London-based platform engineering consultancy ClearRoute has released runkora, an open-source control plane designed to manage the identities, permissions and activities of autonomous AI agents operating across enterprise software engineering systems.
The software, developed by ClearRoute’s Australian engineering team, has been released in alpha following its introduction at PlatformCon Live Day in Sydney. It is designed to give organisations a common set of controls for AI agents working with development infrastructure, repositories and other enterprise services.
Runkora runs existing agent harnesses within a Kubernetes environment rather than requiring developers to adopt a separate agent framework. Its published implementation supports Claude Code, Codex and OpenCode, allowing different agents to operate under a common identity, access and execution structure.
Each agent receives a verifiable workload identity, with permissions associated with the tasks it is authorised to perform. When an agent acts on behalf of an employee, its delegated authority is restricted to the access available to that person, limiting the actions it can undertake independently.
Policy enforcement applies to the tools and resources an agent can use. The platform is designed to evaluate actions against defined rules before execution, allowing administrators to control access to repositories, services and other systems according to the agent’s assigned responsibilities.
Credential management is handled separately from the agent’s execution environment. Instead of placing passwords or access tokens inside a working session, runkora uses controlled gateways to supply the credentials needed for authorised requests. This reduces the need to distribute persistent secrets across autonomous workloads.
Agents execute within isolated environments that can be discarded when their work is complete. Activity records are retained, providing information about the tasks performed, resources accessed and actions taken during a particular run.
The platform also supports coordinated groups of agents, known as agent factories, in which different agents perform related tasks towards a shared objective. Work can begin through schedules, application interfaces, queues or human instructions, while approval points allow employees to supervise selected activities.
ClearRoute is positioning the technology around the wider software delivery process, rather than AI-assisted code generation alone. Engineering teams must still manage testing, security checks, change approvals, deployment and ongoing system operations after code has been produced.
James Jarvis, chief executive of ClearRoute, said: “AI might help developers produce code faster, but unless businesses address the engineering processes around it, they risk accelerating development without actually improving their time to market.”
The consultancy’s Route to Live research has identified approval processes that can take four to eight weeks at some organisations. The introduction of autonomous agents creates additional requirements for access control and accountability when software tasks involve connected production systems.
Runkora’s implementation uses Kubernetes alongside infrastructure components including Istio and SPIRE, which support network controls and workload identity. Deployment documentation also describes requirements for supporting databases, caching and storage services.
The project is distributed under the Apache 2.0 licence, allowing organisations to inspect, use and modify its code under the terms of that licence. ClearRoute also offers engineering support for organisations seeking to integrate the platform into their existing infrastructure.
As an alpha release, runkora is available for technical evaluation while its development continues. The public repository includes installation instructions, architecture documentation and example workflows that engineering teams can use to examine how identity, permissions and isolated execution operate across autonomous software tasks.
You must be logged in to post a comment.