Vector Stores Overview

Vector Stores in SeekrFlow are high-performance semantic indexes built from your documents. They allow language models to retrieve meaningful, contextually relevant information — powering workflows like agents, assistants, and context-grounded fine-tuning.

This page covers everything you need to know to create, configure, and manage Vector Stores.

🧠 What Is a Vector Store?

A Vector Store (also called a vector database or VectorDB) stores vector representations of your content — meaning each chunk of your documents is converted into an embedding, making it searchable by semantic similarity rather than just keywords.

Use a Vector Store when you want to:

• Enable semantic retrieval for Agents or Assistants
• Generate Context Grounded datasets for fine-tuning
• Create long-term memory systems from documentation or internal knowledge

🛠 How to Create a Vector Store

1. Navigate to Data Engine → Storage → Vector Stores
2. Click Create Vector Store

1. In the modal, enter:

   • Vector Store Name (required)
   • Description (optional)
   • Embedding Model – select from available options

1. Click Create Vector Store

You’ll return to the Vector Stores tab where your new store appears in the list with:

• Name
• File count
• Total size
• Date modified

📋 Vector Store Details View

Click on any vector store to open the details view, which displays:

• Vector Store ID
• Selected embedding model
• Created at
• Description
• Section to Add Files to the store
• List of any existing embedded files

➕ Adding Files to a Vector Store

To attach documents to your store:

1. Click Add Files in the detail view

2. A modal will open, giving you two options:

   • Select from existing files in File Storage
   • Upload new files directly into this vector store

You can drag and drop or browse to upload.

Upload Limits:

• Up to 20 files at a time

• Maximum 150MB per file

• Supported file types:

  • .pdf
  • .docx
  • .ppt
  • .md
  • .json

❌

Unsupported files (like .jsonl, .parquet, or .txt) cannot be added to a vector store.

🔧 Chunking Settings

After selecting your files, configure how your content will be segmented:

• Intelligent Chunking (default):
  Automatically detects logical content breaks (e.g., headings, paragraphs)

• Manual override options:

  • Chunk Size (token count)
  • Chunk Overlap (token count)

This ensures precise control over how your data is broken down before embedding.

🚀 File Processing Flow

Once you click Add Files, SeekrFlow will handle everything automatically:

1. Ingesting — Converts compatible files into Markdown if needed
2. Chunking — Breaks content into manageable segments
3. Embedding — Transforms each chunk into a high-dimensional vector
4. Indexing — Stores vectors into the semantic index

You’ll return to the Vector Stores tab where file processing begins in real-time.

📂 File View in Vector Store

When processing completes, embedded files appear in the detail view with:

• File name
• Status (Queued, Ingesting, Chunking, Embedding, Complete)
• Size

If a file fails to process, you’ll receive an error with guidance for resolution.

🔄 Managing Your Vector Store

Inside the detail view, you can:

• Add or remove files
• Monitor file status
• Rebuild the store if content or structure changes
• Track where the vector store is being used (coming soon)

🔗 How Vector Stores Power SeekrFlow

Vector Stores connect directly into major workflows:

• Agents & Assistants — power retrieval-augmented memory
• Context Grounded AI-Ready Data Jobs — generate training data from retrieved chunks
• Fine-Tuning — use structured content to improve model alignment

Each store is fully reusable and optimized for high-performance semantic access.