> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kelasa.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Approvals

> Keep humans in the loop on agent actions

Approvals are how Kelasa keeps people in control of automation. There are two
kinds:

## Confirmations — before the assistant acts

When you ask the [assistant](/concepts/assistant) for something with side
effects, it shows a confirmation of exactly what it's about to do — the action,
the target, the parameters. Approve and it runs; decline and nothing happens.
In Slack this is a message card with buttons; on the web it's inline in the
chat.

This is per-action and immediate: it protects against the assistant doing
something you didn't mean, every single time.

## Approvals — gates inside workflows and skills

Workflow steps and skill runs can require sign-off from someone with the right
role before proceeding. The run pauses at the gate; approvers see the pending
request under **Approvals** in the web app (and as a Slack card, if installed),
review the context, and approve or deny. The run then continues or stops, and
the decision is recorded in the run history.

Use approval gates for the steps that matter: spending money, touching
production, contacting customers, granting access.

## Who can approve

Approvals respect workspace roles — approval requests go to people authorized
for that gate, not to whoever happens to be online. Workspace
[policies](/concepts/governance) can also force approval for specific tool
actions across the board, independent of how a workflow was authored.
