> ## 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.

# Assistant

> A tool-calling AI assistant with human-in-the-loop confirmation

The Kelasa **assistant** is how your team talks to everything else on the
platform. Ask it a question or ask it to do something — it discovers the
capabilities available in your workspace (workflows, skills, knowledge search,
integration actions) and uses them, with you in the loop for anything that has
side effects.

## What it can do

* **Answer from your knowledge.** Questions are grounded in your workspace's
  [knowledge bases](/concepts/knowledge-bases) — documents, websites, and
  synced tool content — not just the model's general knowledge.
* **Run workflows and skills.** "Run the weekly report" or "onboard Priya" —
  the assistant finds the matching workflow or skill and runs it.
* **Act in your tools.** Through connected
  [integrations](/concepts/integrations), the assistant can look things up and
  take actions in the tools your workspace has connected.

## Human-in-the-loop by default

Reading is free; acting is confirmed. Whenever a requested action would have
side effects — sending a message, creating a ticket, changing data — the
assistant pauses and shows exactly what it's about to do. Nothing runs until
you confirm, and in Slack that confirmation is a tap on a message card.

Workspace admins control who can run side-effecting actions at all: by
default only admins can, and a workspace setting extends that to all members.
Every action also passes the workspace's safety and policy checks — see
[Governance](/concepts/governance).

## Where it lives

The assistant is available as **Chat** in the web app and as a DM in
[Slack](/guides/slack) — same capabilities, same guardrails, same history.
Conversations are streamed live, and every session is kept in your workspace
history.