@xapp/stentor-service-generative-ai

Generative AI service for LLM-backed features: analyzing leads, inbound emails, chat transcripts, and websites, plus summarizing conversations. It defines a common LLMService interface with implementations for Anthropic's Claude API (ClaudeService), AWS Bedrock (AWSBedrockService), and OpenAI (OpenAIService), plus a FallbackLLMService that retries against a secondary provider when the primary returns a rate-limit/throttling error.

Usage

import { GenerativeAIService, ClaudeService } from "@xapp/stentor-service-generative-ai";

const llmService = new ClaudeService({
    keyGetter: () => process.env.ANTHROPIC_API_KEY || "",
    // Required. There is no default model: providers retire versions on a schedule and a
    // hardcoded fallback outlives them silently.
    model: "claude-haiku-4-5-20251001"
});

const generativeAIService = new GenerativeAIService(llmService);

const leadAnalysis = await generativeAIService.analyzeLead(lead, {
    businessDescription: "A local plumbing company"
});

Required Environment Variables

None are required. API keys and AWS credentials are passed to each LLMService constructor (keyGetter for ClaudeService/OpenAIService; optional region, defaulting to us-east-1, and credentials for AWSBedrockService, which otherwise uses the AWS SDK default credential chain), so the caller decides where they come from.

• STUDIO_APP_ID - optional; when lead and transcript analysis render a chat transcript into the prompt, messages from this id (or "bot") are labelled as the agent

Documentation

• Anthropic API Docs
• OpenAI API Docs
• AWS Bedrock Docs