Skip to main content
Agents are only useful in an enterprise if you can control what they do. Governance in Kelasa isn’t a report you read after the fact — it sits in the execution path: every side-effecting action passes safety and policy checks before it runs, and everything leaves an audit trail.

​
The layers

1

Emergency controls

Workspace-level kill switches checked before anything else. If something goes wrong, stop agent activity immediately — no deploy, no support ticket.
2

Policy

Workspace policies decide whether a given tool action is allowed, requires approval, or is denied — evaluated at a single enforcement point every action must pass, whether it came from the assistant, a skill, or a workflow.
3

Confirmation and approvals

Side-effecting assistant actions pause for explicit confirmation. Workflow steps and skill runs can require approval from someone with the right role — granted in the web app or from a Slack card. See Approvals.
4

Audit

Runs, deliveries, approvals, and policy decisions are recorded — who asked for what, what was approved, what ran, and what it did.

​
Roles and access

Workspace members are admins or members. Admins manage people, integrations, policies, and settings. By default, only admins can run side-effecting capabilities through the assistant; a workspace setting extends that to all members when you’re ready.

​
Data controls

Workspaces have data-retention controls, and secrets used by workflows are encrypted at rest and never shown back once saved.