Agents inside your network, on your models
The runtime is a container you run. We hold your organization, members and licence; never your data.
AgentKube is licensed software you deploy in your own cloud account or data center. The builder, the agents, the conversations and the integrations run there. The only outbound traffic to AgentKube is a check-in with versions and counts.
Integrate and build
Embed agents in your apps and intranet with one element. Your teams build them in AgentKube Designer; your developers extend them with tools and MCP servers.
Deploy in your cloud
One image runs on Docker, Podman, Cloud Run, App Service and AWS. Environments per place it runs, with tags, containers and the URLs of connected services.
Security
Sign-in through Microsoft Entra ID or Okta, owners and members, API keys with scopes, domain restrictions, an audit trail. The container never runs as root.
What enterprises build first
Eligibility, life events, plan comparison, registration, confirmation.
Intake, documents, status, handoff to an adjuster.
Checks against your systems before anything is promised.
Tickets, policies, access requests, with handoff.
Bring one use case to a technical session
We build, test and deploy it with your team, in your environment.
Questions people ask
Does AgentKube see our conversations?
No. The runtime runs inside your network. Conversations, memory, knowledge and data stores stay there. AgentKube receives version check-ins and usage counts only.
Which clouds are supported?
Google Cloud Run, Azure App Service and AWS today, plus your own servers with Docker or Podman. Kubernetes and OpenShift are on the roadmap.
Can we use our own models?
Yes. Every provider is bring-your-own-key: OpenAI, Anthropic, Gemini, Mistral, Meta, OpenRouter, Ollama, and models in your own AWS account through Bedrock.
How is access controlled?
Sign-in through your identity provider, owners and members per organization, API keys with scopes, domain restrictions for the widget, and an audit trail of who did what.