Agentic artificial intelligence identity startup AgentField today is launching its platform as an Apache 2.0 open-source project so it can do for AI agents what Kubernetes did for software containers.

Its mission is to transform AI agents into “scalable and governed production infrastructure” by solving what it says is a “coordination crisis” in autonomous software. To fix this, it assigns each AI agent with a cryptographic identity, making it easier for teams to set enforceable permissions for each one, together with mathematical proof of authority for access to sensitive information.

AgentField says there’s an urgent need for identities in autonomous software. The problem is that most AI agents tend to be specialized for one specific task, which means they’re required to collaborate with other AI agents to get things done.

However, that causes problems, because when one agent assigns a task such as updating a patient’s record or moving funds to another agent, it sets off a cascade of automated decisions that cannot be tracked. Traditional authorization tools such as OAuth and IAM struggle with this because they were built for deterministic human workflows and rely on broad application programming interface keys and synchronous checks. But they’re insufficient for autonomous, agentic workflows.

AgentField Chief Executive Oktay Goktas said AI agents transform information technology infrastructure into digital workforces that rapidly multiply. “They don’t log in once a day,” he said. “They run constantly in the background, fan out across hundreds of APIs and execute critical operations at machine speed. You cannot manage them with human-centric SSO or API keys. The identity and orchestration layer has to be rebuilt.”

The startup’s solution is to issue each agent a decentralized and tamper-proof cryptographic identity and combine this with Kubernetes-style orchestration. That means that a supply chain agent, for example, can show cryptographic proof that it has been authorized by a specific policy, and that each action it takes can be traced via a chain of agents to a human or system decision.

Security teams also receive cryptographic receipts of each action taken by an AI agent. The entire platform has been made open-source, because the startup’s founders deem open standards to be critical for the emerging “agent economy.”

“When a $250,000 transfer happens at 3:17 a.m., you need to know exactly who authorized it and under what policy,” AgentField Chief Technology Officer Santosh Kumar Radha explained. “AgentField ensures that even if an agent functions autonomously five hops down a chain, its authority can be mathematically verified.”

By doing this, AgentField acts as a centralized control plane for autonomous software, similar to how Kubernetes does the same for software containers. On the orchestration side, it provides operational primitives including long-running execution to support agents that work autonomously for hours on end and asynchronous execution for handling multi-agent workflows. That means it can directly enforce access control and identity access management policies.

Holger Mueller of Constellation Research said AgentField’s ability to create identities for agents and orchestrate them is likely to interest a lot of enterprises, because they ultimately want to use the technology to reinvent enterprise automation, without any restrictions. “The introduction of traceability and digital breadcrumbs is going to be a critical capability going forward in the agentic AI era, and enterprises will need help to do this,” the analyst said.

The startup said its platform is designed to work with any autonomous workflow, but is most useful for high-stakes industries such as finance, healthcare, energy, insurance and logistics. In these industries, small errors can escalate rapidly to become very serious security problems that AgentField can help to prevent.

Enterprises can access AgentField on GitHub, where it’s now available under an Apache 2.0 license.

Image: SiliconANGLE/Dreamina