@renderify/llm

LLM provider implementations and registry for Renderify.

@renderify/llm provides built-in interpreters for OpenAI, Anthropic, Google, Ollama, and LM Studio, plus a provider registry API for custom providers.

Install

pnpm add @renderify/llm
# or
npm i @renderify/llm

Built-in Providers

• OpenAILLMInterpreter (provider: "openai")
• AnthropicLLMInterpreter (provider: "anthropic")
• GoogleLLMInterpreter (provider: "google")
• OllamaLLMInterpreter (provider: "ollama")
• LMStudioLLMInterpreter (provider: "lmstudio")

Factory API

• createLLMInterpreter({ provider, providerOptions })
• LLMProviderRegistry
• createDefaultLLMProviderRegistry()

Quick Start

import { createLLMInterpreter } from "@renderify/llm";

const llm = createLLMInterpreter({
  provider: "openai",
  providerOptions: {
    apiKey: process.env.RENDERIFY_LLM_API_KEY,
    model: "gpt-4o-mini",
  },
});

const response = await llm.generateResponse({
  prompt: "return a simple RuntimePlan JSON",
  context: {},
});

console.log(response.text);

Local Models Example

import { createLLMInterpreter } from "@renderify/llm";

const ollama = createLLMInterpreter({
  provider: "ollama",
  providerOptions: {
    baseUrl: "http://127.0.0.1:11434",
    model: "qwen2.5-coder:7b",
  },
});

const lmstudio = createLLMInterpreter({
  provider: "lmstudio",
  providerOptions: {
    baseUrl: "http://127.0.0.1:1234/v1",
    model: "qwen2.5-coder-7b-instruct",
  },
});

Reliability Controls

Each provider supports request reliability options through providerOptions.reliability:

const llm = createLLMInterpreter({
  provider: "openai",
  providerOptions: {
    apiKey: process.env.RENDERIFY_LLM_API_KEY,
    reliability: {
      maxRetries: 2,
      retryBaseDelayMs: 250,
      retryMaxDelayMs: 2000,
      retryJitterMs: 50,
      retryStatusCodes: [408, 429, 500, 502, 503, 504],
      retryOnNetworkError: true,
      circuitBreakerFailureThreshold: 5,
      circuitBreakerCooldownMs: 15000,
    },
  },
});

Defaults include bounded retry/backoff and circuit breaking for repeated upstream failures.

Custom Provider

import { LLMProviderRegistry, createLLMInterpreter } from "@renderify/llm";

const registry = new LLMProviderRegistry();
registry.register({
  name: "my-provider",
  create: () => {
    const templates = new Map();
    return {
      configure() {},
      async generateResponse() {
        return { text: "{}", tokensUsed: 0 };
      },
      setPromptTemplate(name, content) {
        templates.set(name, content);
      },
      getPromptTemplate(name) {
        return templates.get(name);
      },
    };
  },
});

const llm = createLLMInterpreter({ provider: "my-provider", registry });

Notes

• Provider implementations follow the LLMInterpreter interface from @renderify/core.
• Streaming support is available through generateResponseStream() when provided by the selected interpreter.