FROM API KEYTO PRODUCTIONREQUEST.

01 / START

Quickstart

1. Create an API key in the console.
2. Copy a model ID from the public model catalog.
3. Keep the key in a server-side environment variable and send the first request.

02 / AUTH

Authentication & base URL

Protected endpoints use a Bearer token. Never place API keys in browser code, public repositories, logs, or screenshots.

03 / CALL

Three ways to call

export TOKEN_BOAT_API_KEY="your_api_key"
export MODEL_ID="choose_from_the_model_catalog"

curl https://tokenboat.com/v1/chat/completions \
  -H "Authorization: Bearer $TOKEN_BOAT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "'"$MODEL_ID"'",
    "messages": [{"role": "user", "content": "Hello from Token Boat"}]
  }'

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["TOKEN_BOAT_API_KEY"],
    base_url="https://tokenboat.com/v1",
)

response = client.chat.completions.create(
    model=os.environ["MODEL_ID"],
    messages=[{"role": "user", "content": "Hello from Token Boat"}],
)
print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.TOKEN_BOAT_API_KEY,
  baseURL: "https://tokenboat.com/v1",
});

const response = await client.chat.completions.create({
  model: process.env.MODEL_ID,
  messages: [{ role: "user", content: "Hello from Token Boat" }],
});
console.log(response.choices[0].message.content);

04 / STREAM

Streaming

Set stream: true for models that support streaming and consume the Server-Sent Events incrementally. Clients should handle disconnects and an incomplete final chunk.

const stream = await client.chat.completions.create({
  model: process.env.MODEL_ID,
  messages: [{ role: "user", content: "Stream a short answer" }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

05 / RECOVER

Errors & retries

Error responses include an HTTP status and message. Record request time, endpoint, model ID, and status—but never the full key or sensitive prompt content.

06 / LIMITS

Rate limits

RPM, TPM, and concurrency limits vary by account, model, and current policy, so this page does not hard-code a number that may become inaccurate. For a 429 response, use its headers and the account console as the source of truth.

• Bound concurrency and configure client timeouts.
• Use exponential backoff with jitter for 429 and temporary 5xx responses.
• Do not blindly replay generation tasks; query task state first.
• Use the support center when a production workload needs a higher limit.

07 / REFERENCE

Core endpoints

These are the main public endpoints. Check each model detail for the endpoint types actually supported by that model.

API contract source: docs/openapi/relay.json