schema to JSON Schema format for Claude AI
+ * Convert Zod schema to JSON Schema format for provider APIs.
+ *
+ * Handles three input types:
+ * 1. Zod schemas (have `_def.typeName`) — converted via zod-to-json-schema
+ * 2. AI SDK `jsonSchema()` wrappers (have `.jsonSchema` property) — extracted directly
+ * 3. Plain JSON Schema objects (have `type`/`properties` but no `_def`) — returned as-is
  */
 export declare function convertZodToJsonSchema(zodSchema: ZodUnknownSchema): object;
 /**