AI API GETTING STARTED

From sign-up to your first successful API call

No generic AI theory. This guide helps you create a Zivv API key, choose a model, and complete your first request in Claude Code, Codex, or a desktop client.

About 6 minutesNo API experience requiredBased on the current setup flow

01 / FOUR THINGS TO KNOW

What does an API call need?

Every tool needs the same four building blocks. Understanding them makes setup and troubleshooting much easier.

Zivv account

Used to add balance, create keys, review usage, and manage billing.

API key

A credential beginning with sk-. It represents your account, so never publish or share it.

API endpoint

OpenAI-compatible tools normally use https://zivv.pro/v1; Claude Code uses https://zivv.pro.

Model name

The model determines capability, speed, and cost. Copy a current name from Model Hub.

02 / CHOOSE A MODEL

Choose for the task, not by price alone

Models change quickly. Treat Model Hub as the source for current names and pricing, then adjust based on your own results.

Claude

Claude Code, complex coding, long documents

Start with a Claude Sonnet model for coding; move to a stronger option only when the task needs it.

GPT

Codex, OpenAI SDKs, general tasks

Choose a current GPT model for OpenAI-compatible tools and broad everyday work.

Gemini

Long context, images, multimodal input

Choose Gemini when you need to process large source material or mixed text and images.

View current models and pricing

03 / CREATE A KEY

Get ready in four steps

1

Create an account

Register with an email address you regularly use, then sign in.

Sign up
2

Add a small balance

Test a real task with a small amount before committing more.

Add balance
3

Create an API key

Open API Keys and name the key after its purpose so usage stays easy to track.

Create key
4

Copy a model name

Copy the exact name from Model Hub instead of typing it from memory.

Choose model

Treat every API key like a password

Never put a key in a public repository, screenshot, chat, or browser application. If exposure is possible, delete it and create a new one immediately.

04 / MAKE YOUR FIRST CALL

Choose how you want to connect

These options use the same Zivv account, but their endpoint settings differ.

Claude Code

Choose a Claude-compatible group when creating the key, set these two environment variables, reopen your terminal, and start Claude Code.

export ANTHROPIC_BASE_URL=https://zivv.pro
export ANTHROPIC_AUTH_TOKEN=sk-your-key

Commands differ between Windows PowerShell, macOS, and Linux. Use the full guide for your operating system.

Claude Code setup guide

05 / TROUBLESHOOTING

Four common problems

401

Invalid key

Check for missing characters or spaces and confirm the key still exists. Your account password is not an API key.

404

Wrong endpoint or model

OpenAI tools normally need /v1; Claude Code does not. Copy the exact model name from Model Hub.

429

Request temporarily limited

Reduce concurrency and retry later. If it continues, check the status page or contact support.

Balance

Calls have stopped

Check account balance and key limits, then review Usage to see whether the latest request was recorded.

Still stuck? Send support the time, tool, model name, and error message, but never send your complete API key.

Contact support

06 / FAQ

Before you start

Do I need a separate key for every model?

Usually not. A key can call multiple models allowed by its group and permissions. The exact range depends on the settings used when creating it.

How much balance should I add?

Start small and test your real workload. Confirm model quality, speed, and cost before adding more.

Why can I sign in on the web but get a 401 in my tool?

Website sign-in and API keys are different credentials. Use the sk- key created on the API Keys page, not your account password.

Can I put my key in a browser application?

Do not put a private key in public frontend code. Use secure local storage for personal tools, and separate keys and limits for team members.

Make your first API call now

Create one key and test a small, real task. Once it works, tune the model and spending limits from actual usage.