Embeddings
The Embeddings component — turn text into vectors with an embedding model, chunking documents first, for vector pipelines you build yourself.
The Embeddings component turns text into vectors — lists of numbers that capture meaning, so that similar passages sit close together. It's the building block for vector pipelines you assemble yourself, such as storing vectors in your own Qdrant collection.
Searching documents? Use a dataset
Datasets chunk, embed, store, and search your documents for you. Reach for Embeddings when you need vectors directly.
Inputs
- Model — a provider connection whose model has the Embedding capability (OpenAI and LM Studio offer embedding models). Required.
- input — what to embed. It can be:
- a document's text — split into chunks first, then each chunk embedded;
- a list of strings — each embedded as-is;
- a list of objects — each object keeps all its fields and gains a
vector, embedding the field named in Text field.
Output
Always a list. For text, each chunk comes back as {index, text, heading, vector};
for lists, one entry per item, in the same order.
Settings
| Setting | Default | What it does |
|---|---|---|
| Text field | text | For a list of objects, which field to embed. |
| Chunk size | 1000 | The longest a chunk may be, in characters. |
| Overlap | 200 | Characters repeated across each cut when a single paragraph is longer than a chunk. |
| Split on headings | On | Start a new chunk at every Markdown heading. |
| Batch size | 100 | How many texts go to the model per request. |
| Parallel batches | 5 | How many requests run at once. |
The chunking settings apply to a document's text; lists are embedded item by item. Chunking works the same way as in dataset pipelines.