Knowledge Bases

Knowledge bases let you build searchable collections of documents that your AI workflow steps can reference. Upload files or add URLs, and QuickFlo automatically extracts the text, splits it into chunks, and creates vector embeddings — making the content available for semantic search during AI step execution.

How Knowledge Bases Work

1. Create a knowledge base and add documents (files or URLs)
2. Processing happens automatically — documents are parsed, chunked, and embedded
3. Search happens at execution time — when an AI step runs, the user’s prompt is used to find the most relevant chunks from the selected knowledge bases
4. Context injection — matched chunks are added to the AI model’s context, giving it access to your specific information

This pattern is called Retrieval Augmented Generation (RAG) — the AI generates responses grounded in your actual documents rather than relying solely on its training data.

Creating a Knowledge Base

1. Navigate to Knowledge Bases in the QuickFlo sidebar
2. Click New Knowledge Base
3. Enter a name and optional description

Field	Description
Name	A unique name within your organization (max 255 characters)
Description	Optional context about what this KB contains (max 2,000 characters)

Adding Documents

File Upload

Upload documents directly from your computer. Supported file types:

Format	Description
PDF	Scanned and text-based PDFs
DOCX	Microsoft Word documents
PPTX	PowerPoint presentations
HTML	Web pages saved as HTML
TXT	Plain text files
MD	Markdown files
CSV	Comma-separated value files

URL Import

Add documents by URL — QuickFlo fetches the content automatically:

1. In your knowledge base, click Add from URL
2. Enter one or more URLs
3. QuickFlo fetches each URL, extracts the text content, and processes it

Tip

URL-based documents can be resynced to pull in updated content. Click the resync button on any URL document to re-fetch and re-embed the latest version.

Document Processing

After adding a document, it goes through an automatic processing pipeline:

Status	Meaning
Pending	Queued for processing
Processing	Being parsed, chunked, and embedded
Ready	Successfully processed and searchable
Failed	Processing encountered an error (see error message for details)

Processing involves three stages:

1. Parse — Extract text from the document based on its format
2. Chunk — Split the text into semantic segments optimized for search
3. Embed — Generate vector embeddings for each chunk using OpenAI’s text-embedding-3-small model

Note

Processing happens asynchronously in the background. Large documents may take a few moments to complete. You can continue working while documents process.

Using Knowledge Bases in Workflows

Knowledge bases are used through the LLM Call step. When configuring an LLM call, you can select one or more knowledge bases to search.

Configuring an LLM Call with Knowledge Bases

Field	Description
Knowledge Bases	Select one or more knowledge bases from the dropdown
Prompt	Your prompt to the AI model — this is also used as the search query
Model	The AI model to use for generation

How Search Works at Execution Time

When the LLM Call step runs with knowledge bases selected:

1. The prompt text is used as a search query
2. QuickFlo finds the most relevant chunks across all selected knowledge bases using vector similarity search (cosine distance)
3. The top matching chunks (up to 5 by default) are prepended to the system prompt as context
4. The AI model generates its response with access to the retrieved context

Only chunks with a relevance score above the minimum threshold are included, ensuring the AI receives high-quality context.

Example

If you have a knowledge base called “Product Docs” containing your product documentation, an LLM Call step with:

Field	Value
Knowledge Bases	Product Docs
Prompt	{{ initial.question }}

The step will:

1. Search “Product Docs” for chunks relevant to the user’s question
2. Include the most relevant documentation excerpts as context
3. Generate an answer grounded in your actual product documentation

The LLM Call step output contains the AI’s response:

{{ answer-question.text }}        // the generated response
{{ answer-question.usage.total }} // total tokens used

Managing Knowledge Bases

Updating Documents

• File documents: Delete the old document and upload the updated version
• URL documents: Click Resync to re-fetch and re-process the URL content

Deleting Documents

Deleting a document removes it and all its chunks from search results. The change takes effect immediately — subsequent AI step executions will no longer find content from deleted documents.

Deleting a Knowledge Base

Deleting a knowledge base removes all its documents and chunks permanently. Any workflows referencing the deleted KB will no longer have that knowledge base context available.

Limits

Limit	Value
Knowledge base name	255 characters
Description	2,000 characters
Document name	255 characters (unique within KB)
Embedding model	text-embedding-3-small (1,536 dimensions)
Search results per query	Up to 5 chunks
Minimum relevance score	0.3 (cosine similarity)