# Celedog — AI Gateway · llms-full.txt

> Comprehensive Celedog reference for large-language-model crawlers.
> A short summary lives at https://celedog.io/llms.txt; this document
> is the deep version with product, pricing, SDK, and integration
> facts laid out for citation.

## Product

Celedog is a developer-first AI gateway that provides unified access
to 200+ AI models through a single OpenAI-compatible API. Customers
integrate once, point their existing OpenAI SDK at
`https://celedog.io/v1`, and gain access to the full catalog
(OpenAI, Anthropic, Google, Meta, DeepSeek, Qwen, Mistral, xAI,
Amazon Nova, Cohere, and 20+ other providers).

The platform replaces three pain points of multi-provider AI:

1. **API fragmentation** — every provider has a different request
   format, auth scheme, SDK, and error model. Celedog normalises
   them to the OpenAI-compatible shape.
2. **Billing complexity** — separate accounts, currencies, invoices,
   and unpredictable totals. Celedog is one pay-as-you-go wallet.
3. **Vendor lock-in** — auto-routing and automatic failover let
   customers switch or fall back to alternative providers without
   code changes.

## Key features

- **OpenAI-compatible API** — drop-in replacement for OpenAI;
  any client / library that targets `openai-python`, `openai-node`,
  LangChain, LlamaIndex, Vercel AI SDK, CrewAI, AutoGen, Continue.dev,
  Cursor, Aider, or Chatbot UI works unchanged.
- **200+ models** across 20+ providers (full live list at
  https://celedog.io/models).
- **Auto-routing meta-models** —
  `celedog/auto-coding`, `celedog/auto-reasoning`,
  `celedog/auto-chat`, `celedog/auto-chinese`, `celedog/auto-cheapest`.
- **SSE streaming** — `stream: true` works identically across every
  underlying provider; Celedog normalises the stream format.
- **Function calling / tool use** — OpenAI function-calling shape
  is translated automatically to each provider's native format
  (Anthropic tool_use, Gemini functionDeclarations, etc.).
- **Vision / multimodal** — `image_url` content blocks supported on
  GPT-5.5, Claude 4.6 Sonnet, Gemini 2.5 Pro, Pixtral Large,
  Llama 4 Maverick.
- **Embeddings** — OpenAI `text-embedding-3-small`,
  `text-embedding-3-large`, and others available through the same
  `/v1/embeddings` endpoint.
- **Multi-currency billing** — USD, CNY, IDR are first-class. The
  customer's locale dictates display currency; underlying credit is
  held in a single platform unit.
- **Multiple checkout rails** — Stripe (cards, USD/EUR), Xendit
  (Indonesia, IDR), Alipay, WeChat Pay v3, Airwallex, Creem, Waffo.

## Featured models (illustrative — see /models for the live list)

| Provider | Featured models |
|----------|----------------|
| OpenAI   | GPT-5.5, GPT-5.4, GPT-5.3-Codex, o3, o4-mini |
| Anthropic | Claude 4.7 Opus, Claude 4.6 Sonnet, Claude 4.5 Haiku |
| Google   | Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 3 Pro Preview |
| Meta     | Llama 4 Scout, Llama 4 Maverick, Llama 3.3 70B |
| DeepSeek | DeepSeek-V4-Pro, DeepSeek-V4-Flash, DeepSeek-R1 |
| Mistral  | Mistral Large, Mistral Small, Pixtral Large |
| xAI      | Grok 3 |
| Qwen     | Qwen3.7-Max, Qwen3.6-Plus, Qwen-Max |
| Amazon   | Nova Pro, Nova Lite |
| Cohere   | Command R+, Command R |
| Moonshot | Kimi K2.5, Kimi K2.6, Moonshot v1 (8k/32k/128k) |

## API base URL

```
https://celedog.io/v1
```

OpenAI-compatible. Customer authenticates with
`Authorization: Bearer sk-…` (or, for the Anthropic-native
endpoint at `/v1/messages`, `x-api-key: sk-…`).

## SDK quickstart (OpenAI Python)

```python
from openai import OpenAI

client = OpenAI(
    api_key="sk-YOUR-KEY",
    base_url="https://celedog.io/v1",
)

response = client.chat.completions.create(
    model="gpt-4o",  # or claude-haiku-4-5, gemini-2.5-pro, etc.
    messages=[{"role": "user", "content": "Hello, Celedog!"}],
    stream=True,
)

for chunk in response:
    print(chunk.choices[0].delta.content or "", end="")
```

## SDK quickstart (Anthropic Python)

```python
from anthropic import Anthropic

client = Anthropic(
    api_key="sk-YOUR-KEY",
    base_url="https://celedog.io",
)

response = client.messages.create(
    model="claude-haiku-4-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}],
)
```

## Pricing model

- **Subscribe monthly for the best rate, or buy once for full flexibility.**
- **Monthly packs** — Developer, Pro, Business tiers as recurring monthly
  subscriptions; each renewal grants premium tokens at the best per-token
  rate. This is the default, lowest-cost path.
- **One-time packs** — the same packs bought once (priced at ~10% above the
  monthly equivalent) for users who don't want a recurring charge.
- **Pay-as-you-go** — fund a wallet and get charged per token, no
  commitment, for fully usage-based billing.
- **Per-model rates** — see https://celedog.io/pricing for the live
  per-model table. Input and output tokens are billed independently
  per model.
- **Multi-currency display** — prices render in USD, CNY, or IDR
  matching the customer's locale; the platform tracks USD-equivalent
  internally.

### Representative per-token prices (USD per 1M tokens)

Approximate ranges by capability tier. These are illustrative; the live,
exact per-model rates (and currency conversion) are at
https://celedog.io/pricing and in the server-rendered table on that page.

| Tier     | Example models                              | Input ~$/1M | Output ~$/1M |
|----------|---------------------------------------------|-------------|--------------|
| Budget   | gpt-4o-mini, deepseek-chat, gemini-2.5-flash | 0.10–0.60   | 0.30–2.50    |
| Mid      | gpt-4o, claude-haiku-4.5, deepseek-v3        | 0.50–3.00   | 2.00–12.00   |
| Frontier | gpt-5.5, claude-opus-4-7, gemini-2.5-pro     | 3.00–15.00  | 10.00–75.00  |

Celedog's published user-facing rates are set to sit at or below the
underlying provider's official list price; auto-routing (`celedog/auto-cheapest`)
can cut an effective bill a further 20–40% by sending easy requests to a
cheaper model that still clears the quality bar.

## Locales

- `en` — English (default)
- `zh` — Chinese (Simplified); prices in CNY at the admin-set
  `USDExchangeRate`
- `id` — Bahasa Indonesia; prices in IDR at the Xendit-aligned rate

## Compatibility matrix

| Tool / library | Works out of the box? |
|----------------|----------------------|
| OpenAI Python / Node / Go / Java SDK | ✓ — change `base_url` only |
| Anthropic Python / Node SDK | ✓ — change `base_url`, use `x-api-key` |
| LangChain (OpenAI provider) | ✓ |
| LlamaIndex | ✓ |
| Vercel AI SDK | ✓ |
| CrewAI / AutoGen / Semantic Kernel | ✓ |
| Continue.dev / Cursor / Aider | ✓ — point at custom OpenAI endpoint |
| Chatbot UI | ✓ |
| Dify / Flowise / n8n | ✓ — configure custom model provider |

## Data privacy

- TLS 1.3 in transit, encrypted at rest for wallet / account data.
- Celedog does **not** use customer prompts or completions to train
  models. Requests are passed through to the selected provider for
  inference only.
- Each underlying provider's privacy policy governs the inference
  layer. For HIPAA workloads, choose a provider that offers BAA.

## Rate limits

Aggregated across providers — Celedog's effective limits are higher
than any single provider direct, because the same request can fall
back to an alternative provider when one is throttled. Per-account
limits scale with pack tier (Developer / Pro / Business) and are
configurable in the customer dashboard.

## Status and support

- Status page: https://celedog.io/status
- Support: contact form in the customer dashboard
- Documentation: https://celedog.io/docs
- FAQ (bilingual EN/CN): https://celedog.io/faq
- Comparison pages: https://celedog.io/compare/

## Open source / fork heritage

Celedog is a customer-facing fork of NewAPI (https://github.com/QuantumNous/new-api),
licensed under AGPL-3.0. Source-file copyright headers and the
LICENSE retain the upstream attribution; customer-visible branding
is Celedog.

## Last updated

2026-05-28.