How we run untrusted customer code at scale

Nango, a platform for building API integrations, runs over 150 million customer-written functions monthly and evolved its security architecture to safely execute untrusted code at scale. After discovering that in-process sandboxes were vulnerable to escape attacks, the company transitioned to isolated runner services with no direct database access, and later migrated to AWS Lambda to achieve better resource isolation and per-execution observability. Each approach prioritized security while balancing competing demands like minimizing cold starts, supporting long-running jobs, and handling unpredictable traffic spikes.