Embedding (EMBEDDING)

Turns a text into a list of numbers (a vector) that describes its meaning, so texts can be compared by meaning. It is a step only. It cannot be used as an Agent’s tool.

Category: Models & Agents · Deployed: yes · Tool-capable: no

Ports

Port Kind Purpose

Text (in_text)

Input

The text to embed. It is the whole output of the previous module

Vector (out_vector)

Output

The vector, as a JSON list of numbers, for example [0.012, -0.034, …]

Properties

Click Select AI Server & Model at the top of the panel to fill the first four fields from a registered AI server.

Field Required Meaning

Model Provider (model.provider)

Yes

OLLAMA, OPEN_AI or GEMINI. There is no ANTHROPIC, because Anthropic offers no embedding models. A server of type Anthropic in the picker does not work here

Model Name (model.name)

Yes

An embedding model, for example nomic-embed-text or text-embedding-3-small. Vectors from different models cannot be compared, so keep one model for everything you want to compare

Model URL (model.url)

Yes

The address of the model server, for example localhost:11434. With OPEN_AI, /v1 is added when the address does not end with it. localhost and 127.0.0.1 are replaced by the Docker host, so the agent can reach a server on your computer

API Key (model.apiKey)

Needed for OPEN_AI and GEMINI

Leave it empty for OLLAMA. The field is hidden like a password

Model Timeout (min) (model.timeoutMinutes)

No

How long to wait for the model, 1–120 minutes. Default: 20. Higher → a slow or busy server can finish. Lower → a stuck call fails sooner

Model Provider, Model Name and Model URL must be filled in, even for GEMINI. If one is empty, the pipeline cannot be assigned to an agent, and the designer lists the missing field. If a saved pipeline is deployed anyway, the module is left out.

The key from the picker is a copy. The picker writes the server’s current key into the module. If you change the key on the AI server later, the module keeps the old one. Pick the server and model again to copy the new key.

What happens at run time

  • The module takes the whole output of the previous module as one text and sends it to the model.

  • Embed one text piece at a time. A long text, or a list of many pieces, is not split up here. Use a Text Splitter and an Iterator in front, so each piece is embedded on its own.

  • If the previous module sends nothing or only spaces, the step fails with the message that there is nothing to embed.

  • If the model cannot be reached or answers with an error, the step fails and shows the model’s message.

  • The vector has a fixed length that depends on the model.

Use cases

  • Custom ingestion chain: File Read → Text Splitter → Iterator → Embedding → KV Store Write. This is the usual use. The vector goes to the next module as text.

  • Compare meaning: embed two texts and compare the vectors in a Python Script.

Notes

  • Use an embedding model, not a chat model. A chat model does not give usable vectors.

  • The module only creates the vector. It does not store it or search with it. To keep vectors, pass them on, for example to KV Store Write.

  • For ordinary questions on your documents, a RAG Topic already does all of this. Use this module only when you need your own vector workflow.