Credential security
- Tokens should live only in server-side environment variables or a secret manager; never commit them to public repositories or embed them in front-ends or mobile app bundles.
- Create separate tokens for different business systems for easier permission control and fault isolation.
- When a token is leaked or suspected leaked, immediately disable or rotate it in the console.
- The three credential types (model token, profile access token, query authorization token) have different purposes and security boundaries — never mix them (see Authentication).
Access control
- The platform uses a three-level role model (owner / administrator / member) and the principle of least privilege.
- Query authorization tokens can limit the readable scope to current user / specified users / all users, and can bind only read-only permission points (see Query authorization).
- IP whitelists restrict token calls to trusted sources.
Rate limiting & protection
- Model calls are rate-limited by RPM (requests per minute) / TPM (tokens per minute) to prevent abuse and misconfiguration.
- Hitting the limit returns
429; retry with a backoff strategy.
Data & audit
- Call logs and key operations are traceable for auditing and troubleshooting.
- Log exports contain only your account’s (or authorized scope’s) data; sensitive fields are desensitized per permission.
- Desensitize any sensitive information in logs you share with support.
Compliance recommendations
- Assess data-cross-border and model-capability compliance against the laws and regulations of your business region.
- Review the “pre-launch checks” in the Quickstart before going to production.