IronClaw

IronClaw is an open-source secure runtime designed to run AI agents inside hardware-backed encrypted enclaves on the NEAR AI Cloud. Its primary purpose is to provide strong isolation, confidentiality, and integrity guarantees for AI workloads, ensuring that models, prompts, and data remain protected even from underlying infrastructure operators. By combining confidential computing with a runtime tailored for autonomous agents, IronClaw enables organizations to deploy powerful AI systems without sacrificing security or control over sensitive assets.

At its core, IronClaw provides a sandboxed execution environment where AI agents can operate with tightly scoped permissions, audited access to external resources, and verifiable runtime integrity. It integrates with NEAR AI Cloud to leverage secure enclaves for encryption-in-use, preventing unauthorized access to code and data during execution. IronClaw supports policy-based control over what agents can read, write, or call, including network access, file systems, and external APIs, reducing the risk of data exfiltration or unintended actions. Open-source transparency allows security teams to inspect, audit, and extend the runtime to meet internal compliance and governance requirements.