Skip to main content

What citations do

When an agent answers a question using a knowledge base or a web search, MagOneAI can attach citations to the reply — small chips under the message that tell the user exactly where the information came from. Citations turn an opaque answer into a verifiable one: the user can click through to the source document or web page and confirm it for themselves. Citations are a presentation layer. They never change the agent’s reasoning or the text it generates — they’re collected from the tools the agent actually used and shown alongside the answer.

What a citation chip contains

Each chip represents one source the agent drew from. Depending on the source, a chip shows:
  • Source name — the document filename or the page title
  • Location — the page number for a document (e.g. policy.pdf (p.3)), when available
  • A link — for SharePoint documents and web results, the chip is clickable and opens the original in a new tab
  • A source icon — distinguishing a SharePoint/document source from a web result

Knowledge base

A document the agent retrieved from an attached knowledge base. Manually-uploaded files show as a non-clickable chip; SharePoint-synced files link back to SharePoint.

Web result

A page returned by a web search or fetched by the agent. Always links to the URL.

SharePoint

A synced SharePoint document, linking directly to the file in SharePoint with author and last-modified detail.

Where source types come from

Every citation is tagged with a source type so the chip can render correctly:
Source typeOriginChip behavior
manualA document uploaded by hand to a knowledge baseNon-clickable (no source URL)
sharepointA document mirrored from SharePoint (see Knowledge bases)Clickable — links to the file in SharePoint
webA web search result or fetched pageClickable — links to the URL
The source type and attribution metadata are recorded when documents are indexed, so toggling citations on or off never requires re-processing your documents.

Enabling citations per use case

Citations are controlled by a single per–use case toggle. In the use case’s edit settings, the “Source citations” switch turns chips on or off for every reply that use case produces.
  • The toggle is on by default — new use cases show citations automatically.
  • It takes effect immediately on the next reply; you don’t need to republish the workflow.
  • Turning it off suppresses chips entirely, even though the underlying agent still retrieves and uses the same sources.
Leave citations on for anything user-facing — support assistants, policy Q&A, research helpers — where trust and verifiability matter. Turn them off for internal automation steps where the chips add noise rather than value.

Knowledge base citations

When an agent retrieves from a knowledge base, each retrieved document can become a citation. The chip carries the knowledge base name, filename, section, and page, plus — for SharePoint documents — the source URL, the last person to modify the file, and when. Documents you uploaded manually appear as a plain, non-clickable chip because there’s no external link to point to. This works in both auto and agentic retrieval modes, and across multiple attached knowledge bases.

Web search citations

When an agent uses a web search tool, the results it pulls in are cited automatically. Each web result becomes a clickable chip labeled with the page title (falling back to the domain or URL), so users can open the exact page the agent referenced. Tools that don’t return web-style results — a database query, for example — don’t produce citations, so chips stay meaningful.

How citations reach the user

Citations are gathered from every tool the agent called during a reply, de-duplicated (by URL, or by filename and page), and attached to the assistant’s message. In MagOneAI Hub they render as chips beneath the answer. The full structured output of the run remains available in the execution history — citation chips are an additional, lightweight enrichment on top of the conversation.

Next steps

Knowledge bases and RAG

How documents are retrieved and how SharePoint sync attributes sources

Web search

Give agents the ability to search the web

Triggers & execution

Inspect full execution history and tool calls

Agent node

Configure agent retrieval and tool behavior