✅ What Are Text Embeddings (Simple Explanation)

Text embeddings turn chunks of your document into a list of numbers (a “vector”) that represents meaning.
When ZotAI needs to answer a question about a document, embeddings help it quickly find the most relevant passages inside that same document — even if your wording differs.

Example: If your question says “brain development” and the document uses “neural maturation,” embeddings help ZotAI pick the right parts of the doc anyway.

🔍 Why Embeddings Matter in ZotAI (Single-Document Workflow)

• Handle bigger docs: Models have a context (token) limit. With embeddings, ZotAI selects only the most relevant parts of your current document, so you can work with files that would otherwise be too large.
• Lower costs: Instead of sending the whole document to the AI model, ZotAI sends smaller, relevant chunks — fewer tokens = lower cost.
• Better answers from the same doc: Because the passages are chosen by meaning (not just keywords), answers stay focused and on-topic.

⚙️ How To Enable Text Embeddings in ZotAI

1. Open ZotAI
2. Go to Settings → AI Embeddings
3. Toggle Enable Embeddings to ON
4. Choose a provider (see below)

Once enabled, ZotAI will create embeddings for the document you’re analysing, so it can pull the best passages when answering.

🌐 Choosing an Embeddings Provider

ZotAI supports several different embeddings providers. You can choose the one that best fits your balance between cost, speed, and privacy.

• OpenAI — Fast, accurate, and reliable. Uses models like text-embedding-3-small and text-embedding-3-large. Great default choice for most users.
• Google Gemini — Uses Google’s Gemini API. Works well for modern language data, with similar pricing to OpenAI.
• Voyage AI — Specialised for deep, academic, or research-style text understanding. Slightly higher cost but excellent precision.
• Together AI — Combines multiple open-source models. Affordable and flexible for users who like experimenting.
• Local Models (Ollama / LM Studio) — Run directly on your device. 100% private and free after setup, though speed depends on your computer’s hardware.

You can switch providers anytime in Settings → AI Embeddings. ZotAI will use your chosen source for all new embedding generation.

💬 Embeddings vs No Embeddings (In ZotAI)

• With Embeddings: ZotAI indexes your current document into meaning-based chunks, then sends only the most relevant snippets to the chat model.
  This fits within context limits and reduces cost.
• Without Embeddings: ZotAI must send larger raw text blocks (or full sections) to the chat model.
  This can hit token limits sooner and cost more.

💰 Cost & Performance

• Lower token usage: Sending targeted snippets is cheaper than sending long sections.
• Faster responses: Smaller, focused prompts generally run faster.
• Provider billing: If you use a cloud provider (e.g., OpenAI, Gemini, Voyage, or Together), small per-document fees apply.
  You pay your provider directly — ZotAI never processes payments.
• Local option: Using Ollama or LM Studio is completely free after setup, but may use more of your computer’s resources.

🔒 Privacy

Embeddings are generated only for the document you’re analysing and are used solely to identify relevant passages.
When using cloud providers, only small text chunks are temporarily sent for processing.
With local providers, nothing leaves your computer.
Your data is never shared with other users or stored externally by ZotAI.

✅ Should You Enable Embeddings?

• Yes, if you work with long PDFs or want cheaper, more focused answers from a single document.
• Maybe not needed if your documents are short and straightforward.

📩 Need Help?

Questions about embeddings in ZotAI? Email us at
[email protected]