Keep/keep-notes/lib/ai/providers/ollama.ts

import { createOpenAI } from '@ai-sdk/openai';
import { generateText as aiGenerateText, stepCountIs } from 'ai';
import { AIProvider, TagSuggestion, TitleSuggestion, ToolUseOptions, ToolCallResult } from '../types';

export class OllamaProvider implements AIProvider {
  private baseUrl: string;
  private modelName: string;
  private embeddingModelName: string;
  private model: any;

  constructor(baseUrl: string, modelName: string = 'llama3', embeddingModelName?: string) {
    if (!baseUrl) {
      throw new Error('baseUrl is required for OllamaProvider')
    }
    // Ensure baseUrl ends with /api for Ollama API
    this.baseUrl = baseUrl.endsWith('/api') ? baseUrl : `${baseUrl}/api`;
    this.modelName = modelName;
    this.embeddingModelName = embeddingModelName || modelName;

    // Create OpenAI-compatible model for streaming support
    // Ollama exposes /v1/chat/completions which is compatible with the OpenAI SDK
    const cleanUrl = this.baseUrl.replace(/\/api$/, '');
    const ollamaClient = createOpenAI({
      baseURL: `${cleanUrl}/v1`,
      apiKey: 'ollama',
    });
    this.model = ollamaClient.chat(modelName);
  }

  async generateTags(content: string, language: string = "en"): Promise<TagSuggestion[]> {
    try {
      const promptText = language === 'fa' 
        ? `متن زیر را تحلیل کن و مفاهیم کلیدی را به عنوان برچسب استخراج کن (حداکثر ۱-۳ کلمه).
قوانین:
- کلمات ربط را حذف کن.
- عبارات ترکیبی را حفظ کن.
- حداکثر ۵ برچسب.
پاسخ فقط به صورت لیست JSON با فرمت [{"tag": "string", "confidence": number}]
متن: "${content}"`
        : language === 'fr'
        ? `Analyse la note suivante et extrais les concepts clés sous forme de tags courts (1-3 mots max).
Règles:
- Pas de mots de liaison.
- Garde les expressions composées ensemble.
- Normalise en minuscules sauf noms propres.
- Maximum 5 tags.
Réponds UNIQUEMENT sous forme de liste JSON d'objets : [{"tag": "string", "confidence": number}].
Contenu de la note: "${content}"`
        : `Analyze the following note and extract key concepts as short tags (1-3 words max).
Rules:
- No stop words.
- Keep compound expressions together.
- Lowercase unless proper noun.
- Max 5 tags.
Respond ONLY as a JSON list of objects: [{"tag": "string", "confidence": number}].
Note content: "${content}"`;

      const response = await fetch(`${this.baseUrl}/generate`, {
        method: 'POST',
        headers: { 'Content-Type': 'application/json' },
        body: JSON.stringify({
          model: this.modelName,
          prompt: promptText,
          stream: false,
        }),
      });

      if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);

      const data = await response.json();
      const text = data.response;

      const jsonMatch = text.match(/\[\s*\{[\s\S]*\}\s*\]/);
      if (jsonMatch) {
        return JSON.parse(jsonMatch[0]);
      }

      // Support pour le format { "tags": [...] }
      const objectMatch = text.match(/\{\s*"tags"\s*:\s*(\[[\s\S]*\])\s*\}/);
      if (objectMatch && objectMatch[1]) {
        return JSON.parse(objectMatch[1]);
      }
      
      return [];
    } catch (e) {
      console.error('Erreur API directe Ollama:', e);
      return [];
    }
  }

  async getEmbeddings(text: string): Promise<number[]> {
    try {
      const response = await fetch(`${this.baseUrl}/embeddings`, {
        method: 'POST',
        headers: { 'Content-Type': 'application/json' },
        body: JSON.stringify({
          model: this.embeddingModelName,
          prompt: text,
        }),
      });

      if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);

      const data = await response.json();
      return data.embedding;
    } catch (e) {
      console.error('Erreur embeddings directs Ollama:', e);
      return [];
    }
  }

  async generateTitles(prompt: string): Promise<TitleSuggestion[]> {
    try {
      const response = await fetch(`${this.baseUrl}/generate`, {
        method: 'POST',
        headers: { 'Content-Type': 'application/json' },
        body: JSON.stringify({
          model: this.modelName,
          prompt: `${prompt}\n\nRéponds UNIQUEMENT sous forme de tableau JSON : [{"title": "string", "confidence": number}]`,
          stream: false,
        }),
      });

      if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);

      const data = await response.json();
      const text = data.response;

      // Extraire le JSON de la réponse
      const jsonMatch = text.match(/\[\s*\{[\s\S]*\}\s*\]/);
      if (jsonMatch) {
        return JSON.parse(jsonMatch[0]);
      }

      return [];
    } catch (e) {
      console.error('Erreur génération titres Ollama:', e);
      return [];
    }
  }

  async generateText(prompt: string): Promise<string> {
    try {
      const response = await fetch(`${this.baseUrl}/generate`, {
        method: 'POST',
        headers: { 'Content-Type': 'application/json' },
        body: JSON.stringify({
          model: this.modelName,
          prompt: prompt,
          stream: false,
        }),
      });

      if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);

      const data = await response.json();
      return data.response.trim();
    } catch (e) {
      console.error('Erreur génération texte Ollama:', e);
      throw e;
    }
  }

  async chat(messages: any[], systemPrompt?: string): Promise<any> {
    try {
      const ollamaMessages = messages.map(m => ({
        role: m.role,
        content: m.content
      }));

      if (systemPrompt) {
        ollamaMessages.unshift({ role: 'system', content: systemPrompt });
      }

      const response = await fetch(`${this.baseUrl}/chat`, {
        method: 'POST',
        headers: { 'Content-Type': 'application/json' },
        body: JSON.stringify({
          model: this.modelName,
          messages: ollamaMessages,
          stream: false,
        }),
      });

      if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);

      const data = await response.json();
      return { text: data.message?.content?.trim() || '' };
    } catch (e) {
      console.error('Erreur chat Ollama:', e);
      throw e;
    }
  }

  getModel() {
    return this.model;
  }

  async generateWithTools(options: ToolUseOptions): Promise<ToolCallResult> {
    const { tools, maxSteps = 10, systemPrompt, messages, prompt } = options
    const opts: Record<string, any> = {
      model: this.model,
      tools,
      stopWhen: stepCountIs(maxSteps),
    }
    if (systemPrompt) opts.system = systemPrompt
    if (messages) opts.messages = messages
    else if (prompt) opts.prompt = prompt

    const result = await aiGenerateText(opts as any)
    return {
      toolCalls: result.toolCalls?.map((tc: any) => ({ toolName: tc.toolName, input: tc.input })) || [],
      toolResults: result.toolResults?.map((tr: any) => ({ toolName: tr.toolName, input: tr.input, output: tr.output })) || [],
      text: result.text,
      steps: result.steps?.map((step: any) => ({
        text: step.text,
        toolCalls: step.toolCalls?.map((tc: any) => ({ toolName: tc.toolName, input: tc.input })) || [],
        toolResults: step.toolResults?.map((tr: any) => ({ toolName: tr.toolName, input: tr.input, output: tr.output })) || []
      })) || []
    }
  }
}