Quickstart.

Authentication

Every request carries your key in one of two headers. Use whichever your client supports.

• Authorization: Bearer <key>, which is what OpenAI SDKs send
• X-API-Key: <key>

Keep keys out of source control. Read them from an environment variable, as the examples below do.

First request

Use any model id from the list below. The examples use openai.gpt-oss-120b.

curl https://api.linethree.ai/v1/chat/completions \
  -H "Authorization: Bearer $LINE_THREE_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai.gpt-oss-120b",
    "messages": [
      {"role": "user", "content": "Write table tests for parse_window()."}
    ]
  }'

curl https://api.linethree.ai/v1/chat/completions \
  -H "X-API-Key: $LINE_THREE_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "openai.gpt-oss-120b", "messages": [{"role": "user", "content": "…"}]}'

Models

Every model runs on Amazon Bedrock in US West (Oregon) (us-west-2), called through the Amazon Bedrock Runtime API (bedrock-runtime). Send Line Three’s model id exactly as shown; the gateway sends AWS’s bedrock-runtime id for that model, as listed on its AWS page (for gpt-oss-120b, openai.gpt-oss-120b-1:0; for the two Nemotron models, the same id as Line Three’s).

What is supported

Line Three accepts the OpenAI chat completions format, not every OpenAI API. A client should work if it needs only what is listed here.

Clients

Set the client’s base URL to https://api.linethree.ai/v1 and its API key to your Line Three key. With curl, the URL is the request itself, as in the first request above.

Some editors relay your request through the editor maker’s own servers before it reaches the endpoint you set. If yours does, that company is in your request path too. Check your tool’s documentation. These tools keep their own chat history on the machine or server where they run; Line Three keeps no copy, but the tool may.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.linethree.ai/v1",
    api_key=os.environ["LINE_THREE_KEY"],
)

completion = client.chat.completions.create(
    model="openai.gpt-oss-120b",
    messages=[{"role": "user", "content": "Explain this module before I refactor it."}],
)
print(completion.choices[0].message.content)

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.linethree.ai/v1',
  apiKey: process.env.LINE_THREE_KEY,
});

const completion = await client.chat.completions.create({
  model: 'openai.gpt-oss-120b',
  messages: [{ role: 'user', content: 'Summarize this instruction into a checklist.' }],
});
console.log(completion.choices[0].message.content);

Streaming

Set "stream": true to receive the answer as it is written, as server-sent events ending in data: [DONE]. Add "stream_options": {"include_usage": true} for a final chunk with the token counts.

At the cap: a request that starts while the month’s cap has room is allowed to finish, even if it goes over the cap. The next request gets a 402. If a request is still running when the cap is reached or a prepaid balance reaches zero, Line Three absorbs that overrun and never bills it.

Errors

Errors on /v1/chat/completions have the OpenAI shape, with a request id to quote if you contact us:

{
  "error": {
    "message": "The service is at capacity. Please retry shortly.",
    "type": "service_unavailable",
    "code": "capacity",
    "request_id": "1842"
  }
}

Limits

Per key

Per-key caps are optional limits inside the account’s cap. The $50 minimum is per account, not per key.

Rate

Rate limits follow the cap, for card and purchase-order accounts alike: 60 requests a minute per key below a $1,000 monthly cap, and 600 a minute per key at $1,000 or more.

Context

gpt-oss-120b: 128K tokens; NVIDIA Nemotron 3 Super 120B: 256K tokens; NVIDIA Nemotron Nano 3 30B: 256K tokens.

Request size

Up to 10 MB of JSON per request.

Timeout

10 minutes per request.