How to connect OpenClaw to the Tokenator API

OpenClaw runs on your own server and reaches the model over plain HTTP. As long as the tool reads the base URL and key from environment variables, Tokenator plugs in without code patches.

Get an API key API documentation

What you end up with

Once configured, OpenClaw stops talking to its default provider and sends every request to Tokenator. Billing comes out of your key's token limit, and the list of available models is defined by the key rather than by the tool.

Tool	OpenClaw
Protocol	OpenAI-compatible
Base URL	`https://api.tokenator.top/v1`
Key	`sk-your-tokenator-key`

Requirements and supported operating systems

Windows 10/11, macOS 12+, or a modern 64-bit Linux.
Outbound HTTPS access to the Tokenator domain.
A Tokenator API key that looks like sk-your-tokenator-key.

You need the machine or server where you deploy the agent, plus the ability to set its environment variables.

Installing OpenClaw

Deploy the agent following its own repository instructions — Tokenator does not affect the install method and plugs in when you configure the model provider.

Setting up Tokenator

Register in your account and buy a token bundle.
The key appears in the account right after the payment is confirmed — copy it whole, including the prefix.
The examples below use the placeholder sk-your-tokenator-key. Never publish a real key in repositories or screenshots.

Configuration on Windows

PowerShell

$env:OPENAI_BASE_URL = "https://api.tokenator.top/v1"
$env:OPENAI_API_KEY = "sk-your-tokenator-key"

Anthropic-compatible variant

$env:ANTHROPIC_BASE_URL = "https://api.tokenator.top/anthropic"
$env:ANTHROPIC_AUTH_TOKEN = "sk-your-tokenator-key"

Exact variable names depend on the agent version and change between releases — check its own documentation. On the Tokenator side only three things are fixed: the base URL, the key and the model ID from the catalog.

Configuration on macOS and Linux

bash / zsh

export OPENAI_BASE_URL="https://api.tokenator.top/v1"
export OPENAI_API_KEY="sk-your-tokenator-key"

Anthropic-compatible variant

export ANTHROPIC_BASE_URL="https://api.tokenator.top/anthropic"
export ANTHROPIC_AUTH_TOKEN="sk-your-tokenator-key"

Your first request

Before starting the tool it helps to confirm that the key and the model both work. A single curl request answers both questions.

OpenAI endpoint check

curl https://api.tokenator.top/v1/chat/completions \
  -H "Authorization: Bearer sk-your-tokenator-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "messages": [{"role": "user", "content": "ping"}]
  }'

Start the agent the usual way — with the variables in place it will route through Tokenator.

A successful response means the key is valid, the model is allowed for it and the upstream is reachable. Usage appears in your account right after the request.

Switching models

The model is the Model ID field in provider settings. The base URL and key stay the same when you switch models.

Examples of model IDs available right now:

claude-haiku-4-5
claude-opus-4-6
claude-opus-4-7
claude-sonnet-4-6
gpt-5.5
deepseek-v4-flash

Current IDs and specs live in the AI model catalog, which also shows which models are available right now. The model comparison helps choose between close options.

Common errors and how to fix them

Symptom	Cause	Fix
`401` or an invalid-key message	The key was copied with stray spaces, was revoked, has expired, or landed in the wrong environment variable.	Check the key in your account and make sure the environment variable actually reached the process running the tool.
`403`	The model is not in the key's allowed-model list.	Check the key's model list in your account and pick an ID from the model catalog.
`429`	The per-minute or per-hour request limit, or the concurrent-stream limit for the key, was exceeded.	Lower the agent's concurrency and retry with exponential backoff. Key limits are visible in your account.
Model not found / not supported	The request used an ID that is not in the catalog, or an alias unavailable to this key.	Copy the exact API ID from the model page — case and dots matter.
The request hangs or breaks on a long answer	The client closed the connection on its own timeout before the model finished generating.	Turn on streaming (`stream: true`) and raise the client timeout: with long reasoning answers the first token can take a while.
A bare "Request error" with no details	The upstream provider returned an error. Its text is not passed through to the client.	Retry: Tokenator fails over to the next provider for the model on its own. If it persists, contact support.

FAQ

Do I need a OpenClaw subscription or a separate provider account?

No. The tool talks to Tokenator, and billing comes from your Tokenator key's tokens.

Where do I see usage after connecting?

In your account: it shows requests, spent tokens and the key's remaining limit.

Can one key be used in several tools?

Yes. The limits are the key's own: requests per minute and hour, plus the number of concurrent streams.

What happens if tokens run out mid-session?

Requests start returning an error. Topping up the same key restores the limit — no need to reinstall the tool.

How do I switch back to the previous provider?

Remove the settings you added — the environment variables or the provider block in the config. The tool reverts to its default behaviour.

Where can I read about the API format itself?

In the Tokenator API documentation and on the OpenAI-compatible API page.