Create, store, rotate, and revoke API keys

Create and manage API keys on the token management page. A token can spend up to your whole account balance, so give every token limits such as a quota and keep it on the server side. If a key may be exposed, disable it first, then review call logs. Based on the Terms and Data Policy, last verified October 5, 2026.

Create an API key

Sign in to the LaoZhang API console.
Open token management.
Select a name, group, credit limit, or other restrictions currently available in the console.
Copy the new key and store it securely.
Validate the endpoint, model, and billing group with a minimal request.

Do not send a complete API key in documentation, tickets, or public chats. Support normally needs the account, request time, model, error, and a redacted key identifier.

Recommended storage

Environment variable

Set the key as an environment variable, then read it in code. The Python example uses the OpenAI SDK (pip install openai):

export LAOZHANG_API_KEY="YOUR_LAOZHANG_API_KEY"

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["LAOZHANG_API_KEY"],
    base_url="https://api.laozhang.ai/v1",
)

For production, prefer a cloud secret manager, container Secret, or encrypted deployment-platform environment variable. Do not print keys with echo, debug logs, or exception messages.

Limit each token

When you create or edit a token, you can set four limits:

Limit	What it does	Use it for
Quota	With unlimited quota off, the token can spend at most this amount	Every token; always for test tokens
IP allowlist	Only requests from the listed IPs or ranges can use the token	Production servers with a fixed egress IP
Allowed models	The token can call only the listed models; empty means no limit	Single-purpose tokens and tokens you hand to others
Expiry	The token stops working after this time	Short-lived jobs and tokens for AI tools

Quota matters most. If a token with unlimited quota leaks, you can lose your entire balance. A quota is only a spending cap and doesn’t reserve any of your balance; all tokens draw from the same account balance.
IP allowlist takes your server’s public egress IP, as a single address or a CIDR range such as 203.0.113.0/24. Home and office connections change their egress IP, and every call fails once it changes, so use a quota there instead.
Allowed models rejects any model not on the list. Use IDs exactly as they appear in the model catalog. For everyday use and testing, leave it empty so switching models doesn’t require a console change.

Use separate tokens for development, testing, and production, and for each service or automation, with names you can recognize, such as prod-image-service. Then you can disable one without affecting the rest. Never embed a server API key in a browser, mobile app, or public client.

Group, fallback groups, and billing mode: what’s the difference?

When you create a token, its group, fallback groups, and billing mode are three separate settings. Usage first (按量优先) is a billing mode, not a group.

Setting	What it decides
Group (the primary group)	Which group serves the request, which models the token can call, and which multiplier applies
Fallback groups (optional)	Groups tried in order only when the primary group can’t serve the model
Billing mode	Usage first, Usage-based, or Per-call: whether the token can call usage-based models, per-call models, or both

Whenever the primary group can serve the requested model, the primary group handles the request. Adding a fallback group doesn’t send requests there.
To use a model or billing method that only an official-forward group offers, such as token-based billing for official-forward GPT Image, create a token whose primary group is that group. Setting it only as a fallback group isn’t enough.
Usage first is the right billing mode in most cases. See token billing modes for the models each mode can call, and token groups and multipliers for what each group is for.

Keep keys out of these places

Repositories: a key committed to Git stays in the history even after you delete the file. Add .env to .gitignore and scan before you commit:
grep -rnE '(^|[^A-Za-z0-9])sk-[A-Za-z0-9]{10,}' .
Docs, screenshots, and logs: console screenshots, terminal recordings, and error logs can all contain a full key, so use placeholders in examples. Public repositories are scanned by bots around the clock, and leaked keys can be abused quickly.
Conversations with AI tools: a key pasted into an AI coding tool usually stays in plain text in local session history and file snapshots. Give AI tools a separate short-lived token with a small quota, delete it when the task is done, and never hand over a production key.

Rotate and revoke

Rotate or revoke immediately when:

a key appears in Git, logs, screenshots, tickets, or public messages;
unexplained requests or balance changes occur;
employee, contractor, or system access changes;
a dependency, server, or deployment credential is compromised;
support or security personnel request action.

Safe rotation sequence:

Create a new key and validate it in a controlled environment.
Update the server-side secret.
Observe calls using the new key.
Revoke the old key.
Review call logs for the affected period.

Your organization should set rotation intervals based on its own risk and audit requirements. The documentation does not promise a universal rotation period or unlimited credit.

Investigate unexplained usage

If a token shows calls you don’t recognize:

Find where the key went

Search for the key in chat history, repositories and their commit history, and environment variables on deployment and CI platforms. Most unexplained usage turns out to be an old configuration that’s still running.

Disable it if you can't find the source

Disable the token in token management. Tokens are independent, so disabling one doesn’t affect the others; create a replacement once you know you need it.

Trace the caller in call logs

Filter call logs by the token and look at the models, times, and source IPs. Watch for models you don’t use, calls outside your working hours, and volumes that don’t match your traffic.

Lock down the account

Change the account password and limit who can sign in to the console. Staff who only need to check usage don’t need console access.

API key vs. personal access token

Normal model calls use an API key. Account endpoints such as balance and log export use a personal access token (starts with nap_); the two credentials are not interchangeable. The old system AccessToken was retired on November 9, 2026, so create a personal access token under Security & Access in your personal settings; see Create a personal access token.

What to provide to support

Provide the account, key creation time, redacted key prefix and suffix, suspicious-call time, model, and error. Do not send the complete API key. If the key was exposed, revoke or rotate it before continuing the investigation.

​Create an API key

​Recommended storage

​Environment variable

​Limit each token

​Group, fallback groups, and billing mode: what’s the difference?

​Keep keys out of these places

​Rotate and revoke

​Investigate unexplained usage

​API key vs. personal access token

​What to provide to support

​Related documentation

Create an API key

Recommended storage

Environment variable

Limit each token

Group, fallback groups, and billing mode: what’s the difference?

Keep keys out of these places

Rotate and revoke

Investigate unexplained usage

API key vs. personal access token

What to provide to support

Related documentation