Skip to content

Creating a Knowledge Base

This guide walks you through creating a new Knowledge Base. After creation, you will add documents to it either by uploading files directly or connecting an external document source.

Prerequisites

Before you begin, ensure the following:

  • You have the Manage Knowledge Bases permission in your role.
  • At least one AI integration with an embedding model is configured in Settings > Integrations. Common options include OpenAI (text-embedding-3-small), Vertex AI (text-embedding-005), or a self-hosted model.

Step-by-Step

1. Open the Knowledge Bases Section

Navigate to Building > Knowledge Bases in the Operations Portal sidebar. You will see a list of existing Knowledge Bases (or an empty state if this is your first).

2. Click New Knowledge Base

Click the New Knowledge Base button in the top-right corner. A creation form will open.

imageNew Knowledge Base creation form showing name, description, embedding model, and chunking settings fields
The Knowledge Base creation form

3. Enter Basic Details

FieldRequiredDescription
NameYesA descriptive name (e.g., "Product Documentation", "HR Policies"). Must be unique within your tenant.
DescriptionNoA brief summary of what this KB contains. Shown in the list view and helps teammates understand its purpose.

4. Select an Embedding Model

Choose the AI integration and model that will be used to generate vector embeddings for your documents. This is a critical choice -- once documents are indexed with a specific model, switching models requires a full re-index.

OptionDescription
IntegrationThe AI provider integration configured in Settings (e.g., "OpenAI Production", "Vertex AI")
ModelThe specific embedding model from that provider

WARNING

All documents in a Knowledge Base must use the same embedding model. If you change the model after indexing, you will need to re-index every document. Choose your model carefully before uploading large document sets.

imageEmbedding model selector dropdown showing AI integration picker and model options including text-embedding-3-small, text-embedding-3-large, and text-embedding-005
Selecting an embedding model

5. Configure Chunking Settings

Chunking controls how your documents are split into smaller text segments before embedding. These settings directly affect retrieval quality.

SettingDefaultDescription
Chunk Size512 tokensMaximum number of tokens per chunk. Larger chunks preserve more context but reduce precision.
Chunk Overlap50 tokensNumber of tokens shared between adjacent chunks. Overlap helps ensure relevant passages are not split across chunk boundaries.

TIP

For general-purpose Q&A, the defaults (512 size / 50 overlap) work well. For long-form documents like legal contracts or research papers, consider increasing chunk size to 1024 with 100 overlap to preserve more surrounding context.

6. Save

Click Create Knowledge Base. The KB is created immediately with an empty document set. You will be redirected to the KB detail view where you can start adding documents.

What Happens Next

A newly created Knowledge Base contains no documents and cannot be used in flows until content is indexed. Your next steps are:

  1. Upload documents -- Add files directly from your computer. See Managing Documents.
  2. Connect external sources -- Sync files from SharePoint, Google Drive, S3, or other storage. See External Document Sources.
  3. Verify indexing -- Monitor document processing status to confirm chunks are generated.
  4. Add a KB Search node -- Use the Knowledge Base in a flow by adding a KB Search node and selecting this KB.

Naming Conventions

If you plan to use dynamic KB lookup via the sys_page_label variable, use clear, predictable names. The lookup performs a case-insensitive match, so Product FAQ will match a page label of product faq.

Good:   "Product FAQ", "Billing Support", "Onboarding Guide"
Avoid:  "KB1", "test_kb", "docs (copy)"

Limits

LimitValue
Knowledge Bases per tenant50
Documents per Knowledge Base500
Maximum file size per document50 MB
Supported embedding dimensionsUp to 3072

TIP

These limits apply to the standard plan. Contact your administrator if you need higher limits for your tenant.

OmniBots AI Bot Platform