Skip to content
NoviaHub Docs

Guides

Base URLs and protocols

Which protocol to pick for your tool and model, which base URL to enter, and which header carries the API key.

NoviaHub offers three API protocols, that is, three request formats: OpenAI, Anthropic (Claude) and Google Gemini. They all use the same API key and differ only in the address and the request format. This page helps you decide which one to use.

The short answer

Section titled “The short answer”
What you are using Protocol Base URL
Claude Code, or another tool that only supports the Anthropic API Anthropic Messages https://noviahub.com
Codex OpenAI Responses https://noviahub.com/v1
Google’s official Gen AI SDK (google-genai / @google/genai) Gemini https://noviahub.com
The official OpenAI SDK, and most apps that “support OpenAI-compatible APIs” OpenAI Chat Completions https://noviahub.com/v1

If you are unsure, choose OpenAI Chat Completions: as of 2026-09-29, every model on NoviaHub supports it.

For step-by-step setup of each tool, see Integrations.

Which protocols a model supports

Section titled “Which protocols a model supports”

Open Models & Pricing and click a model. The labels below its name are its endpoint labels:

Label Protocol and path
Chat OpenAI Chat Completions, POST /v1/chat/completions
Response OpenAI Responses, POST /v1/responses
Anthropic Anthropic Messages, POST /v1/messages
Gemini Gemini, POST /v1beta/models/{model}:generateContent
Image OpenAI Images, POST /v1/images/generations
openai-response-compact, openai-alpha-search Two additional endpoints; see Responses · Responses Compact and /v1/alpha/search

Calling a model only through the protocols in its labels gives the most reliable results.

The models online on 2026-09-29, as an example (models are added and removed, so check the model page):

  • Every model has Chat.
  • Claude models: Anthropic, Chat.
  • Gemini models: Gemini, Chat.
  • GPT models: Chat, Response, Anthropic.
  • glm-5.3-flash: Anthropic, Chat.
  • DeepSeek models and kimi-k3: Chat, Response, Anthropic and Gemini, plus openai-response-compact and openai-alpha-search.
  • gpt-image-2: Image, Chat.

When you call a model through a protocol other than its own, NoviaHub converts the format in between. Conversion brings some differences, for example a field that has no counterpart in the other protocol. See Protocol conversion.

Should the base URL end in /v1?

Section titled “Should the base URL end in /v1?”

This is the easiest thing to get wrong, because each vendor’s SDK builds the path differently:

Protocol Enter What the SDK actually requests
OpenAI (Chat, Responses, Images) https://noviahub.com/v1 https://noviahub.com/v1/chat/completions and so on
Anthropic https://noviahub.com The SDK appends /v1/messages
Gemini https://noviahub.com The SDK appends /v1beta/models/...

What you see when it’s wrong:

  • An extra /v1 on the Anthropic or Gemini address: the request goes to an address such as /v1/v1/messages or /v1/v1beta/models/... and gets HTTP 404 with a message like Invalid URL (POST /v1/v1/messages).
  • A missing /v1 on the OpenAI address: the request goes to https://noviahub.com/chat/completions. That address is not an API, so NoviaHub answers HTTP 200 with a whole web page (HTML). Programs usually report that they can’t parse JSON, or an error like Unexpected token '<'. If you see one of these, first check that the address ends in /v1.
  • The full path entered as the base URL (for example https://noviahub.com/v1/chat/completions): the SDK appends the path again, and the result is again a 404 Invalid URL.

Which header carries the API key

Section titled “Which header carries the API key”
Header Works on
Authorization: Bearer <API key> Every endpoint
x-api-key: <API key> Only /v1/messages and /v1/models
x-goog-api-key: <API key>, or ?key=<API key> added to the URL Only /v1beta/models/... and /v1/models

With the wrong header, NoviaHub can’t find the key and returns HTTP 401 Invalid token, for example when you send x-api-key to /v1/chat/completions.

The official SDKs send the right header on their own; just give the key to the SDK. If you build HTTP requests yourself, Authorization: Bearer is the safe choice everywhere.

Next steps

Section titled “Next steps”