> ## Documentation Index
> Fetch the complete documentation index at: https://docs.terminal.fyi/llms.txt
> Use this file to discover all available pages before exploring further.

# Quickstart

> Make your first Terminal Chat Completions request.

# Quickstart

Terminal exposes an OpenAI-compatible Chat Completions endpoint.

## 1. Set your API key

```bash theme={null}
export TERMINAL_API_KEY="your_api_key"
```

To request an API key, email [shubham@terminal.fyi](mailto:shubham@terminal.fyi).

## 2. Choose latency and cost per request

Different agent steps have different requirements. Use Standard when a user is waiting, and use Flex when a background step can trade time for lower cost.

Both windows use the same endpoint. Add `service_tier: "flex"` when the request can be patient. Terminal charges per token and stays fully serverless, so you do not need to manage deployments or plan around rate limits.

For finer request-level control, Terminal is adding SLA headers for latency and cost preferences. See [Custom SLAs](/custom-slas).

## 3. Make a Standard window request

Standard is the default window. It is designed for fast responses.

<CodeGroup>
  ```python Python SDK theme={null}
  from openai import OpenAI

  client = OpenAI(
      base_url="https://api.terminal.fyi/v1",
      api_key="TERMINAL_API_KEY",
  )

  response = client.chat.completions.create(
      model="gemma-4-31b",
      messages=[
          {
              "role": "user",
              "content": "Write a short release note for a lower-cost inference window.",
          }
      ],
      max_tokens=8000,
  )

  print(response.choices[0].message.content)
  ```

  ```typescript TypeScript SDK theme={null}
  import OpenAI from "openai";

  const client = new OpenAI({
    baseURL: "https://api.terminal.fyi/v1",
    apiKey: process.env.TERMINAL_API_KEY,
  });

  const response = await client.chat.completions.create({
    model: "gemma-4-31b",
    messages: [
      {
        role: "user",
        content: "Write a short release note for a lower-cost inference window.",
      },
    ],
    max_tokens: 8000,
  });

  console.log(response.choices[0].message.content);
  ```

  ```python Python HTTP theme={null}
  import requests

  response = requests.post(
      "https://api.terminal.fyi/v1/chat/completions",
      headers={
          "Authorization": "Bearer TERMINAL_API_KEY",
          "Content-Type": "application/json",
      },
      json={
          "model": "gemma-4-31b",
          "messages": [
              {
                  "role": "user",
                  "content": "Write a short release note for a lower-cost inference window.",
              }
          ],
          "max_tokens": 8000,
      },
      timeout=60,
  )
  response.raise_for_status()

  completion = response.json()
  print(completion["choices"][0]["message"]["content"])
  ```

  ```typescript TypeScript HTTP theme={null}
  const response = await fetch("https://api.terminal.fyi/v1/chat/completions", {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.TERMINAL_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      model: "gemma-4-31b",
      messages: [
        {
          role: "user",
          content: "Write a short release note for a lower-cost inference window.",
        },
      ],
      max_tokens: 8000,
    }),
  });
  if (!response.ok) throw new Error(await response.text());

  const completion = await response.json();
  console.log(completion.choices[0].message.content);
  ```

  ```bash cURL theme={null}
  curl -sS https://api.terminal.fyi/v1/chat/completions \
    -H "Authorization: Bearer $TERMINAL_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "gemma-4-31b",
      "messages": [
        {
          "role": "user",
          "content": "Write a short release note for a lower-cost inference window."
        }
      ],
      "max_tokens": 8000
    }' | jq
  ```
</CodeGroup>

## 4. Make a Gemma Flex window request

Flex uses the same endpoint and waits for completion. It is designed for async-style workloads where cost matters more than latency. Allow long client timeouts.

<CodeGroup>
  ```python Python SDK theme={null}
  from openai import OpenAI

  client = OpenAI(
      base_url="https://api.terminal.fyi/v1",
      api_key="TERMINAL_API_KEY",
      timeout=1800.0,
  )

  response = client.chat.completions.create(
      model="gemma-4-31b",
      service_tier="flex",
      messages=[
          {
              "role": "user",
              "content": "Summarize this overnight evaluation run for the platform team.",
          }
      ],
      max_tokens=8000,
  )

  print(response.choices[0].message.content)
  ```

  ```typescript TypeScript SDK theme={null}
  import OpenAI from "openai";

  const client = new OpenAI({
    baseURL: "https://api.terminal.fyi/v1",
    apiKey: process.env.TERMINAL_API_KEY,
    timeout: 1_800_000,
  });

  const response = await client.chat.completions.create({
    model: "gemma-4-31b",
    service_tier: "flex",
    messages: [
      {
        role: "user",
        content: "Summarize this overnight evaluation run for the platform team.",
      },
    ],
    max_tokens: 8000,
  });

  console.log(response.choices[0].message.content);
  ```

  ```python Python HTTP theme={null}
  import requests

  response = requests.post(
      "https://api.terminal.fyi/v1/chat/completions",
      headers={
          "Authorization": "Bearer TERMINAL_API_KEY",
          "Content-Type": "application/json",
      },
      json={
          "model": "gemma-4-31b",
          "service_tier": "flex",
          "messages": [
              {
                  "role": "user",
                  "content": "Summarize this overnight evaluation run for the platform team.",
              }
          ],
          "max_tokens": 8000,
      },
      timeout=1800,
  )
  response.raise_for_status()

  completion = response.json()
  print(completion["choices"][0]["message"]["content"])
  ```

  ```typescript TypeScript HTTP theme={null}
  const response = await fetch("https://api.terminal.fyi/v1/chat/completions", {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.TERMINAL_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      model: "gemma-4-31b",
      service_tier: "flex",
      messages: [
        {
          role: "user",
          content: "Summarize this overnight evaluation run for the platform team.",
        },
      ],
      max_tokens: 8000,
    }),
    signal: AbortSignal.timeout(1_800_000),
  });
  if (!response.ok) throw new Error(await response.text());

  const completion = await response.json();
  console.log(completion.choices[0].message.content);
  ```

  ```bash cURL theme={null}
  curl -sS --max-time 1800 https://api.terminal.fyi/v1/chat/completions \
    -H "Authorization: Bearer $TERMINAL_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "gemma-4-31b",
      "service_tier": "flex",
      "messages": [
        {
          "role": "user",
          "content": "Summarize this overnight evaluation run for the platform team."
        }
      ],
      "max_tokens": 8000
    }' | jq
  ```
</CodeGroup>

## Request shape

```json theme={null}
{
  "model": "gemma-4-31b",
  "service_tier": "standard",
  "messages": [
    {
      "role": "user",
      "content": "Classify this support ticket by urgency and next owner."
    }
  ],
  "max_tokens": 8000
}
```

`service_tier` is optional. If omitted, Terminal uses `standard`.