Momento/memento-note/lib/ai/providers/custom-openai.ts

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

export class CustomOpenAIProvider implements AIProvider {
  private model: any;
  private embeddingModel: any;
  private apiKey: string;
  private baseUrl: string;

  constructor(
    apiKey: string,
    baseUrl: string,
    modelName: string = 'gpt-4o-mini',
    embeddingModelName: string = 'text-embedding-3-small'
  ) {
    this.apiKey = apiKey;
    this.baseUrl = baseUrl.endsWith('/') ? baseUrl.slice(0, -1) : baseUrl;
    // Create OpenAI-compatible client with custom base URL
    // Use .chat() to force /chat/completions endpoint (avoids Responses API)
    const customClient = createOpenAI({
      baseURL: baseUrl,
      apiKey: apiKey,
      fetch: async (url, options) => {
        const headers = new Headers(options?.headers);
        headers.set('HTTP-Referer', 'https://localhost:3000');
        headers.set('X-Title', 'Memento AI');
        return fetch(url, { ...options, headers });
      }
    });

    this.model = customClient.chat(modelName);
    this.embeddingModel = customClient.embedding(embeddingModelName);
  }

  async generateTags(content: string): Promise<TagSuggestion[]> {
    try {
      const { object } = await generateObject({
        model: this.model,
        schema: z.object({
          tags: z.array(z.object({
            tag: z.string().describe('Short tag name in lowercase'),
            confidence: z.number().min(0).max(1).describe('Confidence level between 0 and 1')
          }))
        }),
        prompt: `Analyze the following note and suggest 1 to 5 relevant tags.
        Note content: "${content}"`,
      });

      return object.tags;
    } catch (e) {
      console.error('Error generating tags (Custom OpenAI):', e);
      return [];
    }
  }

  async getEmbeddings(text: string): Promise<number[]> {
    try {
      const { embedding } = await embed({
        model: this.embeddingModel,
        value: text,
      });
      return embedding;
    } catch (e) {
      console.error('Error generating embeddings (Custom OpenAI):', e);
      return [];
    }
  }

  async generateTitles(prompt: string): Promise<TitleSuggestion[]> {
    try {
      const { object } = await generateObject({
        model: this.model,
        schema: z.object({
          titles: z.array(z.object({
            title: z.string().describe('Suggested title'),
            confidence: z.number().min(0).max(1).describe('Confidence level between 0 and 1')
          }))
        }),
        prompt: prompt,
      });

      return object.titles;
    } catch (e) {
      console.error('Error generating titles (Custom OpenAI):', e);
      return [];
    }
  }

  async generateText(prompt: string): Promise<string> {
    try {
      const { text } = await aiGenerateText({
        model: this.model,
        prompt: prompt,
      });

      return text.trim();
    } catch (e) {
      console.error('Error generating text (Custom OpenAI):', e);
      throw e;
    }
  }

  async chat(messages: any[], systemPrompt?: string): Promise<any> {
    try {
      const { text } = await aiGenerateText({
        model: this.model,
        system: systemPrompt,
        messages: messages,
      });

      return { text: text.trim() };
    } catch (e) {
      console.error('Error in chat (Custom OpenAI):', e);
      throw e;
    }
  }

  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 })) || []
      })) || []
    }
  }

  getModel() {
    return this.model;
  }
}