Three lines to your first token.

Quick start

Create an API key in the console, top up a balance, then point any OpenAI SDK at the LATIAO base URL. No other code changes are required.

1. Create an account at api.latiaoapi.com/register.
2. Create a key under API Keys. The plaintext is shown once.
3. Send a request with the base URL https://api.latiaoapi.com/v1.

curl https://api.latiaoapi.com/v1/chat/completions \
  -H "Authorization: Bearer $LATIAO_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4.5",
    "messages": [{"role": "user", "content": "Ship it."}]
  }'

A successful response includes a platform request id in the x-latiao-request-id header. Quote it when contacting support — never post a key in a ticket.

Authentication

Every request requires a bearer token. Keys are scoped to a group of models and a set of limits; create separate keys per environment.

Authorization: Bearer lt_live_xxxxxxxxxxxxxxxxxxxx

• Keys are shown in plaintext once, at creation. Store them in a secret manager.
• Revoking a key takes effect immediately on the next request.
• Never send keys from a browser or mobile app; proxy through your backend.
• Console sessions use short-lived cookies; API keys are not accepted on the console.

Compatibility

Three protocol families are served from one host:

Model IDs are quoted exactly as shown on /models. Aliases such as claude-sonnet-latest resolve to the current stable build of that family.

Streaming

Add "stream": true to receive server-sent events. Chunks are OpenAI-compatible deltas; the final data frame carries usage and cost.

from openai import OpenAI

client = OpenAI(base_url="https://api.latiaoapi.com/v1", api_key="lt-...")

with client.chat.completions.create(
    model="claude-sonnet-4.5",
    messages=[{"role": "user", "content": "Explain failover in one line."}],
    stream=True,
) as stream:
    for chunk in stream:
        print(chunk.choices[0].delta.content or "", end="")

Tool calls

Function calling follows the OpenAI schema. Tools are forwarded to the selected model; unsupported combinations return LATIAO_MODEL_UNAVAILABLE instead of silently dropping the request.

{
  "model": "gpt-6-astra",
  "messages": [{ "role": "user", "content": "What is the weather in Lisbon?" }],
  "tools": [{
    "type": "function",
    "function": {
      "name": "get_weather",
      "description": "Current weather for a city",
      "parameters": {
        "type": "object",
        "properties": { "city": { "type": "string" } },
        "required": ["city"]
      }
    }
  }]
}

Streamed tool calls arrive as incremental tool_calls deltas with an index.

Errors

Errors use one envelope: { "error": { "code", "message", "request_id" } }. Messages are safe to show to end users; they never contain upstream provider details.

Client setup

These clients work without a plugin. Set the base URL and key, then pick a model ID.

ANTHROPIC_BASE_URL=https://api.latiaoapi.com
ANTHROPIC_AUTH_TOKEN=lt-...

OPENAI_BASE_URL=https://api.latiaoapi.com/v1
OPENAI_API_KEY=lt-...

Base URL: https://api.latiaoapi.com/v1
API key: lt-...

provider: openai-compatible
baseURL: https://api.latiaoapi.com/v1

API Provider: OpenAI Compatible
Base URL: https://api.latiaoapi.com/v1

API Host: https://api.latiaoapi.com
API Key: lt-...

Rate limits

Limits are enforced per key and per group: requests per minute, tokens per minute and concurrent streams. Paid plans raise the defaults; enterprise limits are set per account.

• Responses include x-ratelimit-limit-requests and x-ratelimit-remaining-requests.
• 429 responses include Retry-After in seconds.
• Bursts are smoothed over a short window; sustained overage is rejected, not queued.

Need higher limits? Contact us with your expected requests per minute and model mix.

Changelog

• 2026-11-15 · 0.1 draft — Documentation skeleton published with the brand site preview.
• Planned · 0.2 — Wire examples to the frozen v1 OpenAPI contract, add Cmd+K search and per-code snippets.