Gavrun is the agent action control plane — a decision engine that sits between your AI agents and the systems they touch. Every tool call gets a verdict before it executes: allow, block, or hold for a human.
AI agents are moving from demos to production. They book meetings, process refunds, send emails, and touch databases. Most of the time that's fine. Occasionally it's catastrophic. Gavrun is the layer that makes the difference — catching the bad calls before they land, routing the risky ones to a person, and keeping a record of everything.
Developers install Langman (pip install langman) inside their agent applications. Langman intercepts each tool call and asks Gavrun for a preflight decision. Gavrun matches the call against your policies — tool allow-lists, approval gates, token ceilings, PII rules — and returns a verdict in milliseconds. The agent never changes; the call never executes without a decision on record.
For teams that need to run governance on their own infrastructure, Zumie Gateway brings the full Gavrun stack into your AWS environment. Same decisions, same audit trail — behind your own VPC. Available on AWS Marketplace.
Gavrun is currently in limited preview. Sign in to access an existing workspace, or deploy Zumie Gateway on AWS to run it on your own infra.