Configure OpenClaw with a cheap API: point the provider at the gateway and it runs
OpenClaw lets you declare custom model providers. Declare the Cho Tai Khoan gateway as an OpenAI-compatible provider and your agent can call 26 models (Claude, GPT, Gemini, DeepSeek…) with a single key, paying a fixed package from $0.72 (≈ 19.000đ).
Step 1 — Get a key first
Go to the API gateway home page, pick a package and pay (USDT, VietQR or MoMo) — the system issues a redeem code automatically.
Redeem the code right on the page and receive a key starting with sk-ctk-.
Step 2 — Declare the provider in OpenClaw
Open the model / provider configuration of the OpenClaw instance you run.
Add a custom provider of the OpenAI Compatible type with 3 details:
Base URL:https://api.chotaikhoan.com/v1
API Key: your sk-ctk-... key
Model: the name of the model you want the agent to use — for example claude-sonnet-5, gpt-5.5 or deepseek-v4-pro (copy the exact name from the Models & pricing section on the home page)
Set this provider as the default for your agent (or select it when creating a new session), then save the configuration.
After saving, restart OpenClaw / the agent so the new configuration takes effect — if you edit the configuration without restarting, the agent keeps running the old one.
Field names may differ slightly between OpenClaw versions — the principle never changes: a custom provider + the Base URL above + your key + the exact model name.
Step 3 — Check
Send your agent a simple message, for example: "Reply with exactly one word: running". If the agent answers, you are running through the gateway. Then open the Check key page on the gateway to cross-check the tokens just charged.
Common errors
The agent reports an authentication error: the key is missing characters, or the provider has not been set as the default.
The agent still uses the old provider: the new provider has not been set as the default, or you have not restarted after editing the configuration.
Model not found: the model name is typed wrong — copy the exact name from the Models & pricing page again.
Out of quota: redeem another code — the quota stacks onto your old key, validity follows the new package.
System busy: a temporary error, and that request was not charged. Wait a few minutes and try again.
An agent running 24/7 consumes tokens steadily — start with the Max or Team package to avoid interruptions, and check your balance regularly on the Check key page. See the API pricing page to pick a package that matches your usage rhythm.
One key for a long-running agent
OpenClaw eats tokens by the hour — the Max package at 139.000đ / 500M tokens is a sensible starting point, and you can stack more onto the same key when it runs out.