Skip to main content

Vector Stores - Your Own Documents as Knowledge Source

Vector Stores enable your agent to use your own documents as a knowledge source – PDFs, text files, FAQs and more.

What is a Vector Store?​

A Vector Store is a knowledge database that:

  • Stores and makes documents searchable
  • Converts text into mathematical "vectors"
  • Enables semantic search (by meaning, not just keywords)
  • Is managed by OpenAI

Simply explained: Imagine you upload your product catalogs, FAQs or manuals. The agent can then "look up" information in these documents when needed.

Use Cases​

📚 Product Documentation​

Upload: Product manuals, specifications, catalogs
Agent can: Look up technical details, explain features

❓ FAQ Database​

Upload: FAQ document with questions & answers
Agent can: Give consistent answers to common questions

📋 Guidelines & Policies​

Upload: Return policy, shipping guidelines, terms
Agent can: Provide correct information about policies

🏢 Company Knowledge​

Upload: Training materials, internal docs
Agent can: Access employee knowledge

Create Vector Store​

Step 1: Open OpenAI Dashboard​

  1. Go to platform.openai.com
  2. Sign in
  3. Navigate to Storage → Vector Stores
  4. Or directly: platform.openai.com/storage/vector_stores

Step 2: Create New Vector Store​

  1. Click on "+ Create vector store"
  2. A dialog will open

Step 3: Assign Name​

Name: Enter a descriptive name

Examples:
- "Product Catalog Winter Collection 2025"
- "Customer Service FAQ"
- "Technical Documentation"
- "Return and Shipping Policies"

Tip: Choose names that are easy to identify later.

Step 4: Upload Files​

Supported File Formats:​

  • PDF (.pdf) - Ideal for documents, catalogs
  • Text (.txt) - Simple text files
  • Markdown (.md) - Formatted texts
  • Word (.docx) - Microsoft Word documents
  • HTML (.html) - Website content

File Limits:​

  • Max. file size: 512 MB per file
  • Max. count: 10,000 files per Vector Store
  • Max. total size: Depends on your OpenAI plan

Upload Process:​

  1. Click "Upload files" or drag files via drag & drop
  2. Select one or more files
  3. Wait for upload to complete
  4. Status: "Uploading" → "Processing" → "Completed"
Multiple Files

You can upload multiple files simultaneously! Ideal if you have, for example, a complete product catalog with 50 PDFs.

Step 5: Wait for Processing​

OpenAI processes the files automatically:

  1. Chunking: Document is divided into small sections
  2. Embedding: Each section is converted into a vector
  3. Indexing: Vectors are made searchable

Duration:

  • Small file (10 pages): ~30 seconds
  • Large file (500 pages): ~5-10 minutes
  • Many files: Correspondingly longer

Check status:

  • Completed ✅ - Done, ready to use
  • Processing ⏳ - Being processed, wait
  • Failed ❌ - Error occurred

Step 6: Copy Vector Store ID​

After successful processing:

  1. Open the Vector Store
  2. Find the Vector Store ID
  3. Format: vs_abc123...
  4. Click "Copy ID"

In Shopware Backend:​

  1. Go to 5E OAI Agent Manager
  2. Open your agent (or create a new one)
  3. Find the "Vector Store ID" field
  4. Paste the copied ID: vs_abc123...
  5. Save
Linked!

Your agent can now access the documents in the Vector Store!

Mention Vector Store in Instructions​

Important: Tell your agent to use the documents!

Example: System Instructions​

You have access to our product catalog via a Vector Store.

IMPORTANT:
- Use the information from the Vector Store for product details
- If asked about specific product features, search the Vector Store FIRST
- Never give information that is NOT in the Vector Store
- If something is not in the documents, say honestly: "I don't have this information"

EXAMPLE:
Customer: "Does jacket X have a hood?"
You: [Search in Vector Store for "jacket X" and "hood"]
You: "Yes, jacket X has a removable hood with fur trim"

Example: FAQ Agent​

You have access to our FAQ database in the Vector Store.

PROCESS:
1. For every question: Search the Vector Store FIRST
2. Use the exact answer from the FAQ
3. If no matching FAQ found: Use other tools or refer to support

NEVER answer based on your general knowledge when it comes to shop-specific questions!

Best Practices for Documents​

1. Structure & Formatting​

Well structured:

# Product Name: Premium Winter Jacket

## Description
A high-quality winter jacket for extreme cold.

## Features
- Material: 100% Polyester
- Filling: 90% down, 10% feathers
- Water column: 10,000 mm
- Breathability: 8,000 g/m²/24h
- Hood: Yes, removable
- Pockets: 4 (2 outside, 2 inside)

## Sizes
S, M, L, XL, XXL

## Care
- Machine wash at 30°C
- Do not bleach
- Do not tumble dry

Poorly structured:

Premium Winter Jacket, polyester, down, waterproof, hood,
4 pockets, wash at 30 degrees, sizes S-XXL

2. Clear Headings​

✅ Use headings (H1, H2, H3) ✅ Structure logically ✅ Use lists ✅ Clearly separate topics

3. Avoid Redundancy​

❌ Not: Same information in 10 documents ✅ Better: One central document per topic

4. Ensure Currency​

  • Update documents regularly
  • Delete outdated information
  • Version documents (e.g., "FAQ_v2_2025.pdf")

5. Optimize File Size​

For PDFs:

  • Compress images
  • Remove unnecessary pages
  • Use text PDFs (not scanned text!)

Optimal:

  • 1-50 pages per document
  • 1-5 MB per file
  • Text searchable

Use Multiple Vector Stores​

Attention: Currently, an agent can have only ONE Vector Store.

Workaround for multiple knowledge sources:

  1. Everything in one Vector Store:

    • Upload all documents into ONE Vector Store
    • Use clear filenames for organization

  2. Multiple Agents:

    • Agent A: Vector Store "Products"
    • Agent B: Vector Store "Support"
    • Agent C: Vector Store "Policies"

Manage Vector Store​

Add Files​

  1. Open the Vector Store in OpenAI
  2. Click "Add files"
  3. Upload new files
  4. Wait for processing

Delete Files​

  1. Open the Vector Store
  2. Select the file
  3. Click "Delete"
  4. Confirm
Caution!

Deleted files cannot be recovered!

Delete Vector Store​

  1. Go to Vector Stores overview
  2. Select the Vector Store
  3. Click "Delete vector store"
  4. Confirm

Important: Remove the ID from your agent in Shopware first!

Costs​

Vector Store Costs​

OpenAI charges for Vector Stores:

Storage: ~$0.10 per GB per day

Example calculation:

10 PDF files at 5 MB = 50 MB = 0.05 GB
Cost per day: 0.05 GB × $0.10 = $0.005 (half a cent)
Cost per month: $0.005 × 30 = $0.15

Conclusion: Vector Stores are very affordable!

Usage Costs​

When the agent searches in the Vector Store, additional costs arise:

  • Embedding costs: ~$0.00002 per search
  • Token costs: Found texts count as input tokens

But: Good answers often save more than the search costs!

Troubleshooting​

Problem: "Vector store not found"​

Solution:

  1. Check the ID (format: vs_abc123...)
  2. Check if the Vector Store still exists in OpenAI
  3. Copy the ID again

Problem: Files are not being processed​

Causes:

  • File format not supported
  • File is corrupted
  • File too large (> 512 MB)
  • PDF is protected/encrypted

Solution:

  • Convert to supported format
  • Repair the file
  • Split large files
  • Remove password protection

Problem: Agent doesn't find information​

Solutions:

  1. Search is too specific:

    • Vector Store searches semantically, not keyword-based
    • Formulate questions more broadly

  2. Information not present:

    • Check if the info really is in the document
    • Use Ctrl+F in the original document

  3. Document poorly structured:

    • Improve the structure
    • Use headings
    • Upload again

  4. Mention in instructions:

    Use the Vector Store for product information.
    ALWAYS search the Vector Store first before answering.

Advanced Techniques​

Chunk Strategy​

Vector Stores divide documents into "chunks" (sections).

Optimal chunk size:

  • One chunk = 1 logical section
  • Use headings for separation
  • Avoid overly long paragraphs (> 500 words)

Example: Well structured

## Product Name

Description (100-200 words)

## Features

- Feature 1
- Feature 2

## Specifications

| Property | Value |
|----------|-------|
| Material | Polyester |

Use Metadata​

In filenames, you can encode metadata:

Product_Winter_Jacket_Premium_Category_Outdoor.pdf
FAQ_Shipping_USA_2025.pdf
Policy_Returns_EU.pdf

This helps with organizing large Vector Stores.

Alternatives to Vector Stores​

If Vector Stores don't fit your use case:

1. get_meta_information Tool​

For: Short shop info (contact, hours)
Advantage: Faster, simpler
Disadvantage: Limited to plugin config field

2. fetch_url Tool​

For: Dynamic content from your website
Advantage: Always current
Disadvantage: Slower responses, external dependency

3. search_logs Tool​

For: Self-learning FAQ database
Advantage: Grows automatically
Disadvantage: Needs time to build up

4. Hardcoded in Instructions​

For: Few, static information
Advantage: Very fast, no extra costs
Disadvantage: Instructions become long, hard to maintain

Summary: Vector Stores Checklist​

  • OpenAI Vector Store created
  • Documents uploaded (PDF, TXT, etc.)
  • Processing completed (Status: Completed)
  • Vector Store ID copied
  • ID entered in agent configuration
  • Mentioned in instructions ("Use Vector Store for...")
  • Tested in backend chat
  • Response quality checked
  • Documents optimized as needed

Next Steps​

➡️ Logs & Monitoring - Track usage

➡️ Knowledge Management - Self-learning agents

➡️ Best Practices - Optimization tips