@juspay/neurolink 9.68.19 → 9.68.20

package/dist/factories/providerRegistry.js

@@ -112,10 +112,10 @@ export class ProviderRegistry {
                 return new MistralProvider(modelName, sdk, undefined, mistralCreds);
             }, MistralModels.MISTRAL_LARGE_LATEST, ["mistral"]);
             // Register Ollama provider
-            ProviderFactory.registerProvider(AIProviderName.OLLAMA, async (modelName, _providerName, _sdk, _region, credentials) => {
+            ProviderFactory.registerProvider(AIProviderName.OLLAMA, async (modelName, _providerName, sdk, _region, credentials) => {
                 const ollamaCreds = credentials;
                 const { OllamaProvider } = await import("../providers/ollama.js");
-                return new OllamaProvider(modelName, ollamaCreds);
+                return new OllamaProvider(modelName, sdk, undefined, ollamaCreds);
             }, process.env.OLLAMA_MODEL || OllamaModels.LLAMA3_2_LATEST, ["ollama", "local"]);
             // Register LiteLLM provider
             ProviderFactory.registerProvider(AIProviderName.LITELLM, async (modelName, _providerName, sdk, _region, credentials) => {

package/dist/lib/factories/providerRegistry.js

@@ -112,10 +112,10 @@ export class ProviderRegistry {
                 return new MistralProvider(modelName, sdk, undefined, mistralCreds);
             }, MistralModels.MISTRAL_LARGE_LATEST, ["mistral"]);
             // Register Ollama provider
-            ProviderFactory.registerProvider(AIProviderName.OLLAMA, async (modelName, _providerName, _sdk, _region, credentials) => {
+            ProviderFactory.registerProvider(AIProviderName.OLLAMA, async (modelName, _providerName, sdk, _region, credentials) => {
                 const ollamaCreds = credentials;
                 const { OllamaProvider } = await import("../providers/ollama.js");
-                return new OllamaProvider(modelName, ollamaCreds);
+                return new OllamaProvider(modelName, sdk, undefined, ollamaCreds);
             }, process.env.OLLAMA_MODEL || OllamaModels.LLAMA3_2_LATEST, ["ollama", "local"]);
             // Register LiteLLM provider
             ProviderFactory.registerProvider(AIProviderName.LITELLM, async (modelName, _providerName, sdk, _region, credentials) => {

package/dist/lib/providers/ollama.d.ts

@@ -1,149 +1,71 @@
 import type { AIProviderName } from "../constants/enums.js";
-import { BaseProvider } from "../core/baseProvider.js";
-import type { StreamOptions, StreamResult, ZodUnknownSchema } from "../types/index.js";
-import type { LanguageModel, Schema } from "../types/index.js";
+import type { NeurolinkCredentials, StreamOptions, TextGenerationOptions } from "../types/index.js";
+import { OpenAIChatCompletionsProvider } from "./openaiChatCompletionsBase.js";
 /**
- * Ollama Provider v2 - BaseProvider Implementation
+ * Ollama Provider — direct HTTP, no AI SDK.
  *
- * PHASE 3.7: BaseProvider wrap around existing custom Ollama implementation
+ * Wraps a local (or remote/Cloud) Ollama server via its OpenAI-compatible
+ * `/v1` API. All request / stream / multi-step tool-loop orchestration lives
+ * in `OpenAIChatCompletionsProvider`; this class declares configuration plus
+ * the Ollama-specific behaviour:
  *
- * Features:
- * - Extends BaseProvider for shared functionality
- * - Preserves custom OllamaLanguageModel implementation
- * - Local model management and health checking
- * - Enhanced error handling with Ollama-specific guidance
+ *   1. `/v1` base-URL normalization (accepts a bare `OLLAMA_BASE_URL` host).
+ *   2. No-auth-by-default with an optional `OLLAMA_API_KEY` for Ollama Cloud.
+ *   3. Configurable per-model tool gating via `OLLAMA_TOOL_CAPABLE_MODELS` /
+ *      `modelConfig` (`supportsTools`).
+ *   4. Elevated request timeout for slow large local models (5-minute base
+ *      default, overridable via `OLLAMA_TIMEOUT`).
+ *   5. Native `/v1/embeddings` (`embed` / `embedMany`).
+ *   6. Rich, actionable error mapping (`ollama serve` / `ollama pull` hints).
+ *
+ * Model discovery uses the base's `/v1/models` probe (Ollama supports it).
+ *
+ * @see https://docs.ollama.com/api/openai-compatibility
  */
-export declare class OllamaProvider extends BaseProvider {
-    private ollamaModel;
-    private baseUrl;
-    private timeout;
-    constructor(modelName?: string, credentials?: {
-        baseURL?: string;
-    });
+export declare class OllamaProvider extends OpenAIChatCompletionsProvider {
+    constructor(modelName?: string, sdk?: unknown, _region?: string, credentials?: NeurolinkCredentials["ollama"]);
     protected getProviderName(): AIProviderName;
     protected getDefaultModel(): string;
-    /**
-     * Returns the Vercel AI SDK model instance for Ollama.
-     *
-     * OllamaLanguageModel implements OllamaAsLanguageModel which is structurally
-     * compatible with LanguageModelV2 (specificationVersion "v2", modelId, provider,
-     * supportedUrls, doGenerate, doStream).
-     */
-    protected getAISDKModel(): LanguageModel;
-    /**
-     * Ollama Tool Calling Support (Enhanced 2025)
-     *
-     * Uses configurable model list from ModelConfiguration instead of hardcoded values.
-     * Tool-capable models can be configured via OLLAMA_TOOL_CAPABLE_MODELS environment variable.
-     *
-     * **Configuration Options:**
-     * - Environment variable: OLLAMA_TOOL_CAPABLE_MODELS (comma-separated list)
-     * - Configuration file: providers.ollama.modelBehavior.toolCapableModels
-     * - Fallback: Default list of known tool-capable models
-     *
-     * **Implementation Features:**
-     * - Direct Ollama API integration (/v1/chat/completions)
-     * - Automatic tool schema conversion to Ollama format
-     * - Streaming tool calls with incremental response parsing
-     * - Model compatibility validation and fallback handling
-     *
-     * @returns true for supported models, false for unsupported models
-     */
-    supportsTools(): boolean;
-    /**
-     * Extract images from multimodal messages for Ollama API
-     * Returns array of base64-encoded images
-     */
-    private extractImagesFromMessages;
-    /**
-     * Convert multimodal messages to Ollama chat format
-     * Extracts text content and handles images separately
-     */
-    private convertToOllamaMessages;
-    protected executeStream(options: StreamOptions, analysisSchema?: ZodUnknownSchema | Schema<unknown>): Promise<StreamResult>;
-    /**
-     * Execute streaming with Ollama's function calling support
-     * Uses conversation loop to handle multi-step tool execution
-     */
-    private executeStreamWithTools;
-    /**
-     * Execute streaming without tools using the generate API
-     * Fallback for non-tool scenarios or when chat API is unavailable
-     */
-    private executeStreamWithoutTools;
-    /**
-     * Convert AI SDK tools format to Ollama's function calling format
-     */
-    private convertToolsToOllamaFormat;
-    /**
-     * Parse tool calls from Ollama API response
-     */
-    private parseToolCalls;
-    /**
-     * Process Ollama streaming response and stream content to controller
-     * Returns aggregated content, tool calls, and finish reason
-     */
-    private processOllamaResponse;
-    /**
-     * Process individual stream data chunk from Ollama
-     */
-    private processOllamaStreamData;
-    /**
-     * Create stream generator for Ollama chat API with tool call support
-     */
-    private createOllamaChatStream;
-    /**
-     * Format tool calls for display when tools aren't executed directly
-     */
-    private formatToolCallForDisplay;
-    /**
-     * Convert AI SDK tools to ToolDefinition format
-     */
-    private convertAISDKToolsToToolDefinitions;
-    /**
-     * Execute a single tool and return the result
-     */
-    private executeSingleTool;
-    /**
-     * Execute tools and format results for Ollama API
-     * Similar to Bedrock's executeStreamTools but for Ollama format
-     */
-    private executeOllamaTools;
-    /**
-     * Convert ReadableStream to AsyncIterable for compatibility with StreamResult interface
-     */
-    private convertToAsyncIterable;
-    /**
-     * Create stream generator for Ollama generate API (non-tool mode)
-     */
-    private createOllamaStream;
-    private createOpenAIStream;
+    protected getFallbackModelName(): string;
     protected formatProviderError(error: unknown): Error;
     /**
-     * Check if Ollama service is healthy and accessible
+     * Ollama proxies many local models with varying tool support. When
+     * `OLLAMA_TOOL_CAPABLE_MODELS` (or `modelConfig`'s
+     * `modelBehavior.toolCapableModels`) is configured, gate tools on a
+     * substring match against the current model; with no list configured,
+     * assume tools are supported (don't disable on absent evidence).
      */
-    private checkOllamaHealth;
+    supportsTools(): boolean;
     /**
-     * Get available models from Ollama
-     */
-    getAvailableModels(): Promise<string[]>;
+     * Local models are slow; the base already defaults Ollama to a 5-minute
+     * timeout and honors a per-call `options.timeout`. Preserve the legacy
+     * `OLLAMA_TIMEOUT` env override for callers who relied on it, applied only
+     * when no explicit per-call timeout is set. Parsed with the shared
+     * `parseTimeout` so both millisecond numbers ("240000") and duration strings
+     * ("4m", "30s") work; a malformed value is ignored in favour of the default.
+     */
+    getTimeout(options: TextGenerationOptions | StreamOptions): number;
+    /**
+     * Health check: probe `/v1/models` and require at least one installed model.
+     * A reachable server with zero models would let `resolveModelName()` fall
+     * back to a model that the first real request can't serve, so report unusable.
+     */
+    validateConfiguration(): Promise<boolean>;
+    getConfiguration(): {
+        provider: AIProviderName;
+        model: string;
+        defaultModel: string;
+        baseURL: string;
+    };
     /**
-     * Check if a specific model is available
+     * Generate an embedding for a single text input via native /v1/embeddings.
+     * Uses `OLLAMA_EMBEDDING_MODEL` (default `nomic-embed-text`); the embedding
+     * model must be pulled locally (`ollama pull nomic-embed-text`).
      */
-    isModelAvailable(modelName: string): Promise<boolean>;
+    embed(text: string, modelName?: string): Promise<number[]>;
     /**
-     * Get recommendations for tool-calling capable Ollama models
-     * Provides guidance for users who want to use function calling locally
+     * Generate embeddings for multiple text inputs via native /v1/embeddings.
      */
-    static getToolCallingRecommendations(): {
-        recommended: string[];
-        performance: Record<string, {
-            speed: number;
-            quality: number;
-            size: string;
-        }>;
-        notes: Record<string, string>;
-        installation: Record<string, string>;
-    };
+    embedMany(texts: string[], modelName?: string): Promise<number[][]>;
+    private callEmbeddings;
 }
-export default OllamaProvider;