Skip to main content
The Knowledge Base allows you to upload, organize, and search documents using AI-powered semantic search. Documents in your Knowledge Base can be accessed by Stella AI to provide accurate, context-aware responses.

Key Concepts

Datasets

Collections that organize related documents together. Each dataset can have its own configuration and access settings.

Documents

Individual files uploaded to datasets. Supported formats include PDF, Word, text, and markdown files.

Stella Search

Enable AI-powered search on datasets to make them available to Stella AI agents.

Connected Sources

Import documents automatically from external sources like Intercom Help Centers.

What You Can Do

ActionWho Can Do ItDescription
View datasetsAll usersBrowse datasets and their documents
Create datasetsAdminsCreate new document collections
Upload documentsAdminsAdd PDF, Word, text, or markdown files
Delete documentsAdminsRemove documents from datasets
Configure datasetsAdminsEnable Stella search, connect sources

Supported File Types

FormatExtensionsMax Size
PDF.pdf10 MB
Word.docx, .doc10 MB
Text.txt10 MB
Markdown.md10 MB

How AI Search Works

When you enable “Searchable in Stella” on a dataset:
  1. Document Processing: Each document is automatically chunked into semantic segments
  2. Embedding Generation: Segments are converted to vector embeddings using AI
  3. Indexing: Embeddings are stored for fast similarity search
  4. Search: Stella can find relevant content by meaning, not just keywords
Processing happens automatically in the background. Large documents may take a few minutes to become searchable.

Getting Started

1

Create a Dataset

Go to Knowledge Base and click Add Dataset. Give it a name and optional description.
2

Upload Documents

Open your dataset and click Upload Documents. Drag and drop files or browse to select.
3

Enable Stella Search

Edit your dataset settings and toggle Searchable in Stella to make documents available to AI.
4

Use with Stella

Stella AI will automatically search enabled datasets when answering questions or performing tasks.

Next Steps