Introduction to the Knowledge Base

Learn how the Spojit Knowledge Base lets your workflows work with your business documents.

Overview

The Knowledge Base is a document storage and retrieval feature built into your Spojit workspace. You upload documents into named collections, Spojit processes them into vector embeddings, and your workflows can then search the contents in natural language instead of relying on rigid keyword matching.

It is the bridge between your unstructured business content (PDFs, spreadsheets, emails, contracts) and the structured logic of a workflow. A workflow can pull facts out of a policy document, look up a SKU in a product catalogue, or answer a customer question by citing the most relevant paragraph of your support docs, all without you writing custom search code.

How It Works

The Knowledge Base has three phases:

Upload - You add files to a collection through the Knowledge page in the sidebar, or programmatically from a workflow.
Embed - Spojit parses the document, splits it into chunks, and generates vector embeddings so it can be searched semantically.
Query - A Knowledge node in a workflow searches a chosen collection with a natural language query and returns the most relevant matches.

Concepts

Collection - A named group of related documents (for example product-specs or company-policies). Collections are workspace-scoped. A workflow can also use a Transient collection, which lives for one workflow run and is automatically cleaned up - ideal for processing a single document end-to-end without persisting it.
Document - A single uploaded file inside a collection.
Embedding - The vector representation Spojit creates for each chunk of a document so it can be matched against a query.
Query - A natural language question sent to a collection. The query returns ranked passages from the documents.

Common Use Cases

An internal FAQ that support workflows query before drafting a reply.
A product catalogue lookup that enriches incoming orders with specs.
Invoice and contract parsing where a workflow extracts named fields from uploaded files.
A policy lookup feeding an approval workflow with the relevant rules.

Tips

Start with one collection per topic rather than a single catch-all. Focused collections produce sharper search results.
Test queries from the Knowledge page before wiring them into a workflow. It is faster than running the workflow each time.
Treat the Knowledge Base as your AI-readable source of truth: when source documents change, replace them in the collection.

Common Pitfalls

Mixing unrelated content in one collection. Queries lose relevance when a single collection covers many topics.
Forgetting that embeddings represent the document as it was uploaded. Editing the original file outside Spojit does not update the index until you re-upload.
Expecting exact-match keyword search. The Knowledge Base is semantic, so phrase your queries as questions rather than search terms.
Ignoring storage limits. Large PDFs and image-heavy decks consume more of your plan allowance than plain text.

Learn More

Knowledge base documentation