> For the complete documentation index, see [llms.txt](https://oodlz-ai-studio.gitbook.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://oodlz-ai-studio.gitbook.io/docs/inbound-agent/the-knowledge/knowledge-upload/embed-and-upsert.md).

# Embed And Upsert

One automated pass from form submission to live vectors — embeddings generated, metadata preserved, deterministic IDs prevent collisions.

> **Submit the form, the knowledge is live in seconds — no manual vector generation, no separate upload step.**

Every uploaded document is embedded and upserted into the destination namespace in a single automated pass. The pipeline runs end-to-end on form submission. No manual vector generation. No separate upload step. The agent reads the new content on the next retrieval call.

## How It Works

After extraction and normalisation, the text is sent to the embeddings model and a vector is returned for each chunk. The pipeline merges the returned vector with the document metadata into a single payload — original text, question and answer fields if present, document type, source file name, upload timestamp — and upserts the payload into the destination retrieval namespace.

Vector IDs are generated deterministically from document type, timestamp, and chunk index. The combination guarantees uniqueness across uploads — re-uploads of the same content with the same parameters produce the same ID, which means upserts overwrite cleanly instead of producing silent duplicates.

## What You Get

* One pipeline run from form submission to live vectors — no intermediate steps.
* Full metadata persisted with every vector — every retrieval result carries traceability back to the source document and upload date.
* Deterministic vector IDs — re-uploads overwrite, never duplicate.
* Live retrieval within seconds of upload — the agent picks up new content on the next call.