# Document Text Embedding

Extract text records from a document visual manifest, generate vectors, and write them to a text index table.

Use it alongside image indexing in multimodal flows. For ordinary document arrays going directly to a knowledge base, use [Embed and Add to Knowledge Base](../ai-processing/knowledge-index.md).

## Bind the manifest and text index

Check template-provided bindings, or supply these inputs when building a flow yourself.

| Input | Description |
| --- | --- |
| Visual manifest | Upstream manifest object or manifest file reference |
| `text_vector_table` | Required target table for text vectors |
| `embedding_model` | Required text embedding model |
| `enabled` | When false, skips writes and returns `disabled` |

Multi-level indexing is off by default. Enable it if needed.

Bind a manifest containing both text and source information.

## Use the output

`documents` contains text records built from the manifest, and `documents_count` counts them. `written` counts vector rows actually written. Check `text_vector_table` and `embedding_model` against the intended configuration.

Use `status` to distinguish writes from a disabled node.

## Example: add text retrieval

Flow: Build Document Visual Manifest → Document Text Embedding.

Bind the manifest and specify the text table and embedding model. Connect Image Embedding separately for visual retrieval, then check the write results of both indexes.

## Related pages

- [Variables and data flow](../../variables-and-data.md)
- [Run and debug](../../run-debug.md)