Getting started · Quickstart

Quickstart

Get from zero to your first optimised LLM call in under five minutes. This guide assumes you have a Node.js or Python environment and an API key from your Cosavu dashboard.

01Get an API key

Sign in to the dashboard, create a new project, and copy the API key from the keys page. We recommend storing it in your environment as COSAVU_API_KEY.

# add to .env
COSAVU_API_KEY=cv_live_a1b2c3d4e5f6...

02Install the SDK

Cosavu ships official SDKs for TypeScript and Python. They expose every endpoint as a typed function and handle retries, telemetry, and connection pooling for you.

$ npm install @cosavu/sdk

03Make your first call

The fastest way to feel ContextAPI is to optimise a bloated prompt. Drop this snippet into a file, set the environment variable, and run it.

import { Cosavu } from "@cosavu/sdk"
 
const cosavu = new Cosavu({ apiKey: process.env.COSAVU_API_KEY })
 
const result = await cosavu.context.optimize({
  prompt: "Could you please kindly explain in great detail what RAG is and how it works step by step if that makes sense at all",
  budget: 512,
})
 
console.log(result.optimizedPrompt)
console.log(`Saved ${result.tokensSaved} tokens (${(result.compressionPct * 100).toFixed(0)}%)`)

You should see something like this in your terminal:

$ npx ts-node quickstart.ts
"Explain RAG pipeline. Step by step."
Saved 493 tokens (58%)

04Understand the response

Every optimize call returns a structured payload. The shape is identical across SDKs:

{
  "color:#3b82f6">"optimizedPrompt": "Explain RAG pipeline. Step by step.",
  "color:#3b82f6">"originalTokens": 847,
  "color:#3b82f6">"outputTokens": 354,
  "color:#3b82f6">"tokensSaved": 493,
  "color:#3b82f6">"compressionPct": 0.582,
  "color:#3b82f6">"latencyMs": 14,
  "color:#3b82f6">"ir": {
    "color:#3b82f6">"blocks": [
      { "color:#3b82f6">"blockType": "IDENTITY", ">"tokens": 12 },
      { "color:#3b82f6">"blockType": "INSTRUCTION", ">"tokens": 18 }
    ]
  }
}

Response fields

Next steps

You've seen ContextAPI. Here's where to go from here: