llms-sdk

A single interface for OpenAI-compatible chat completions and the Anthropic Messages API, with native bindings for Rust, TypeScript, and Python.

Features

One request model for OpenAI and Anthropic.
Multimodal input — text, image, audio (OpenAI), and document (Anthropic) message parts.
Structured JSON output via JSON Schema.
Tool / function calling with provider-specific serialization.
Streaming responses with text, tool, and reasoning deltas.
Configurable retry policy for transient failures.
Optional CLI (llms) for quick testing.

Supported providers

Provider	Endpoint style
OpenAI	`https://api.openai.com/v1`
Anthropic	`https://api.anthropic.com/v1`

Pick your language

Rust

Native crate with full feature parity. Get started

TypeScript

NAPI-RS bindings for Node.js. Get started

Python

PyO3 bindings for Python 3.10+. Get started

Quick example

use llms_sdk::{LLM, LLMRequest, Message, MessagePart, MessageRole, TextPart};

let request = LLMRequest::builder()
    .api_key(std::env::var("OPENAI_API_KEY")?)
    .messages(vec![Message {
        role: MessageRole::User,
        content: vec![MessagePart::Text(TextPart::new("Hello!"))],
    }])
    .model("gpt-5.4-mini")
    .build();

let llm = LLM::default();
let response = llm.respond(request).await?;

import { Llm, ApiType, MessageRole } from '@cle-does-things/llms-sdk'

const request = {
  apiType: ApiType.OpenAI,
  apiKey: process.env.OPENAI_API_KEY!,
  model: 'gpt-5.4-mini',
  messages: [{
    role: MessageRole.User,
    content: [{ text: 'Hello!', type: 'text' }],
  }],
  stream: false,
  parallelToolCalls: false,
}

const llm = new Llm()
const response = await llm.respond(request)

from llms_sdk_py import LLMRequest, Message, TextPart, LLM

request = LLMRequest(
    model="gpt-5.4-mini",
    api_key="sk-...",
    messages=[Message("user", [TextPart("Hello, world!")])],
    stream=False,
)

client = llm.LLM()
response = await client.respond(request)