Semantic Search

​

Semantic Search node

Semantic search allows you to search over uploaded files or over the datasets you create.

​

Details

Embeddings for datapoints are done using Cohere’s embed-multilingual-v3.0 model.

​

Configuration

​

Threshold

Threshold represents minimal semantic similarity between the query and the search result for it to be selected. Valid range: 0.0 - 1.0. For example, if the threshold is set at 0.42, only the contents with similarity above or equal to 0.42 to the search query will be selected

​

Limit

Limit is the maximum number of contents with similarity over threshold to be returned. Valid range: non-negative integers.

​

Template

You can control the output through a template where keys are placed into {{mustache_syntax}}. This template is applied to every content retrieved by semantic search.

All retrieved contents, rendered according to template, are joined via a newline character \n.

Read more about variable requirements in: String template node and JSON extractor node

​

Reserved keys

• relevance_index - An index, which starts from 1, representing the order in the list of retrieved contents ordered by similarity to the query. 1 means this content is the most semantically similar.
• Additionally, you can select any key name from your dataset "data" as a key.

​

Datasources

You can click “Add datasource” button and add as many datasources as you want. It means the semantic search will search for the most semantically similar texts among all datasources.

Datasources of semantic search nodes are simply datasets. You can go to “datasets” page, create a dataset, add datapoints, index them, and then use it here (Read more). If datapoints are not indexed, then there will be no entries in our vector database to search through.

​

Output

The output of a semantic search node is a single string of all rendered (with each key replaced by its value) templates joined via a newline character \n.

​

Examples

​

Example template for searching over files

​

Template

[{{relevance_index}}]

{{content}}

---

​

Rendered output

[1]

This portion of the file is the most relevant

---

[2]

This portion of the file is a little bit less relevant

---

​

Example template for searching over datasets.

This will search over a dataset that has keys short_description and price for a list of products.

​

Template

{{relevance_index}}. {{short_description}}
Price: {{price}}

​

Output

1. This is the best frying pan you can find in household shops.
Price: $25.99

2. The frying pan everyone must get!
Price: $14.49