SDKs & Wrappers - Curtly

Curtly provides an official TypeScript and Node.js SDK (curtly) for automated two-stage token optimization: dynamic input prompt compression and Output Brevity Governor directive injection.

1. Official TypeScript / Node.js SDK

Install the package via your preferred package manager:

npm install curtly

pnpm add curtly

yarn add curtly

bun add curtly

Quickstart Example

import { Curtly } from 'curtly';

const curtly = new Curtly({
  apiKey: process.env.CURTLY_API_KEY, // Or pass ctly_live_... key
  defaultMode: 'balanced',           // 'conservative' | 'balanced' | 'aggressive'
  defaultBrevityLevel: 'full',        // 'lite' | 'full' | 'ultra' | 'off'
});

// Compress any prompt or context block
const result = await curtly.compress(
  "Please note that you should write a function in Python that parses CSV files."
);

console.log(result.compressed);
// Output: "Write Python function that parses CSV files."
console.log(`Saved ${result.stats.savedPercent}% (${result.stats.savedTokens} tokens)`);

2. Drop-In OpenAI Client Wrapper

Optimize all OpenAI SDK calls with zero changes to your existing application code:

import OpenAI from 'openai';
import { Curtly } from 'curtly';

const openai = new OpenAI();
const curtly = new Curtly({ apiKey: process.env.CURTLY_API_KEY });

// 1-Line change: wrap your existing OpenAI client
const client = curtly.wrapOpenAI(openai);

// All completions now automatically benefit from Safe Vault compression and Brevity Governor
const response = await client.chat.completions.create({
  model: 'gpt-6-astra',
  messages: [
    { role: 'user', content: 'Write a TypeScript function to validate RFC 5322 email addresses.' }
  ],
});

console.log(response.choices[0].message.content);

3. Vercel AI SDK Integration

Integrate Curtly directly into Next.js App Router route handlers:

import { streamText } from 'ai';
import { openai } from '@ai-sdk/openai';
import { curtly } from 'curtly';

export async function POST(req: Request) {
  const { messages } = await req.json();

  // Automatically compress user prompts and inject Output Brevity directives
  const { messages: optimizedMessages } = await curtly.optimizeMessages(messages, {
    mode: 'balanced',
    brevityLevel: 'full',
  });

  const result = streamText({
    model: openai('gpt-6-astra'),
    messages: optimizedMessages,
  });

  return result.toDataStreamResponse();
}

4. Python SDK & Requests Integration

For Python backends, use the direct Curtly API client:

import os
import requests

CURTLY_API_KEY = os.environ.get("CURTLY_API_KEY")

def compress_prompt(prompt: str, mode: str = "balanced") -> str:
    try:
        res = requests.post(
            "https://curtly.dev/api/v1/compress",
            headers={"Authorization": f"Bearer {CURTLY_API_KEY}"},
            json={
                "prompt": prompt,
                "mode": mode,
                "protectCodeBlocks": True,
                "protectJson": True,
            },
            timeout=4
        )
        if res.status_code == 200:
            return res.json().get("compressed", prompt)
    except Exception as e:
        print(f"Curtly fallback: {e}")
    return prompt

# LangChain Drop-In Configuration (Claude Fable 5.1 yields the highest token savings)
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="claude-fable-5-1",
    openai_api_base="https://curtly.dev/v1",
    openai_api_key=CURTLY_API_KEY
)