> ## 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.

# API Reference

> Every endpoint, one base URL, one key.

All endpoints share the same base:

```
https://llm.bytespike.ai
```

* `https://llm.bytespike.ai/v1/*` — Anthropic + OpenAI surfaces
* `https://llm.bytespike.ai/v1beta/*` — Gemini Native surface
* `https://llm.bytespike.ai/api/v1/*` — management endpoints (keys, usage, billing)

Auth is a single header (see [Authentication](/authentication)).
Failures don't bill. Every response carries quota + rate-limit headers
so you don't need a side channel for accounting.

## Endpoint families

<CardGroup cols={2}>
  <Card title="Text" href="/api-reference/text/claude-messages">
    Claude Messages, OpenAI Chat Completions, OpenAI Responses, Gemini Native.
  </Card>

  <Card title="Image" href="/api-reference/image/gpt-image-2">
    GPT Image 2, Nano Banana Pro, Nano Banana 2.
  </Card>

  <Card title="Video" href="/api-reference/video/veo-3-1">
    Veo 3.1 / Fast / Lite. Async via `/v1/tasks/*`.
  </Card>

  <Card title="Utility" href="/api-reference/utility/balance">
    `/v1/models`, `/v1/usage`, `/v1/balance`, async `/v1/tasks/{submit,query,cancel}`.
  </Card>
</CardGroup>

## Live pricing

Per-token / per-call rates come from the production gateway and refresh
nightly. The authoritative table lives at
[bytespike.ai/pricing](https://bytespike.ai/pricing) — link directly to
the section you need:

* [Text models →](https://bytespike.ai/pricing#text)
* [Image models →](https://bytespike.ai/pricing#image)
* [Video models →](https://bytespike.ai/pricing#video)
* [Utility endpoints →](https://bytespike.ai/pricing#utility)

## Conventions

**Async vs sync.** Text endpoints are synchronous. Image is synchronous
(≤30s for single-image, large batches may use the async tasks API).
Video is async — `POST /v1/tasks/submit` returns a `task_id`, poll
`/v1/tasks/query` (free) or stream `/v1/tasks/stream/{task_id}` via
SSE. The [`tasks` reference](/api-reference/utility/tasks-submit)
covers the full lifecycle.

**Streaming.** Pass `"stream": true` (or `?stream=true` on Gemini). The
SSE stream is byte-for-byte compatible with each protocol's native
streaming format — Anthropic event names on `/v1/messages`, OpenAI
event names on `/v1/chat/completions` and `/v1/responses`, Gemini's
chunked `streamGenerateContent` on `/v1beta`.

**Accounting headers.** Every response (success **and** failure)
carries the gateway's quota envelope:

| Header                          | What it means                                                       |
| ------------------------------- | ------------------------------------------------------------------- |
| `X-RateLimit-Limit`             | USD cap of the rate-limit bucket closest to constraining you.       |
| `X-RateLimit-Remaining`         | Budget left in that bucket.                                         |
| `X-RateLimit-Reset`             | Bucket reset Unix timestamp.                                        |
| `X-Quota-Remaining-Credits`     | Lifetime credits remaining on this key (USD; `1 USD = 1M credits`). |
| `X-Org-Quota-Remaining-Credits` | Same, at the org wallet level, on org-owned keys.                   |

For per-request cost, query
[`GET /api/v1/usage`](/api-reference/account/me-usage) — it returns
one row per request with `prompt_tokens`, `completion_tokens`, and
billed `credits`. Non-2xx responses don't bill, so
`X-Quota-Remaining-Credits` doesn't move on a failure.