Getting started

Use your API key

Token Relay exposes two upstream-compatible endpoints. Once you have a key, you can keep using your existing OpenAI or Anthropic client libraries — only the base_url and the API key change.

1. Get a key

Sign in to the developer console.
Open API keys.
Click Create API key. Give it a name (something like local-dev or prod-app) and optionally a monthly spend cap in USD.
A token starting with tsk_ appears in the reveal banner. Copy it now — the platform never shows the full token again.

Why reveal-once? The backend only stores a SHA-256 hash of the token. Even an admin with full database access can't recover the original — they can only revoke and reissue.

2. Endpoints

The relay exposes two protocols. Pick the one your client already speaks.

Protocol	Base URL	Auth header
OpenAI Chat Completions	`https://chinzy.com/v1`	`Authorization: Bearer tsk_...`
Anthropic Messages (streaming)	`https://chinzy.com/anthropic/v1`	`x-api-key: tsk_...`

3. Pick a model

Pass any of our public model aliases as the model field. The full list is at Models. A few common ones:

gpt-5.2, gpt-5, gpt-5-mini, gpt-5.4-mini — OpenAI family
claude-sonnet-4.5, claude-haiku-4.5, claude-opus-4.5 — Anthropic family
deepseek-v3.2, kimi-k2-thinking, glm-4.6 — open-weight families

You always pass the platform alias, never the upstream-specific id. Behind the scenes the relay picks a healthy upstream and routes the call there with automatic failover.

4. Examples

curl

curl https://chinzy.com/v1/chat/completions \
  -H "Authorization: Bearer $TOKEN_RELAY_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.2",
    "messages": [
      {"role": "user", "content": "Hi in one sentence"}
    ]
  }'

Python (OpenAI SDK)

from openai import OpenAI

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

resp = client.chat.completions.create(
    model="gpt-5.2",
    messages=[{"role": "user", "content": "Hi in one sentence"}],
)
print(resp.choices[0].message.content)

Node.js (OpenAI SDK)

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.TOKEN_RELAY_KEY,
  baseURL: 'https://chinzy.com/v1',
});

const resp = await client.chat.completions.create({
  model: 'gpt-5.2',
  messages: [{ role: 'user', content: 'Hi in one sentence' }],
});
console.log(resp.choices[0].message.content);

Anthropic SDK (streaming)

import Anthropic from '@anthropic-ai/sdk';

const client = new Anthropic({
  apiKey: process.env.TOKEN_RELAY_KEY,
  baseURL: 'https://chinzy.com/anthropic',
});

const stream = await client.messages.stream({
  model: 'claude-sonnet-4.5',
  max_tokens: 512,
  messages: [{ role: 'user', content: 'Hi in one sentence' }],
});

for await (const event of stream) {
  if (event.type === 'content_block_delta' && event.delta.type === 'text_delta') {
    process.stdout.write(event.delta.text);
  }
}

5. Common errors

Status	Meaning	Fix
`401`	Token missing, malformed, or disabled.	Verify the `Authorization` header value matches the token from the reveal banner. Check API keys — disabled keys show a red badge.
`402` / `403 "balance too low"`	Wallet has no balance to settle this request.	Top up at Billing.
`403 "monthly spend cap reached"`	This specific key's monthly quota is exhausted.	Either wait for the next calendar month, raise the cap on API keys, or use a different key.
`429`	You crossed the per-IP rate limit (300 req/min for relay).	Backoff and retry; widen the gap between requests.
`503`	Every upstream that serves this alias is down. Rare — the relay already auto-retries on intermittent failures.	Try a sibling alias (e.g. `gpt-5-mini` instead of `gpt-5.2`). If persistent, check your status page.

6. Tips

One key per app. Cheap to rotate; lets you disable a leaked credential without breaking other clients.
Set a monthly cap. A runaway dev loop is much easier to recover from when each key is bounded.
Watch usage at /console/usage. The page shows per-model + per-key spend, recent calls with status codes, and a CSV export for accounting.
Streaming is supported on both protocols. Just pass stream: true in the OpenAI SDK or use messages.stream() with Anthropic.

Found something missing or wrong in this doc? Open an issue or ping an admin.