Quickstart

Create an account, choose CLI, SDK, or REST, and send your first routed request.

Get from zero to a working /v1/chat request in a few minutes.

1. Create an account

Sign up at console.latentkit.com. New workspaces land on AI Router, which prepares a default route and API key when Platform Access is available.

2. Choose how you want to connect

Use the path that matches what you are doing:

Path	Best for	Credential
CLI	Humans testing from a terminal, support, debugging, logs, traces, and CI checks	Browser login token or `LATENTKIT_API_KEY`
JavaScript SDK	Node.js, Next.js, Express, serverless, background jobs	`LATENTKIT_API_KEY`
Python SDK	FastAPI, Django, Flask, scripts, workers	`LATENTKIT_API_KEY`
PHP SDK (beta)	WordPress, Drupal, Laravel, PHP workers and scripts	`LATENTKIT_API_KEY`
REST / cURL	Any language, quick tests, AI agents, low-level debugging	`LATENTKIT_API_KEY`

3. Copy an API key for app code

Open AI Router or API Keys. Copy your key once when it is created — the console masks existing secrets. Store it in a server-side environment variable:

export LATENTKIT_API_KEY="lk_..."

Never commit API keys or expose them in browser bundles.

If you only want to test from your terminal, you can start with latentkit login instead of copying a raw API key. The CLI opens a browser approval flow and stores a local account profile. App code and CI should still use LATENTKIT_API_KEY.

4. Send a chat request

npm install -g @latentkit/cli
latentkit login
latentkit whoami
latentkit chat "Say hello from LatentKit."

Use JSON when another program or AI agent needs to read the result:

latentkit chat "Say hello from LatentKit." --json

npm install @latentkit/sdk

import { LatentKit } from '@latentkit/sdk';

const client = new LatentKit();

const response = await client.chat.create({
  messages: [{ role: 'user', content: 'Say hello from LatentKit.' }],
  response_profile: 'balanced',
  max_tokens: 100,
});

console.log(response.content);

pip install latentkit

import os
from latentkit import LatentKit

with LatentKit(api_key=os.environ["LATENTKIT_API_KEY"]) as client:
    response = client.chat.create(
        messages=[{"role": "user", "content": "Say hello from LatentKit."}],
        response_profile="balanced",
        max_tokens=100,
    )
    print(response["content"])

composer require latentkit/latentkit-php:0.1.0-beta.3

<?php

require __DIR__ . '/vendor/autoload.php';

use LatentKit\LatentKit;

$client = new LatentKit();

$response = $client->chat->create(
    messages: [['role' => 'user', 'content' => 'Say hello from LatentKit.']],
    responseProfile: 'balanced',
    maxTokens: 100,
);

echo $response['content'];

curl https://ai.latentkit.com/v1/chat \
  -H "Authorization: Bearer $LATENTKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{ "role": "user", "content": "Say hello from LatentKit." }],
    "response_profile": "balanced",
    "max_tokens": 100
  }'

5. Understand the response

A successful response looks like this:

{
  "id": "req_abc123",
  "provider": "openai",
  "model": "gpt-4o-mini",
  "policy_version_id": "pol_v_xyz789",
  "content": "Hello from LatentKit!",
  "usage": { "input_tokens": 12, "output_tokens": 6, "total_tokens": 18 },
  "cost_usd": 0.0000054
}

content is the assistant's answer; provider and model report which route attempt executed. If something fails, read X-LK-Request-ID from response headers and the JSON error field — see Error handling.

Request fails with NO_HEALTHY_PROVIDER? Your key needs an assigned published route with at least one eligible model. New workspaces get this automatically via AI Router; otherwise check Routes and Connections in the console. See Troubleshooting.

Route-based requests

Do not send model, provider, route, or policy in application requests. Change provider/model selection by editing the API key's assigned route in the console.

Optional response depth

Pass response_profile to prefer speed or depth when the route allows overrides:

Value	Use when
`fast`	Lower latency, lower token use
`balanced`	Default quality and speed
`deep`	More reasoning depth when supported

Next steps

Authentication
Core concepts — how routing, keys, and capabilities fit together
CLI, JavaScript SDK, Python SDK, or PHP SDK
Streaming, Tool calling, and the other API reference pages
Vibe coding setup — copy a stack-aware prompt for Cursor, Claude, or other AI tools

On this page