Skip to content
Eidos Docs

Use Semantic Search

Semantic search allows you to find content based on meaning rather than exact keyword matches. By converting text into vector embeddings, you can search with natural language queries like “machine learning tutorials” and find relevant items even if they don’t contain those exact words.

This guide shows you how to enable semantic search on a text field and use it to search through data. We’ll use Hacker News stories as a practical example.

Prerequisites

Section titled “Prerequisites”
  1. Eidos Desktop: The desktop app is required for embedding generation.
  2. AI Provider: Configure an embedding model in Settings → AI.

Setup

Section titled “Setup”

1. Enable Embedding on a Text Field

Section titled “1. Enable Embedding on a Text Field”

Semantic search works on any Text field. Here’s how to enable it:

  1. Open your table
  2. Click the Text field header → Edit Property
  3. Expand AI Enhancement section
  4. Toggle Enable Embedding
  5. (Optional) Enable Color Hint to see vectorization status

2. Process Embeddings

Section titled “2. Process Embeddings”

Once data is added, Eidos automatically generates embeddings for the enabled field. You can check progress in field properties:

  • Green cells = Up-to-date vectors
  • Yellow cells = Outdated (text was edited)

Click Process to manually trigger embedding generation if needed.

Section titled “Using Semantic Search”

In the Table View

Section titled “In the Table View”
  1. Open your table
  2. Click the search/filter icon
  3. Select Semantic Search
  4. Enter your query in natural language

For example, if you’re searching through Hacker News stories:

QueryFinds stories about…
”AI and machine learning”Neural networks, LLMs, deep learning
”web development tools”Frameworks, build tools, CSS libraries
”database optimization”Query performance, indexing, caching
”career advice”Hiring, interviews, work culture
”debugging techniques”Troubleshooting, logging, observability

Troubleshooting

Section titled “Troubleshooting”

”No embedding model configured”

Section titled “”No embedding model configured””

Go to Settings → AI and add an embedding provider. You can use cloud-based providers (OpenAI, etc.) or local models that run offline.

Slow search performance

Section titled “Slow search performance”
  • Check if all vectors are generated (green cells)
  • Reduce limit parameter (e.g., 10 instead of 100)
  • Ensure your device has sufficient RAM for the embedding model

Model mismatch

Section titled “Model mismatch”

If you switch embedding models, existing vectors become invalid:

  1. Open the field properties
  2. Click Reset Vectors to clear old embeddings
  3. Click Process to regenerate with the new model

Next Steps

Section titled “Next Steps”