@agent-orch/llm

LLM provider implementations with a unified interface for the Agent Orch framework. Currently supports OpenAI (using the Responses API with streaming) and Anthropic (using the Messages API with streaming).

Installation

pnpm add @agent-orch/llm

Key API

| Export | Description | |--------|-------------| | createLLM | Factory function that creates an LLM instance from a model ID | | OpenAILLM | OpenAI provider implementation with streaming support | | AnthropicLLM | Anthropic provider implementation with streaming support |

Usage

import { createLLM } from "@agent-orch/llm";

const llm = createLLM({
  modelId: "openai/gpt-4o",
  apiKey: process.env.OPENAI_API_KEY,
});

const stream = llm.stream(messages, tools);

for await (const event of stream) {
  console.log(event);
}

The modelId format is "provider/model" — the prefix selects the provider and the suffix is passed as the model name.

Supported Providers

| Provider | Prefix | Example | |----------|--------|---------| | OpenAI | openai/ | openai/gpt-4o | | Anthropic | anthropic/ | anthropic/claude-3-5-sonnet-20241022 |

Dependencies

| Package | Purpose | |---------|---------| | @agent-orch/core | Core types and interfaces | | openai | Official OpenAI SDK | | @anthropic-ai/sdk | Official Anthropic SDK | | zod | Runtime schema validation | | zod-to-json-schema | Zod to JSON Schema conversion |

Documentation

See the Agent Orch docs for full framework documentation.