# API Protocols

> LMU AI supports three inbound protocols — Anthropic, OpenAI Compatible, and Gemini native. One table to pick the right Base URL and endpoint and avoid errors.

URL: https://docs.lmuai.ai/docs/guide/api-protocols



LMU AI supports three inbound protocols: **Anthropic, OpenAI Compatible, and Gemini native v1beta**. Every endpoint uses your LMU AI `sk-` API key, and the models you can actually call are decided by the key's group.

<Callout type="info" title="Two things to keep straight first">
  * **"Which protocol"** is decided by the client, SDK, and use case. The Anthropic SDK uses `/v1/messages`, the OpenAI SDK uses `/v1/chat/completions` or `/v1/responses`, and Gemini native image calls use `/v1beta/models/{model}:generateContent`.
  * **"Which models you can call"** is decided by your API key's **group** (the upstream account behind your subscription / top-up plan), which is **a separate dimension from the inbound protocol**.

  In other words: the wrong protocol gives an outright 401 / 404; the right protocol with a model outside your group's range returns a model-unavailable error.
</Callout>

<Callout type="warn" title="The most common error comes from the wrong protocol">
  * The **Anthropic protocol** Base URL does **not** include the `/v1` suffix
  * The **OpenAI protocol** SDK Base URL usually **does** include the `/v1` suffix
  * The **Gemini native protocol** uses `https://api.lmuai.ai` as the host and calls the full `/v1beta/...` path

  Getting it wrong causes 400 / 401 / 404. Before configuring a tool or writing code, confirm which protocol the client expects.
</Callout>

***

## Pick your protocol at a glance [#pick-your-protocol-at-a-glance]

| Protocol                 | Base URL                  | Typical tools                                                                                                                                                         |
| ------------------------ | ------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Anthropic protocol**   | `https://api.lmuai.ai`    | Claude Code (CLI / desktop / VS Code extension), the official `anthropic` SDK, Cherry Studio, Kilo Code with Claude / Chinese models, any Anthropic-compatible client |
| **OpenAI Compatible**    | `https://api.lmuai.ai/v1` | Codex CLI, Codex App, Cursor / Cline / Roo Code / OpenCode, the official `openai` SDK, VS Code extensions with GPT, any OpenAI-compatible client                      |
| **Gemini native v1beta** | `https://api.lmuai.ai`    | Gemini native SDK / HTTP clients, Gemini text-to-image and image-to-image, model list and `generateContent`                                                           |

All three protocols use the LMU AI key that starts with `sk-`:

* Anthropic protocol: `Authorization: Bearer <YOUR_API_KEY>` or `x-api-key: <YOUR_API_KEY>`
* OpenAI Compatible: `Authorization: Bearer <YOUR_API_KEY>`
* Gemini native: `x-goog-api-key: <YOUR_API_KEY>` recommended, Bearer also accepted

***

## Anthropic protocol [#anthropic-protocol]

**Base URL:** `https://api.lmuai.ai` (**no** `/v1`)

**Endpoints:**

* `POST /v1/messages` — message conversation
* `POST /v1/messages/count_tokens` — token counting
* `GET /v1/models` — available model list

**Python SDK example:**

```python
from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.lmuai.ai",
    api_key="sk-xxxxxxxx",
)

resp = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.content[0].text)
```

**curl example:**

```bash
curl -X POST https://api.lmuai.ai/v1/messages \
  -H "Authorization: Bearer sk-xxxxxxxx" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 1024,
    "messages": [{"role":"user","content":"Hello"}]
  }'
```

<Callout type="info" title="Why is the SDK base_url without /v1 but curl with /v1?">
  The Anthropic official SDK's `base_url` convention is to omit `/v1`; the SDK appends paths like `/v1/messages` internally. When you write curl by hand, you write the full path `/v1/messages`. Both point to the same endpoint.
</Callout>

***

## OpenAI protocol [#openai-protocol]

**Base URL:** `https://api.lmuai.ai/v1` (**with** `/v1`)

**Endpoints:**

* `POST /chat/completions` — standard Chat Completions API
* `POST /responses` — OpenAI Responses API (including the `/responses/{id}` sub-paths)
* `POST /images/generations`, `POST /images/edits` — image generation / editing

**Python SDK example:**

```python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.lmuai.ai/v1",
    api_key="sk-xxxxxxxx",
)

resp = client.chat.completions.create(
    model="gpt-5.6-sol",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)
```

**curl example:**

```bash
curl -X POST https://api.lmuai.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.6-sol",
    "messages": [{"role":"user","content":"Hello"}]
  }'
```

***

## Gemini native protocol [#gemini-native-protocol]

**Base URL:** `https://api.lmuai.ai`

**Main endpoints:**

* `GET /v1beta/models` — list Gemini native models
* `GET /v1beta/models/{model}` — query a specific model
* `POST /v1beta/models/{model}:generateContent` — text generation, text-to-image, and image-to-image
* `POST /v1beta/models/{model}:streamGenerateContent?alt=sse` — streaming generation

**Authentication:**

```http
x-goog-api-key: YOUR_API_KEY
```

Minimal text-to-image example:

```bash
curl --request POST \
  'https://api.lmuai.ai/v1beta/models/gemini-3.1-flash-image:generateContent' \
  -H 'x-goog-api-key: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "contents": [{
      "role": "user",
      "parts": [{"text": "an orange cat wearing an astronaut helmet"}]
    }],
    "generationConfig": {
      "responseModalities": ["TEXT", "IMAGE"],
      "imageConfig": {"aspectRatio": "1:1", "imageSize": "1K"}
    }
  }'
```

<Callout type="info" title="The Gemini image endpoint is not /v1/chat/completions">
  Gemini image models use the native `generateContent`. For full text-to-image, image-to-image, 1K / 2K / 4K, and Base64 parsing details, see the [Gemini Image API](/docs/api/gemini-image).

  For GPT image models see the [GPT Image API](/docs/api/gpt-image), and for Grok image models see the [Grok Image API](/docs/api/grok-image). For large offline Gemini jobs see the [Gemini Batch Image API](/docs/api/gemini-image-batch).
</Callout>

***

## Protocol and available models are two different things [#protocol-vs-models]

Which models your key can call is **decided entirely by the upstream account mounted on its group**, independent of which protocol you call with:

| The upstream on your group                                     | The models you can actually call                                                                                                      |
| -------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------- |
| OpenAI account only                                            | GPT series only                                                                                                                       |
| Claude account only                                            | Claude series only (even if you call with the OpenAI protocol, the backend translates the protocol, but the model range is unchanged) |
| Chinese-model upstream only (e.g. GLM / Kimi)                  | the corresponding Chinese models only                                                                                                 |
| Model routing configured in the backend (multi-upstream group) | routed by model name to different upstreams, can span brands — the exact range depends on the group config                            |

So:

* Choosing the Anthropic, OpenAI Compatible, or Gemini native protocol determines the inbound request format; the actual model range is still decided by the API key's group.
* To see which models your current key can call, check the group's model list on the **API Keys** detail page or the **Available Models** page in the console.

***

## About the Claude Max group [#about-the-claude-max-group]

<Callout type="warn" title="The Claude Max group supports the Anthropic protocol only">
  **The Claude Max group is for Claude Code only**, so it can only use the **Anthropic protocol** (`https://api.lmuai.ai`).

  If you are using a key from a Claude Max plan group:

  * ✅ It works in Claude Code (CLI / desktop / VS Code extension)
  * ❌ It **cannot** be used with Codex CLI, Cursor, Cherry Studio, or any tool that uses the OpenAI protocol
  * ❌ It **cannot** be entered as `https://api.lmuai.ai/v1`

  To use an OpenAI-protocol tool, switch to a key from a pay-as-you-go / regular subscription group (the exact available models still depend on the plan group you purchased).
</Callout>

***

## Troubleshooting [#troubleshooting]

| Symptom                                     | Usual cause                                                                                                             | What to do                                                                                                                            |
| ------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------- |
| `401 Unauthorized`                          | Base URL uses the wrong protocol / key typo / IDE not restarted                                                         | Check the Base URL matches the tool's protocol; restart the IDE to reload config                                                      |
| `404 Not Found`                             | OpenAI-protocol URL missing `/v1`, Anthropic URL wrongly including `/v1`, or Gemini path not using `/v1beta/models/...` | Re-check the Base URL and full endpoint against the table above                                                                       |
| Model unavailable / `No available accounts` | Called a model outside your group's range (e.g. calling GPT with a Claude Max group)                                    | Confirm the models your group actually includes on the **Available Models** page, or switch to a group that includes the target model |
| `429 Too Many Requests`                     | Daily quota exhausted                                                                                                   | See the [FAQ](/docs/guide/faq#issue-2)                                                                                                |

For more troubleshooting see the [FAQ](/docs/guide/faq).

***

## Next steps [#next-steps]

Once you have the protocol and Base URL right, pick the tool you want:

* [CC Switch (one-click import, recommended for Claude Code users)](/docs/tools/cc-switch)
* [Claude Code CLI](/docs/tools/claude-code) · [Desktop](/docs/tools/claude-code-desktop) · [VS Code extension](/docs/tools/claude-code-vscode)
* [Codex CLI · Windows](/docs/tools/codex-cli-windows) · [Mac/Linux](/docs/tools/codex-cli-mac) · [Server](/docs/tools/codex-cli-server)
* [Codex App desktop](/docs/tools/codex-app) · [VS Code / Cursor / Trae extension](/docs/tools/vscode-plugin)
* [OpenCode](/docs/tools/opencode) · [Cherry Studio](/docs/tools/cherry) · [IDEA Kilo Code](/docs/tools/kilo-code-idea) · [Hermes Agent](/docs/tools/hermes)
* [Gemini Image API](/docs/api/gemini-image) · [GPT Image API](/docs/api/gpt-image) · [Grok Image API](/docs/api/grok-image) · [Gemini Batch Image API](/docs/api/gemini-image-batch)