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

# Gemini CLI

> Point Google's official Gemini CLI at ByteSpike. Works against gemini-3-5-flash / gemini-3-flash-lite plus every model reachable via Gemini Native translation.

Google's `gemini-cli` (and the family of CLIs / IDE extensions
built on top of it) speaks the Gemini Native protocol — query-param
auth (`?key=...`) against `/v1beta/models/{model}:generateContent`.
ByteSpike serves that protocol verbatim at `llm.bytespike.ai/v1beta`.

## Prerequisites

* A ByteSpike account + a key bound to the **`gemini-default`** group
  (or any group that serves Gemini models). See [Register](/register).
* Gemini CLI installed:
  ```bash theme={null}
  npm install -g @google/generative-ai-cli
  # or whichever Gemini CLI your team uses
  ```

## Configure

Gemini CLIs typically read `GEMINI_API_KEY` and let you override the
base URL via env or flag.

```bash theme={null}
export GEMINI_API_KEY="sk-byts-..."
export GEMINI_BASE_URL="https://llm.bytespike.ai/v1beta"
```

For the official Google CLI that doesn't expose a base-URL flag, set
the variable per its docs (some versions read
`GOOGLE_API_BASE_URL`); or use the raw curl form (next section)
inside a wrapper script.

## Verify

```bash theme={null}
curl "https://llm.bytespike.ai/v1beta/models/gemini-3.1-pro:generateContent?key=$GEMINI_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "say hi"}]}]
  }'
```

Expect a 200 with a `candidates[0].content.parts[0].text` field, plus
the standard `X-Quota-Remaining-Credits` header.

## Switching models

The model name lives in the URL path:

```
/v1beta/models/gemini-3-5-flash:generateContent
/v1beta/models/gemini-3-flash-lite:generateContent
```

Any model id from your key's group works. See
[`/v1beta` reference](/api-reference/text/gemini-generate-content)
for the full request shape.

## Streaming

Switch the method suffix:

```bash theme={null}
curl "https://llm.bytespike.ai/v1beta/models/gemini-3.1-pro:streamGenerateContent?alt=sse&key=$GEMINI_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "explain SSE briefly"}]}]
  }'
```

SSE stream matches Google's native format —
`data: {chunk}\n\n` blocks terminated by a final `[DONE]` marker.

## SDKs

Both Google's official Generative AI SDK and most third-party Gemini
clients accept a `baseUrl` override at client construction:

```python theme={null}
import google.generativeai as genai
genai.configure(
    api_key="sk-byts-...",
    transport="rest",
    client_options={"api_endpoint": "llm.bytespike.ai"}
)
model = genai.GenerativeModel("gemini-3-5-flash")
print(model.generate_content("hello").text)
```

```ts theme={null}
import { GoogleGenerativeAI } from "@google/generative-ai";

const genAI = new GoogleGenerativeAI("sk-byts-...", {
  baseUrl: "https://llm.bytespike.ai",
});
const model = genAI.getGenerativeModel({ model: "gemini-3-5-flash" });
const result = await model.generateContent("hello");
console.log(result.response.text());
```

## Image + video models via Gemini stack

Veo (Google's video model) ships under the Gemini API surface but
the long-running shape uses ByteSpike's async tasks API instead:

```bash theme={null}
curl https://llm.bytespike.ai/v1/tasks/submit \
  -H "Authorization: Bearer $BYTESPIKE_API_KEY" \
  -d '{
    "model": "veo-3-1",
    "params": {"prompt": "...", "duration_seconds": 8}
  }'
```

See [`POST /tasks/submit`](/api-reference/utility/tasks-submit) and
the [Veo 3.1 model page](/models/veo3-1).

## Troubleshooting

| Symptom                       | Cause                                                                                                              | Fix                                                    |
| ----------------------------- | ------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------ |
| `401 API key not valid`       | Wrong key or missing query param                                                                                   | Verify `?key=sk-byts-...` is present and correct       |
| `403 PERMISSION_DENIED`       | Model not in key's group                                                                                           | Switch to `gemini-default` group or pick another model |
| `404 NOT_FOUND` (model)       | Model id typo — ByteSpike uses the dashed form (`gemini-3-5-flash`), not Google's dotted form (`gemini-3.5-flash`) | Use the slugs from `/api/v1/me/available-models`       |
| Stream cuts after first chunk | Some CLIs default to non-streaming endpoint; check the `:streamGenerateContent` suffix                             | Switch to `streamGenerateContent`                      |

## Next

<CardGroup cols={2}>
  <Card title="/v1beta reference" icon="code" href="/api-reference/text/gemini-generate-content">
    Full request / response / streaming protocol.
  </Card>

  <Card title="Gemini models" icon="google" href="/models">
    Models, capabilities, pricing.
  </Card>

  <Card title="Claude Code CLI" icon="terminal" href="/claude-code-cli">
    The Anthropic-native equivalent.
  </Card>

  <Card title="Cursor IDE" icon="code-branch" href="/cursor">
    Editor-level Gemini integration.
  </Card>
</CardGroup>