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

# LLM gateway

> Use llm.corsair.dev as an OpenAI-compatible API with per-developer keys and budgets.

`llm.corsair.dev` is Corsair's internal LLM gateway, an OpenAI-compatible API you can use with the Vercel AI SDK, OpenAI SDK, Cursor, or any client that accepts a custom `baseURL`.

Visiting [llm.corsair.dev](https://llm.corsair.dev) in a browser redirects here.

## Quick start

Add these env vars to your app:

```bash theme={null}
LITELLM_API_KEY=sk-your-virtual-key
LITELLM_BASE_URL=https://llm.corsair.dev/v1
LITELLM_MODEL=gpt-5.4-mini   # optional default
```

Point your client at the gateway and call models by name.

## Available models

| Model           | Typical use                    |
| --------------- | ------------------------------ |
| `gpt-5.6-sol`   | Newest flagship                |
| `gpt-5.6-terra` | Balanced / everyday            |
| `gpt-5.6-luna`  | Fast / cheapest 5.6            |
| `gpt-5.5`       | Previous flagship              |
| `gpt-5.3-codex` | Coding / agentic (Codex)       |
| `gpt-5.4`       | General flagship               |
| `gpt-5.4-mini`  | Fast / balanced (good default) |
| `gpt-5.4-nano`  | Cheapest / highest throughput  |

List models with your key:

```bash theme={null}
curl https://llm.corsair.dev/v1/models \
  -H "Authorization: Bearer $LITELLM_API_KEY"
```

## Check your budget

Each key has a dollar budget. Check how much you've used and how much is left:

```bash theme={null}
curl https://llm.corsair.dev/v1/key/info \
  -H "Authorization: Bearer $LITELLM_API_KEY"
```

Example response:

```json theme={null}
{
  "key": "sk-...",
  "info": {
    "spend": 1.42,
    "max_budget": 25,
    "budget_duration": "30d",
    "models": ["corsair-models"]
  }
}
```

* **`spend`**: dollars used so far in the current budget window
* **`max_budget`**: your total allowance
* **`budget_duration`**: when spend resets (e.g. `30d` = monthly)

Remaining budget: `max_budget - spend` (e.g. `$23.58` left in the example above).

When `spend` reaches `max_budget`, requests return an error until the budget resets or your limit is increased.

## Vercel AI SDK

```ts theme={null}
import { createOpenAI } from '@ai-sdk/openai';
import { generateText } from 'ai';

const corsair = createOpenAI({
  apiKey: process.env.LITELLM_API_KEY,
  baseURL: process.env.LITELLM_BASE_URL ?? 'https://llm.corsair.dev/v1',
});

const { text } = await generateText({
  model: corsair('gpt-5.4-mini'),
  prompt: 'Hello!',
});
```

Streaming works the same way with `streamText`.

## OpenAI SDK

```ts theme={null}
import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.LITELLM_API_KEY,
  baseURL: 'https://llm.corsair.dev/v1',
});

const response = await client.chat.completions.create({
  model: 'gpt-5.4-mini',
  messages: [{ role: 'user', content: 'Hello!' }],
});
```

<Warning>
  Do not use `openai('gpt-4o')` with the default OpenAI provider, which calls OpenAI directly. Always use `createOpenAI({ baseURL, apiKey })` so requests go through the gateway.
</Warning>

## curl

```bash theme={null}
curl https://llm.corsair.dev/v1/chat/completions \
  -H "Authorization: Bearer $LITELLM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4-mini",
    "messages": [{ "role": "user", "content": "hi" }]
  }'
```

## Env var aliases

Some projects use different names. These are equivalent:

| Preferred          | Also accepted                                           |
| ------------------ | ------------------------------------------------------- |
| `LITELLM_API_KEY`  | `OPENAI_API_KEY` (when `baseURL` points at the gateway) |
| `LITELLM_BASE_URL` | none                                                    |
| `LITELLM_MODEL`    | none                                                    |

## Endpoints

| Path                                  | Purpose                          |
| ------------------------------------- | -------------------------------- |
| `https://llm.corsair.dev/v1/*`        | OpenAI-compatible inference      |
| `https://llm.corsair.dev/v1/models`   | List available models            |
| `https://llm.corsair.dev/v1/key/info` | Check spend and remaining budget |
| `https://llm.corsair.dev/health`      | Health check                     |