> 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.md).

# Knowledge Upload

A web form for uploading any document — embedded and upserted into the right namespace in one pass, with clean-slate utility for full replacements.

> **Drop a file in, the agent knows it within seconds — no database access, no API keys, no developer.**

The agent is only as good as the knowledge it can retrieve. The Knowledge Upload workflow is how that knowledge gets in and how it stays accurate. A simple web form accepts any document format, embeds it, and upserts it into the correct namespace in a single pass — so a freshly uploaded FAQ entry or product spec is live in the agent within seconds.

## What You Get

* Upload form accepts CSV, TXT, PDF, DOC, DOCX, MD — every document format the brand actually has.
* Document type dropdown routes content to the correct namespace automatically — FAQ content goes to the FAQ retrieval tool, instructional content goes to the walkthrough tool, brand data stays in the brand catalogue.
* One-pass embedding and upsert — submit the form and the agent knows it within seconds.
* Full metadata stored on every vector — original text, file name, upload timestamp, document type — so retrieval results carry traceability back to source.
* Clean-slate utility for full namespace replacement when content needs to be rebuilt rather than incrementally updated.

## How It Works

The form takes 3 fields: the document file, a document type dropdown, and an optional notes field. On submit, the pipeline detects file type by MIME, routes PDFs through PDF text extraction and other formats through structured extraction, normalises the result into a single text field per chunk, and sends the text through the embeddings model.

The returned vectors are merged with the document metadata — text, document type, file name, source, timestamp — and upserted into the destination namespace derived from the document type dropdown. The whole process runs in a single workflow execution per upload — no manual vector generation, no separate upload step.

For full replacements, a dedicated utility wipes a namespace in one call so the next upload starts on a clean slate.

## Quality Baked In

* File type detection by MIME, not extension — extension renames don't break the routing.
* CSV Q\&A format support — each row's question and answer are concatenated into a single embeddable chunk for FAQ-style content.
* Deterministic vector IDs — built from document type, timestamp, and chunk index, so re-uploads don't produce silent ID collisions.
* Full metadata on every vector — every retrieval result traces back to the source document and upload date.
* Namespace isolation — uploaded FAQ content only ever returns from the FAQ tool; uploaded brand data only ever returns from the brands tool.

## Inputs

* Document file — CSV, TXT, PDF, DOC, DOCX, or MD.
* Document type — selected from a dropdown that maps to a retrieval namespace.
* Optional notes — preserved alongside the upload but not retrieved.

## Outputs

* Vectors upserted into the correct retrieval namespace.
* Metadata persisted with each vector — text, type, file name, timestamp.
* Confirmation that the upload succeeded.

## What This Replaces

* The developer ticket to "add this FAQ to the bot" and the 2-week wait that follows.
* The manual embedding script that runs locally on someone's laptop.
* The stale knowledge base that nobody owns and nobody updates.
* The deferred-forever backlog that quietly degrades retrieval quality month after month.

One form, any document, live knowledge — without touching a database or a deploy pipeline.

***

*This was built. No agent ships this out of the box.*