Govern agent work
Authorize agent workflows before they act, with permissions set by your organization.
AI workflow platform with governance built in
EKKA runs AI workflows using your business context, with access and actions governed by your organization. Choose models and usage limits for each task, and keep a verifiable record of execution.
Start free on Developer: every capability, 1,000 runs a month, no card.
EKKA integrates with NVIDIA OpenShell to govern agent workflows and desk administration. Manage what agents can do, control the desks they work in, and review signed records of governed operations.
Authorize agent workflows before they act, with permissions set by your organization.
Create, start, stop and delete agent desks, and change their access policies through governed workflows.
Review signed execution records to trace governed operations and their reported outcomes.
OpenShell enforces sandbox boundaries. EKKA governs agent workflows and desk administration.
See how EKKA works with NVIDIA OpenShellWatch EKKA check authority before an AI workflow acts.
For teams building AI workflows that need business context, defined permissions, and human decisions.
Provide task inputs from approved business sources.
Choose an approved model for each task.
Use a specific prompt version for each task.
Set response-token limits on supported models.
Assign approval decisions to the appropriate roles.
Review signed records and reported model usage.
Your organization defines the information a workflow uses and the actions it may perform.
EKKA works with what you already use, including NVIDIA OpenShell.
EKKA Kits guide your team through a working AI workflow.
Find a workflow to explore.
Set up EKKA and connect your services.
Inspect permissions and records.
Install the CLI and sign in. Your organization starts on Developer and can run work right away.
Need more people, more Enclaves or an agreement? Tell us about your team and we’ll talk about EKKA Enterprise.
Developer is free: every capability, 1,000 runs a month, no card.
Models run on your own model key.
See pricing