GitLab Duo Agent Platform Custom Agent Deployment for GitOps Automation
The GitLab Duo Agent Platform custom agent deployment workflow is helping engineering teams automate one of the most repetitive parts of Kubernetes operations: onboarding new microservices into GitOps environments. Instead of spending hours updating manifests, delivery pipelines, namespaces, and Flux automation manually, teams can now build AI agents inside GitLab that handle the entire process automatically while keeping everything auditable and version-controlled.
Every engineering organization has repetitive operational tasks that consume valuable developer time. In GitOps environments, onboarding a new microservice is one of them. Developers must generate Kubernetes manifests, configure deployment pipelines, update image automation rules, and ensure every hostname, namespace, and port matches existing conventions. As a result, a single missed configuration can break deployments and delay releases.
However, GitLab is now addressing this problem through its Duo Agent Platform. The platform allows organizations to create custom AI agents trained on their own repository structures, pipeline standards, and deployment workflows. Consequently, platform teams can automate complex onboarding tasks without sacrificing governance or visibility.
GitLab demonstrated the workflow using a fictional company called TanukiBank. The company operates a Kubernetes-based GitOps environment where every microservice uses its own CI/CD pipeline. Meanwhile, a Flux Image Automation Controller continuously monitors container registries for new images and updates deployment manifests automatically. In addition, a Flux CD Controller ensures Kubernetes pods stay synchronized with the desired state stored in Git repositories.
Because all Flux manifests are stored in a dedicated Flux Config project, onboarding a new service normally requires coordinated updates across several repositories and delivery systems. Therefore, the GitLab Duo Agent Platform custom agent deployment model focuses on automating these multi-project changes consistently.
To build the agent, a developer first opens GitLab Agentic Chat within the TanukiBank group and instructs it to study the organization’s repositories, manifests, Dockerfiles, and configuration files. After analyzing the environment, Duo generates a detailed system prompt describing workflow logic, naming conventions, and recommended tools. That prompt effectively becomes the operational knowledge base for the new AI agent.
Next, the developer creates a managed project named application-agents and defines a custom agent called TanukiBank Microservice Onboarder. The recommended tools are enabled, the generated system prompt is added, and the agent is configured for both GitOps repositories: Tanuki Bank – Delivery and Flux Config.
Once a new microservice named intra-account-transfers is ready and its container images are available in GitLab’s registry, the onboarding process begins. The developer opens Agentic Chat inside the delivery project, selects the custom onboarding agent, and provides the service name together with its hostname.
The agent then performs several tasks automatically. First, it locates the service Dockerfile to identify the correct application port. Next, it generates the required Kubernetes manifests and updates the delivery pipeline configuration. Afterward, it requests user approval before committing changes or opening merge requests. Finally, the agent creates two separate merge requests across the GitOps projects and provides a structured deployment summary with links, modified files, and recommended follow-up actions.
After both merge requests are merged, the Kubernetes environment inside GitLab immediately reflects the new deployment state. A kubectl pod listing confirms that three new pods are running successfully. Furthermore, the public service endpoint responds correctly, proving the onboarding process completed successfully.
The GitLab Duo Agent Platform custom agent deployment approach significantly reduces operational overhead for engineering teams. Tasks that previously consumed hours can now finish within minutes. More importantly, organizational knowledge such as Flux conventions, naming standards, and deployment patterns becomes reusable through centralized AI agents.
Because the agents operate entirely inside GitLab, platform teams also maintain full governance and audit visibility. Every generated artifact, merge request, and configuration change remains version-controlled within the same DevSecOps environment. As AI-assisted software delivery continue evolving, GitLab’s approach demonstrates how enterprise teams can automate infrastructure workflows while still preserving security, compliance, and operational control.
