jiva-core 0.3.41 → 0.3.42-dev.b6b238d

package/dist/models/krutrim.js

@@ -1,311 +1,11 @@
 /**
- * Krutrim API Client
+ * @deprecated
+ * This file is kept for backward compatibility only.
+ * Import from './model-client.js' instead.
  *
- * Handles communication with Krutrim Cloud API for both gpt-oss-120b and Llama-4-Maverick-17B models.
- * Implements Harmony format for gpt-oss-120b and standard format for multimodal model.
+ * KrutrimModel  → ModelClient
+ * KrutrimConfig → ModelClientConfig
+ * createKrutrimModel → createModelClient
  */
-import { formatMessagesForHarmony, parseHarmonyResponse } from './harmony.js';
-import { ModelError } from '../utils/errors.js';
-import { logger } from '../utils/logger.js';
-export class KrutrimModel {
-    config;
-    constructor(config) {
-        this.config = config;
-    }
-    supportsVision() {
-        return this.config.type === 'multimodal';
-    }
-    supportsToolCalling() {
-        // Both the reasoning model and dedicated tool-calling model support tool calling
-        return this.config.type === 'reasoning' || this.config.type === 'tool-calling';
-    }
-    async chat(options) {
-        // 'tool-calling' models are treated as reasoning models for tool-call formatting purposes
-        const isReasoningModel = this.config.type === 'reasoning' || this.config.type === 'tool-calling';
-        // Retry logic for transient errors (WAF/rate limiting/server errors)
-        const maxRetries = 4;
-        let lastError = null;
-        for (let attempt = 0; attempt <= maxRetries; attempt++) {
-            if (attempt > 0) {
-                await new Promise(resolve => setTimeout(resolve, this._lastRetryWait ?? 2000));
-            }
-            try {
-                return await this.attemptChat(options, isReasoningModel);
-            }
-            catch (error) {
-                lastError = error;
-                // Retry on transient errors: 403 (WAF), 429 (rate limit), 500/502/503/504 (server errors)
-                if (error instanceof ModelError) {
-                    const is429 = error.message.includes('(429)');
-                    const shouldRetry = error.message.includes('(403)') || // WAF blocking
-                        is429 || // Rate limiting
-                        error.message.includes('(500)') || // Internal server error
-                        error.message.includes('(502)') || // Bad gateway
-                        error.message.includes('(503)') || // Service unavailable
-                        error.message.includes('(504)'); // Gateway timeout
-                    if (!shouldRetry || attempt === maxRetries) {
-                        throw error;
-                    }
-                    // Log the error type for debugging
-                    let errorType = 'Unknown error';
-                    if (error.message.includes('(403)'))
-                        errorType = '403 Access Denied (WAF)';
-                    else if (is429)
-                        errorType = '429 Rate Limited';
-                    else if (error.message.includes('(500)'))
-                        errorType = '500 Internal Server Error';
-                    else if (error.message.includes('(502)'))
-                        errorType = '502 Bad Gateway';
-                    else if (error.message.includes('(503)'))
-                        errorType = '503 Service Unavailable';
-                    else if (error.message.includes('(504)'))
-                        errorType = '504 Gateway Timeout';
-                    // For 429s, parse the actual retry-after time from the error message.
-                    // Groq returns: "Please try again in 8.53s."
-                    if (is429) {
-                        const match = error.message.match(/try again in (\d+(?:\.\d+)?)\s*s/i);
-                        const retryAfterMs = match ? Math.ceil(parseFloat(match[1]) * 1000) + 500 : null;
-                        // Use parsed time, falling back to capped exponential backoff
-                        const exponential = Math.min(Math.pow(2, attempt) * 1000, 30_000);
-                        this._lastRetryWait = retryAfterMs ?? exponential;
-                        logger.warn(`Got ${errorType}, will retry in ${(this._lastRetryWait / 1000).toFixed(1)}s (attempt ${attempt + 1}/${maxRetries + 1})...`);
-                    }
-                    else {
-                        const waitTime = Math.min(Math.pow(2, attempt) * 1000, 30_000);
-                        this._lastRetryWait = waitTime;
-                        logger.warn(`Got ${errorType}, will retry in ${waitTime / 1000}s (attempt ${attempt + 1}/${maxRetries + 1})...`);
-                    }
-                }
-                else {
-                    throw error;
-                }
-            }
-        }
-        throw lastError;
-    }
-    /** Retry wait time in ms, set dynamically based on API hint or exponential backoff. */
-    _lastRetryWait;
-    async attemptChat(options, isReasoningModel) {
-        try {
-            let messages;
-            let tools;
-            const useHarmony = this.config.useHarmonyFormat ?? false;
-            if (isReasoningModel && options.tools && options.tools.length > 0) {
-                if (useHarmony) {
-                    // Convert to Harmony format (Krutrim-specific)
-                    // Tools are embedded in the developer message
-                    tools = options.tools.map(t => ({
-                        name: t.name,
-                        description: t.description,
-                        parameters: t.parameters,
-                    }));
-                    messages = formatMessagesForHarmony(options.messages, tools);
-                }
-                else {
-                    // Standard OpenAI format - tools sent separately
-                    messages = options.messages;
-                }
-            }
-            else {
-                messages = options.messages;
-            }
-            // All OpenAI-compatible APIs support standard roles: system, user, assistant, tool
-            // Convert 'developer' role to 'system' for API compatibility
-            const apiMessages = messages.map((msg) => {
-                if (msg.role === 'developer') {
-                    return { ...msg, role: 'system' };
-                }
-                return msg;
-            });
-            // ── Reasoning effort ────────────────────────────────────────────────────
-            // Per-call value takes precedence over model-level default.
-            const effort = options.reasoningEffort ?? this.config.defaultReasoningEffort;
-            const strategy = this.config.reasoningEffortStrategy ?? 'both';
-            // System-prompt injection: prepend "Reasoning: <level>" as the very first
-            // system message. gpt-oss-120b understands this on any provider — it's the
-            // official fallback for Krutrim/self-hosted endpoints that strip unknown
-            // request body params.
-            if (effort && isReasoningModel && (strategy === 'system_prompt' || strategy === 'both')) {
-                apiMessages.unshift({ role: 'system', content: `Reasoning: ${effort}` });
-            }
-            const requestBody = {
-                model: options.model || this.config.model,
-                messages: apiMessages,
-                temperature: options.temperature ?? 0.2, // Lower default to reduce hallucination
-            };
-            if (options.maxTokens) {
-                requestBody.max_tokens = options.maxTokens;
-            }
-            // API param: native Groq support; silently ignored by providers that don't implement it.
-            if (effort && isReasoningModel && (strategy === 'api_param' || strategy === 'both')) {
-                requestBody.reasoning_effort = effort;
-            }
-            // include_reasoning: request thinking tokens in the response (Groq-specific).
-            // When present, tokens appear in choice.message.reasoning and are logged below.
-            if (this.config.includeReasoning && isReasoningModel) {
-                requestBody.include_reasoning = true;
-            }
-            // ────────────────────────────────────────────────────────────────────────
-            // Send tools in standard OpenAI format if not using Harmony
-            if (!useHarmony && isReasoningModel && options.tools && options.tools.length > 0) {
-                requestBody.tools = options.tools.map(t => ({
-                    type: 'function',
-                    function: {
-                        name: t.name,
-                        description: t.description,
-                        parameters: t.parameters,
-                    },
-                }));
-                requestBody.tool_choice = 'auto';
-            }
-            // Log full request for debugging
-            logger.debug('API Request:', JSON.stringify(requestBody, null, 2));
-            logger.debug('Request size:', JSON.stringify(requestBody).length, 'bytes');
-            logger.debug('Message count:', requestBody.messages.length);
-            const response = await fetch(this.config.endpoint, {
-                method: 'POST',
-                headers: {
-                    'Content-Type': 'application/json',
-                    'Authorization': `Bearer ${this.config.apiKey}`,
-                    'User-Agent': 'Jiva/0.1.0',
-                    'Accept': 'application/json',
-                },
-                body: JSON.stringify(requestBody),
-            });
-            if (!response.ok) {
-                const errorText = await response.text();
-                logger.error(`API Error Response (${response.status}):`, errorText);
-                logger.debug('Request that failed:', JSON.stringify(requestBody, null, 2));
-                // Write failing request to file for debugging
-                if (response.status === 403) {
-                    try {
-                        const fs = await import('fs');
-                        const debugPath = '/tmp/jiva_failed_request.json';
-                        await fs.promises.writeFile(debugPath, JSON.stringify(requestBody, null, 2));
-                        logger.warn(`Failing request saved to: ${debugPath}`);
-                    }
-                    catch (e) {
-                        // Ignore file write errors
-                    }
-                }
-                throw new ModelError(`API error (${response.status}): ${errorText}`, this.config.model);
-            }
-            const data = await response.json();
-            logger.debug('API Response:', JSON.stringify(data, null, 2));
-            if (!data.choices || data.choices.length === 0) {
-                throw new ModelError('No choices in response', this.config.model);
-            }
-            const choice = data.choices[0];
-            const messageContent = choice.message?.content || '';
-            // Log reasoning tokens when include_reasoning=true (Groq returns them in
-            // choice.message.reasoning). Useful for debugging model "thinking" quality.
-            const reasoningTokens = choice.message?.reasoning;
-            if (reasoningTokens) {
-                logger.debug(`[Model reasoning] ${reasoningTokens.substring(0, 2000)}${reasoningTokens.length > 2000 ? '…' : ''}`);
-            }
-            // Parse response based on format used
-            if (useHarmony && isReasoningModel && options.tools && options.tools.length > 0) {
-                // Parse Harmony format response (Krutrim-specific)
-                const parsed = parseHarmonyResponse(messageContent);
-                return {
-                    content: parsed.final || parsed.commentary || messageContent,
-                    toolCalls: parsed.toolCalls.length > 0 ? parsed.toolCalls : undefined,
-                    usage: data.usage ? {
-                        promptTokens: data.usage.prompt_tokens || 0,
-                        completionTokens: data.usage.completion_tokens || 0,
-                        totalTokens: data.usage.total_tokens || 0,
-                    } : undefined,
-                    raw: {
-                        ...data,
-                        parsedHarmony: parsed,
-                    },
-                };
-            }
-            else {
-                // Standard OpenAI format response
-                // Check for tool calls in standard format
-                const toolCalls = choice.message?.tool_calls;
-                return {
-                    content: messageContent,
-                    toolCalls: toolCalls && toolCalls.length > 0 ? toolCalls : undefined,
-                    usage: data.usage ? {
-                        promptTokens: data.usage.prompt_tokens || 0,
-                        completionTokens: data.usage.completion_tokens || 0,
-                        totalTokens: data.usage.total_tokens || 0,
-                    } : undefined,
-                    raw: data,
-                };
-            }
-        }
-        catch (error) {
-            if (error instanceof ModelError) {
-                throw error;
-            }
-            throw new ModelError(`Failed to communicate with API: ${error instanceof Error ? error.message : String(error)}`, this.config.model);
-        }
-    }
-    /**
-     * Helper method for vision tasks using multimodal model
-     */
-    async describeImage(imageUrl, prompt) {
-        if (!this.supportsVision()) {
-            throw new ModelError('This model does not support vision tasks', this.config.model);
-        }
-        const response = await this.chat({
-            model: this.config.model,
-            messages: [
-                {
-                    role: 'user',
-                    content: [
-                        {
-                            type: 'text',
-                            text: prompt || 'Describe this image in detail.',
-                        },
-                        {
-                            type: 'image_url',
-                            image_url: {
-                                url: imageUrl,
-                            },
-                        },
-                    ],
-                },
-            ],
-        });
-        return response.content;
-    }
-    /**
-     * Test connectivity to the API endpoint
-     * Returns true if connection is successful, throws error otherwise
-     */
-    async testConnectivity() {
-        const startTime = Date.now();
-        try {
-            logger.debug(`Testing connectivity to ${this.config.endpoint}...`);
-            // Simple test request with minimal tokens and low reasoning effort (it's just a ping)
-            const testResponse = await this.chat({
-                messages: [{ role: 'user', content: 'test' }],
-                temperature: 0,
-                maxTokens: 5,
-                reasoningEffort: 'low',
-            });
-            const latency = Date.now() - startTime;
-            logger.debug(`Connectivity test passed in ${latency}ms`);
-            return { success: true, latency };
-        }
-        catch (error) {
-            const errorMessage = error instanceof Error ? error.message : String(error);
-            logger.error(`Connectivity test failed: ${errorMessage}`);
-            return {
-                success: false,
-                error: errorMessage,
-            };
-        }
-    }
-}
-/**
- * Factory function to create Krutrim model instances
- */
-export function createKrutrimModel(config) {
-    return new KrutrimModel(config);
-}
+export { ModelClient as KrutrimModel, createModelClient as createKrutrimModel, } from './model-client.js';
 //# sourceMappingURL=krutrim.js.map

package/dist/models/krutrim.js.map

@@ -1 +1 @@
-{"version":3,"file":"krutrim.js","sourceRoot":"","sources":["../../src/models/krutrim.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EACL,wBAAwB,EACxB,oBAAoB,EAGrB,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,UAAU,EAAE,MAAM,oBAAoB,CAAC;AAChD,OAAO,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AAsC5C,MAAM,OAAO,YAAY;IACf,MAAM,CAAgB;IAE9B,YAAY,MAAqB;QAC/B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACvB,CAAC;IAED,cAAc;QACZ,OAAO,IAAI,CAAC,MAAM,CAAC,IAAI,KAAK,YAAY,CAAC;IAC3C,CAAC;IAED,mBAAmB;QACjB,iFAAiF;QACjF,OAAO,IAAI,CAAC,MAAM,CAAC,IAAI,KAAK,WAAW,IAAI,IAAI,CAAC,MAAM,CAAC,IAAI,KAAK,cAAc,CAAC;IACjF,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,OAA8B;QACvC,0FAA0F;QAC1F,MAAM,gBAAgB,GAAG,IAAI,CAAC,MAAM,CAAC,IAAI,KAAK,WAAW,IAAI,IAAI,CAAC,MAAM,CAAC,IAAI,KAAK,cAAc,CAAC;QAEjG,qEAAqE;QACrE,MAAM,UAAU,GAAG,CAAC,CAAC;QACrB,IAAI,SAAS,GAAiB,IAAI,CAAC;QAEnC,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;YACvD,IAAI,OAAO,GAAG,CAAC,EAAE,CAAC;gBAChB,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,IAAI,CAAC,cAAc,IAAI,IAAI,CAAC,CAAC,CAAC;YACjF,CAAC;YAED,IAAI,CAAC;gBACH,OAAO,MAAM,IAAI,CAAC,WAAW,CAAC,OAAO,EAAE,gBAAgB,CAAC,CAAC;YAC3D,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,SAAS,GAAG,KAAc,CAAC;gBAE3B,0FAA0F;gBAC1F,IAAI,KAAK,YAAY,UAAU,EAAE,CAAC;oBAChC,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;oBAC9C,MAAM,WAAW,GACf,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAK,eAAe;wBACnD,KAAK,IAA+B,gBAAgB;wBACpD,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAK,wBAAwB;wBAC5D,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAK,cAAc;wBAClD,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAK,sBAAsB;wBAC1D,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAI,kBAAkB;oBAExD,IAAI,CAAC,WAAW,IAAI,OAAO,KAAK,UAAU,EAAE,CAAC;wBAC3C,MAAM,KAAK,CAAC;oBACd,CAAC;oBAED,mCAAmC;oBACnC,IAAI,SAAS,GAAG,eAAe,CAAC;oBAChC,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;wBAAE,SAAS,GAAG,yBAAyB,CAAC;yBACtE,IAAI,KAAK;wBAAE,SAAS,GAAG,kBAAkB,CAAC;yBAC1C,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;wBAAE,SAAS,GAAG,2BAA2B,CAAC;yBAC7E,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;wBAAE,SAAS,GAAG,iBAAiB,CAAC;yBACnE,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;wBAAE,SAAS,GAAG,yBAAyB,CAAC;yBAC3E,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;wBAAE,SAAS,GAAG,qBAAqB,CAAC;oBAE5E,sEAAsE;oBACtE,6CAA6C;oBAC7C,IAAI,KAAK,EAAE,CAAC;wBACV,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,mCAAmC,CAAC,CAAC;wBACvE,MAAM,YAAY,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC;wBACjF,8DAA8D;wBAC9D,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,GAAG,IAAI,EAAE,MAAM,CAAC,CAAC;wBAClE,IAAI,CAAC,cAAc,GAAG,YAAY,IAAI,WAAW,CAAC;wBAClD,MAAM,CAAC,IAAI,CAAC,OAAO,SAAS,mBAAmB,CAAC,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,OAAO,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,MAAM,CAAC,CAAC;oBAC3I,CAAC;yBAAM,CAAC;wBACN,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,GAAG,IAAI,EAAE,MAAM,CAAC,CAAC;wBAC/D,IAAI,CAAC,cAAc,GAAG,QAAQ,CAAC;wBAC/B,MAAM,CAAC,IAAI,CAAC,OAAO,SAAS,mBAAmB,QAAQ,GAAG,IAAI,cAAc,OAAO,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,MAAM,CAAC,CAAC;oBACnH,CAAC;gBACH,CAAC;qBAAM,CAAC;oBACN,MAAM,KAAK,CAAC;gBACd,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,SAAU,CAAC;IACnB,CAAC;IAED,uFAAuF;IAC/E,cAAc,CAAU;IAExB,KAAK,CAAC,WAAW,CAAC,OAA8B,EAAE,gBAAyB;QACjF,IAAI,CAAC;YACH,IAAI,QAAe,CAAC;YACpB,IAAI,KAA0C,CAAC;YAC/C,MAAM,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,gBAAgB,IAAI,KAAK,CAAC;YAEzD,IAAI,gBAAgB,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAClE,IAAI,UAAU,EAAE,CAAC;oBACf,+CAA+C;oBAC/C,8CAA8C;oBAC9C,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;wBAC9B,IAAI,EAAE,CAAC,CAAC,IAAI;wBACZ,WAAW,EAAE,CAAC,CAAC,WAAW;wBAC1B,UAAU,EAAE,CAAC,CAAC,UAAU;qBACzB,CAAC,CAAC,CAAC;oBAEJ,QAAQ,GAAG,wBAAwB,CACjC,OAAO,CAAC,QAA4B,EACpC,KAAK,CACN,CAAC;gBACJ,CAAC;qBAAM,CAAC;oBACN,iDAAiD;oBACjD,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC;gBAC9B,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC;YAC9B,CAAC;YAED,mFAAmF;YACnF,6DAA6D;YAC7D,MAAM,WAAW,GAAU,QAAQ,CAAC,GAAG,CAAC,CAAC,GAAQ,EAAE,EAAE;gBACnD,IAAI,GAAG,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;oBAC7B,OAAO,EAAE,GAAG,GAAG,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;gBACpC,CAAC;gBACD,OAAO,GAAG,CAAC;YACb,CAAC,CAAC,CAAC;YAEH,2EAA2E;YAC3E,4DAA4D;YAC5D,MAAM,MAAM,GAAG,OAAO,CAAC,eAAe,IAAI,IAAI,CAAC,MAAM,CAAC,sBAAsB,CAAC;YAC7E,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,uBAAuB,IAAI,MAAM,CAAC;YAE/D,0EAA0E;YAC1E,2EAA2E;YAC3E,yEAAyE;YACzE,uBAAuB;YACvB,IAAI,MAAM,IAAI,gBAAgB,IAAI,CAAC,QAAQ,KAAK,eAAe,IAAI,QAAQ,KAAK,MAAM,CAAC,EAAE,CAAC;gBACxF,WAAW,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,cAAc,MAAM,EAAE,EAAE,CAAC,CAAC;YAC3E,CAAC;YAED,MAAM,WAAW,GAAQ;gBACvB,KAAK,EAAE,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,MAAM,CAAC,KAAK;gBACzC,QAAQ,EAAE,WAAW;gBACrB,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,GAAG,EAAE,wCAAwC;aAClF,CAAC;YAEF,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;gBACtB,WAAW,CAAC,UAAU,GAAG,OAAO,CAAC,SAAS,CAAC;YAC7C,CAAC;YAED,yFAAyF;YACzF,IAAI,MAAM,IAAI,gBAAgB,IAAI,CAAC,QAAQ,KAAK,WAAW,IAAI,QAAQ,KAAK,MAAM,CAAC,EAAE,CAAC;gBACpF,WAAW,CAAC,gBAAgB,GAAG,MAAM,CAAC;YACxC,CAAC;YAED,8EAA8E;YAC9E,gFAAgF;YAChF,IAAI,IAAI,CAAC,MAAM,CAAC,gBAAgB,IAAI,gBAAgB,EAAE,CAAC;gBACrD,WAAW,CAAC,iBAAiB,GAAG,IAAI,CAAC;YACvC,CAAC;YACD,2EAA2E;YAE3E,4DAA4D;YAC5D,IAAI,CAAC,UAAU,IAAI,gBAAgB,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACjF,WAAW,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;oBAC1C,IAAI,EAAE,UAAU;oBAChB,QAAQ,EAAE;wBACR,IAAI,EAAE,CAAC,CAAC,IAAI;wBACZ,WAAW,EAAE,CAAC,CAAC,WAAW;wBAC1B,UAAU,EAAE,CAAC,CAAC,UAAU;qBACzB;iBACF,CAAC,CAAC,CAAC;gBACJ,WAAW,CAAC,WAAW,GAAG,MAAM,CAAC;YACnC,CAAC;YAED,iCAAiC;YACjC,MAAM,CAAC,KAAK,CAAC,cAAc,EAAE,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;YACnE,MAAM,CAAC,KAAK,CAAC,eAAe,EAAE,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;YAC3E,MAAM,CAAC,KAAK,CAAC,gBAAgB,EAAE,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;YAE5D,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE;gBACjD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE;oBACP,cAAc,EAAE,kBAAkB;oBAClC,eAAe,EAAE,UAAU,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE;oBAC/C,YAAY,EAAE,YAAY;oBAC1B,QAAQ,EAAE,kBAAkB;iBAC7B;gBACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC;aAClC,CAAC,CAAC;YAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,MAAM,SAAS,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;gBACxC,MAAM,CAAC,KAAK,CAAC,uBAAuB,QAAQ,CAAC,MAAM,IAAI,EAAE,SAAS,CAAC,CAAC;gBACpE,MAAM,CAAC,KAAK,CAAC,sBAAsB,EAAE,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;gBAE3E,8CAA8C;gBAC9C,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;oBAC5B,IAAI,CAAC;wBACH,MAAM,EAAE,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;wBAC9B,MAAM,SAAS,GAAG,+BAA+B,CAAC;wBAClD,MAAM,EAAE,CAAC,QAAQ,CAAC,SAAS,CACzB,SAAS,EACT,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,IAAI,EAAE,CAAC,CAAC,CACrC,CAAC;wBACF,MAAM,CAAC,IAAI,CAAC,6BAA6B,SAAS,EAAE,CAAC,CAAC;oBACxD,CAAC;oBAAC,OAAO,CAAC,EAAE,CAAC;wBACX,2BAA2B;oBAC7B,CAAC;gBACH,CAAC;gBAED,MAAM,IAAI,UAAU,CAClB,cAAc,QAAQ,CAAC,MAAM,MAAM,SAAS,EAAE,EAC9C,IAAI,CAAC,MAAM,CAAC,KAAK,CAClB,CAAC;YACJ,CAAC;YAED,MAAM,IAAI,GAAQ,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;YACxC,MAAM,CAAC,KAAK,CAAC,eAAe,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;YAE7D,IAAI,CAAC,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAC/C,MAAM,IAAI,UAAU,CAAC,wBAAwB,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACpE,CAAC;YAED,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAC/B,MAAM,cAAc,GAAG,MAAM,CAAC,OAAO,EAAE,OAAO,IAAI,EAAE,CAAC;YAErD,yEAAyE;YACzE,4EAA4E;YAC5E,MAAM,eAAe,GAAuB,MAAM,CAAC,OAAO,EAAE,SAAS,CAAC;YACtE,IAAI,eAAe,EAAE,CAAC;gBACpB,MAAM,CAAC,KAAK,CAAC,qBAAqB,eAAe,CAAC,SAAS,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,eAAe,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACrH,CAAC;YAED,sCAAsC;YACtC,IAAI,UAAU,IAAI,gBAAgB,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAChF,mDAAmD;gBACnD,MAAM,MAAM,GAAG,oBAAoB,CAAC,cAAc,CAAC,CAAC;gBAEpD,OAAO;oBACL,OAAO,EAAE,MAAM,CAAC,KAAK,IAAI,MAAM,CAAC,UAAU,IAAI,cAAc;oBAC5D,SAAS,EAAE,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;oBACrE,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;wBAClB,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa,IAAI,CAAC;wBAC3C,gBAAgB,EAAE,IAAI,CAAC,KAAK,CAAC,iBAAiB,IAAI,CAAC;wBACnD,WAAW,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY,IAAI,CAAC;qBAC1C,CAAC,CAAC,CAAC,SAAS;oBACb,GAAG,EAAE;wBACH,GAAG,IAAI;wBACP,aAAa,EAAE,MAAM;qBACtB;iBACF,CAAC;YACJ,CAAC;iBAAM,CAAC;gBACN,kCAAkC;gBAClC,0CAA0C;gBAC1C,MAAM,SAAS,GAAG,MAAM,CAAC,OAAO,EAAE,UAAU,CAAC;gBAE7C,OAAO;oBACL,OAAO,EAAE,cAAc;oBACvB,SAAS,EAAE,SAAS,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;oBACpE,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;wBAClB,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa,IAAI,CAAC;wBAC3C,gBAAgB,EAAE,IAAI,CAAC,KAAK,CAAC,iBAAiB,IAAI,CAAC;wBACnD,WAAW,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY,IAAI,CAAC;qBAC1C,CAAC,CAAC,CAAC,SAAS;oBACb,GAAG,EAAE,IAAI;iBACV,CAAC;YACJ,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,KAAK,YAAY,UAAU,EAAE,CAAC;gBAChC,MAAM,KAAK,CAAC;YACd,CAAC;YAED,MAAM,IAAI,UAAU,CAClB,mCAAmC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAC3F,IAAI,CAAC,MAAM,CAAC,KAAK,CAClB,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,aAAa,CAAC,QAAgB,EAAE,MAAe;QACnD,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,EAAE,CAAC;YAC3B,MAAM,IAAI,UAAU,CAClB,0CAA0C,EAC1C,IAAI,CAAC,MAAM,CAAC,KAAK,CAClB,CAAC;QACJ,CAAC;QAED,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC;YAC/B,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK;YACxB,QAAQ,EAAE;gBACR;oBACE,IAAI,EAAE,MAAM;oBACZ,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EAAE,MAAM,IAAI,gCAAgC;yBACjD;wBACD;4BACE,IAAI,EAAE,WAAW;4BACjB,SAAS,EAAE;gCACT,GAAG,EAAE,QAAQ;6BACd;yBACF;qBACF;iBACF;aACF;SACF,CAAC,CAAC;QAEH,OAAO,QAAQ,CAAC,OAAO,CAAC;IAC1B,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,gBAAgB;QACpB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAE7B,IAAI,CAAC;YACH,MAAM,CAAC,KAAK,CAAC,2BAA2B,IAAI,CAAC,MAAM,CAAC,QAAQ,KAAK,CAAC,CAAC;YAEnE,sFAAsF;YACtF,MAAM,YAAY,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC;gBACnC,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;gBAC7C,WAAW,EAAE,CAAC;gBACd,SAAS,EAAE,CAAC;gBACZ,eAAe,EAAE,KAAK;aACvB,CAAC,CAAC;YAEH,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;YACvC,MAAM,CAAC,KAAK,CAAC,+BAA+B,OAAO,IAAI,CAAC,CAAC;YAEzD,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;QACpC,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,YAAY,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YAC5E,MAAM,CAAC,KAAK,CAAC,6BAA6B,YAAY,EAAE,CAAC,CAAC;YAE1D,OAAO;gBACL,OAAO,EAAE,KAAK;gBACd,KAAK,EAAE,YAAY;aACpB,CAAC;QACJ,CAAC;IACH,CAAC;CACF;AAED;;GAEG;AACH,MAAM,UAAU,kBAAkB,CAAC,MAAqB;IACtD,OAAO,IAAI,YAAY,CAAC,MAAM,CAAC,CAAC;AAClC,CAAC"}
+{"version":3,"file":"krutrim.js","sourceRoot":"","sources":["../../src/models/krutrim.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,OAAO,EACL,WAAW,IAAI,YAAY,EAC3B,iBAAiB,IAAI,kBAAkB,GACxC,MAAM,mBAAmB,CAAC"}

package/dist/models/model-client.d.ts

@@ -0,0 +1,95 @@
+/**
+ * OpenAI-Compatible Model Client
+ *
+ * Generic client for any OpenAI-compatible chat completion API.
+ * Supports: Krutrim, Groq, Sarvam, OpenAI, and any other provider
+ * that exposes a /v1/chat/completions endpoint.
+ *
+ * Provider-specific behaviour is controlled entirely through config flags:
+ *   - useHarmonyFormat    → Krutrim gpt-oss-120b (Harmony tool format)
+ *   - reasoningEffortStrategy → how reasoning effort is communicated
+ *   - defaultMaxTokens   → required for reasoning models (e.g. Sarvam-105B)
+ */
+import { IModel, ChatCompletionOptions, ModelResponse } from './base.js';
+export interface ModelClientConfig {
+    endpoint: string;
+    apiKey: string;
+    model: string;
+    defaultModel?: string;
+    type: 'reasoning' | 'multimodal' | 'tool-calling';
+    /**
+     * Use Harmony format for tool calling.
+     * Only needed for Krutrim's gpt-oss-120b model.
+     * Default: false (standard OpenAI format)
+     */
+    useHarmonyFormat?: boolean;
+    /**
+     * Default reasoning effort for all calls made through this model instance.
+     * Can be overridden per-call via ChatCompletionOptions.reasoningEffort.
+     * Sensible defaults: 'high' for reasoning models, 'medium' for tool-calling models.
+     */
+    defaultReasoningEffort?: 'low' | 'medium' | 'high';
+    /**
+     * How to communicate reasoning effort to the model.
+     *
+     * 'api_param'     — send only as the `reasoning_effort` request body field.
+     *                   Works natively on Groq and Sarvam.
+     * 'system_prompt' — inject only as a leading system message: "Reasoning: <level>".
+     *                   Required for gpt-oss-120b on Krutrim (strips unknown params).
+     * 'both'          — do both (safe default for unknown providers).
+     *
+     * Provider recommendations:
+     *   Krutrim  → 'system_prompt'
+     *   Groq     → 'api_param'
+     *   Sarvam   → 'api_param'
+     *   Other    → 'both'
+     *
+     * Default: 'both'
+     */
+    reasoningEffortStrategy?: 'api_param' | 'system_prompt' | 'both';
+    /**
+     * Request reasoning tokens in the response body (Groq-specific: `include_reasoning`).
+     * When true, the model's thinking tokens are logged at debug level.
+     * Has no effect on providers that do not support the field.
+     * Note: Sarvam always returns reasoning_content regardless of this flag.
+     * Default: false
+     */
+    includeReasoning?: boolean;
+    /**
+     * Default max tokens for completion.
+     *
+     * REQUIRED for reasoning models like Sarvam-105B that spend completion tokens
+     * on their thinking chain before producing output. Without a sufficient budget
+     * the model exhausts the default limit (~2048) reasoning and returns empty content.
+     *
+     * Recommended: 8192 for Sarvam-105B, unset for Groq/Krutrim.
+     */
+    defaultMaxTokens?: number;
+}
+export declare class ModelClient implements IModel {
+    private config;
+    constructor(config: ModelClientConfig);
+    supportsVision(): boolean;
+    supportsToolCalling(): boolean;
+    chat(options: ChatCompletionOptions): Promise<ModelResponse>;
+    /** Retry wait time in ms, set dynamically based on API hint or exponential backoff. */
+    private _lastRetryWait?;
+    private attemptChat;
+    /**
+     * Helper method for vision tasks using multimodal model
+     */
+    describeImage(imageUrl: string, prompt?: string): Promise<string>;
+    /**
+     * Test connectivity to the API endpoint
+     */
+    testConnectivity(): Promise<{
+        success: boolean;
+        error?: string;
+        latency?: number;
+    }>;
+}
+/**
+ * Factory function to create a model client instance
+ */
+export declare function createModelClient(config: ModelClientConfig): ModelClient;
+//# sourceMappingURL=model-client.d.ts.map

package/dist/models/model-client.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-client.d.ts","sourceRoot":"","sources":["../../src/models/model-client.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,MAAM,EAAE,qBAAqB,EAAE,aAAa,EAAW,MAAM,WAAW,CAAC;AAUlF,MAAM,WAAW,iBAAiB;IAChC,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,WAAW,GAAG,YAAY,GAAG,cAAc,CAAC;IAElD;;;;OAIG;IACH,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAE3B;;;;OAIG;IACH,sBAAsB,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IAEnD;;;;;;;;;;;;;;;;OAgBG;IACH,uBAAuB,CAAC,EAAE,WAAW,GAAG,eAAe,GAAG,MAAM,CAAC;IAEjE;;;;;;OAMG;IACH,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAE3B;;;;;;;;OAQG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED,qBAAa,WAAY,YAAW,MAAM;IACxC,OAAO,CAAC,MAAM,CAAoB;gBAEtB,MAAM,EAAE,iBAAiB;IAIrC,cAAc,IAAI,OAAO;IAIzB,mBAAmB,IAAI,OAAO;IAKxB,IAAI,CAAC,OAAO,EAAE,qBAAqB,GAAG,OAAO,CAAC,aAAa,CAAC;IAiElE,uFAAuF;IACvF,OAAO,CAAC,cAAc,CAAC,CAAS;YAElB,WAAW;IAiMzB;;OAEG;IACG,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IAgCvE;;OAEG;IACG,gBAAgB,IAAI,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CA2B1F;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,iBAAiB,GAAG,WAAW,CAExE"}