SigFollow
Sign in

System guide

Knowledge bases

Upload your own documents so an AI Agent answers from your organization's knowledge — create a knowledge base, add documents, and attach it to an AI Agent from AI Capabilities → Knowledge Bases.

A knowledge base is an organization-owned library of documents that AI agents can consult — product manuals, service terms, internal handbooks. Build one by uploading files. Instead of relying only on what the model already knows, an AI Agent retrieves the most relevant content from your dedicated knowledge base and grounds its reply in it. This page covers how to create a knowledge base, add and manage documents, read each document's processing status, and attach a knowledge base to an AI Agent.

The knowledge base list

Each row is one knowledge base. Columns:

ColumnWhat it shows
NameThe knowledge base name
DescriptionA short note, or a dash if left blank
DocumentsHow many documents the knowledge base holds
ChunksTotal number of indexed pieces across all ready documents
StatusActive — eligible for retrieval. Disabled — skipped during retrieval even if an AI Agent is attached.

Use the Actions menu on each row to Edit or Delete the knowledge base. Click a row to open its detail page and manage documents.

Creating a knowledge base

Click New Knowledge Base and fill in the form. The chunking fields have sensible defaults, so for most cases you only need a name.

FieldRequiredNotes
NameYesUp to 100 characters
DescriptionNoUp to 500 characters; describes what the knowledge base contains
Chunk SizeYesCharacters per chunk, 100–4000, default 800 (the FAQ/support sweet spot)
Chunk OverlapYes0–500 and at most half the chunk size, default 100; overlap keeps key sentences from being split across chunks
ActiveYesOn by default; an inactive knowledge base is skipped during retrieval

Click Create to save. The knowledge base appears in the list right away, empty. Open the detail page to start adding documents.

Defaults work for most cases
Chunk size 800 with overlap 100 fits typical FAQ and support content. Only change these if you have a specific reason, and remember that changing them later means re-indexing your existing documents.

Adding documents

Open a knowledge base, then click Add Document. There are two ways to add content — switch between them with the Paste text and Upload file tabs at the top of the dialog.

Paste text

  • Title(required) — a name for the document, for example "Refund Policy".
  • Content (required) — paste the document text into the box. A live character count helps you gauge the size.

Upload file

  • Title (optional) — leave it blank to use the file name.
  • File (required) — drag a file in or click to select one. Accepted formats are .md, .markdown, and .txt, up to 20 MB.

Click Add Document to submit. The document is queued for indexing and appears in the list as Pending; it becomes Ready once indexing finishes.

Managing documents

The detail page header shows the knowledge base name, description, the chunking parameters, and the total chunk count. The document table lists every document with these columns:

  • Title — the document name.
  • Sourcefile for an uploaded file or manual for pasted text.
  • Status — the processing state (see the table below).
  • Chunks — how many indexed pieces the document produced; shown only once it is Ready.
  • Indexed At — when indexing finished; shown only once it is Ready.

Document processing status

StatusMeaning
PendingSubmitted and waiting to be indexed
IndexingCurrently being processed
ReadyFinished and available for retrieval
FailedProcessing failed; click the status to see the error and retry

Per-document actions

Use the actions menu on each document row:

  • Re-index — re-processes the document. Its status returns to Pendingand it is indexed again with the knowledge base's current chunking parameters.
  • Delete — removes the document and every chunk that came from it. This cannot be undone.

Re-indexing everything

The Re-index Allbutton at the top of the detail page re-processes every document that is not currently indexing, using the knowledge base's current chunking parameters. Existing chunks are cleared and the documents return to Pending while they index again in the background; any document that is already indexing is skipped. This is the button to use after you change the chunk size or overlap.

Editing a knowledge base

From the list, choose Edit on a row to update the name, description, chunking parameters, and the Active toggle. The embedding model is fixed when the knowledge base is created and cannot be changed afterward.

Changing chunking does not re-index automatically
If you change the chunk size or overlap, existing documents keep their old chunks until you re-index them. The edit dialog reminds you of this. After saving, open the detail page and use Re-index All (or Re-index on each document) to apply the new settings.

Deleting a knowledge base

Choose Deletefrom a row's actions menu. Deleting permanently removes the knowledge base along with all of its documents and chunks — this cannot be undone. Before you confirm, the dialog checks which AI Agents are attached to this knowledge base and shows one of two results:

  • Safe to delete — no AI Agent is currently attached, so you can delete right away.
  • Agents attached — the dialog lists the AI Agents that use this knowledge base and warns that they will lose retrieval from it on the next customer message. You must tick the acknowledgement checkbox before the Delete button becomes available.

Attaching a knowledge base to an AI Agent

A knowledge base does nothing on its own — you attach it to an AI Agent so the agent can retrieve from it. Open the AI Agent's Edit dialog and find the Knowledge Bases (RAG) section. There are two settings:

  • Authorized Knowledge Bases — tick the active knowledge bases this AI Agent may use. Each option shows its chunk count. If your organization has no active knowledge bases yet, create one first from the Knowledge Bases page.
  • Auto-inject Top-K — how many snippets to retrieve for every customer message before the AI replies (0–5). Setting it to 0 is tool-only mode: the AI decides for itself when to search the knowledge base. A higher value improves recall but uses more tokens.
Pick at least one knowledge base when retrieval is on
If you authorize the AI Agent's knowledge-base search tool, or set Auto-inject Top-K above 0, you must also select at least one knowledge base — otherwise the agent has nothing to search. The edit dialog flags this and blocks saving until it is resolved.

Save the AI Agent. From the next customer message onward, it can ground its answers in the documents you provided.