# The AI Front Door

> Ask the Unifuse assistant about your students, documents, applications and campaigns; how it cites sources, abstains when unsure, and asks before it acts.

Source: https://docs.unifuse.ai/ai-assistant/front-door/

The AI Front Door is the assistant built into Unifuse. Open it from the chat bubble in the
header or the ask bar on the Orchestration Hub. It answers from your institution's knowledge
base and the Expert Nodes configured for your institution, and it says when it does not know.

## What it knows

The left panel, **What this front door knows**, lists the live knowledge-base sources and
the Expert Nodes that are online. The assistant only answers from those; it does not guess.
Student questions return only students in your scope.

Suggested starting points include "How many students are at risk this week?", "Show me
pending document reviews", "Where do this cycle's applications come from, by state?" and
"Summarize recent campaign performance". **Start a campaign** (for roles with *manage
campaigns*) runs a three-question intake that creates a real campaign draft and opens the
builder.

## Reading an answer

An answer can include the steps the assistant took, result cards (a student you can open, or a
chart with its source and record count), numbered citations that link to the source, a
confidence percentage, and an **Abstained** badge when the assistant found nothing it could
answer from. Anything that looks like code, configuration or credentials is replaced with a
refusal.

## Your conversation

Your thread is kept on the server per user, survives navigation and reloads, and is cleared
when you sign out. **New conversation** starts afresh.

If your institution has turned on the **Voice input** module, a microphone appears in the
input. The transcription is done by your browser vendor.

## When the assistant wants to act

Some requests ask the assistant to do something, such as send a nudge to a student. Those
actions are governed by trust tiers: the assistant shows what it wants to run and does not run
it until the right person says yes. See [Approvals and trust tiers](/ai-assistant/approvals/).

## Related

- [Knowledge base answers](/enrollment/knowledge-base/)
