Text Embedder
Text Embedder calculates the numeric fingerprint of the meaning of a specific text. That is what an embedding is: a list of numbers that sums up what a text says, so that two texts with a similar meaning have similar fingerprints, even when they use different words.
Vector databases calculate these fingerprints on their own when they store and search documents. Use Text Embedder when you want the fingerprint of a single text to work with it in the flow, for example to compare two texts with Embedding Similarity.
When to use it
- Compare a customer’s question with a reference sentence to find out whether they are about the same thing.
- Detect whether a new incident is similar to one that is already logged.
- Test which embeddings model best tells your own texts apart before choosing one for your knowledge base.
Configuration
| Connection | What to connect |
|---|---|
| Embedding Model | The embeddings component that calculates the fingerprint, for example OpenAI Embeddings or FastEmbed Embeddings (Local). Connect its Embeddings output here. |
| Field | What it does |
|---|---|
| Message | The text whose fingerprint is calculated. You can type it or receive it from another component, for example from a chat input. |
Output
- Embedding Data: a data item with two parts: the original text (
text) and its fingerprint, the list of numbers (embeddings). Connect it to the Embedding Vectors input of Embedding Similarity to compare it with another text, or to any component that works with data.
Example
The customer service team of Electrodomésticos Olmo wants to know whether a message is about a return.
- Add an embeddings component, for example FastEmbed Embeddings (Local).
- Add two Text Embedder components and connect the same embeddings component to Embedding Model in both.
- In the first one, connect the customer’s message to Message. In the second one, type a reference sentence in Message, such as “I want to return a product and get my money back”.
- Connect both Embedding Data outputs to Embedding Vectors of Embedding Similarity, which tells you how similar they are.
Good practices
- Use the same embeddings component for all the texts you are going to compare: fingerprints from different models cannot be compared with each other.