The Knowledge base is a store of documents agents draw facts from when they
answer (RAG, retrieval-augmented generation). You upload company material,
AiHummer extracts the text, splits it into chunks and indexes it; the agent then
finds the relevant chunks and answers from them.
What you can upload
Three ways to add content:
+ File — pick one or more files. Supported: PDF, Word (.docx), Excel
(.xlsx), CSV, .txt, .md, .rtf; the cap is 20 MiB per file. The server extracts
text and reports the number of chunks produced per file.
+ Text — paste text directly: title, source and body.
+ URL — point at a link; the page content is fetched.
The list
Each document shows its title/source, chunk count and time added. There is search
and sorting by name or date, and a total chunk counter in the header.
How to use it
Upload documents (clean, structured text works best).
Check that the chunk count is non-zero — that means the text was extracted.
Make sure the agent is allowed to use knowledge (the capabilities section)
and that the chosen embedder fits your
language.
Configuration
Extraction, embedding and chunk-size settings belong to the RAG engine — see
Knowledge & RAG. The default embedding model is
multilingual; for large volumes it can be moved to a GPU.
Tips
Split large documents by meaning — retrieval is more precise that way.
Give a meaningful “source”: it helps both people and the agent cite where a fact
came from.
The **Knowledge base** is a store of documents agents draw facts from when they
answer (RAG, retrieval-augmented generation). You upload company material,
AiHummer extracts the text, splits it into chunks and indexes it; the agent then
finds the relevant chunks and answers from them.
## What you can upload
Three ways to add content:
- **+ File** — pick one or more files. Supported: **PDF, Word (.docx), Excel
(.xlsx), CSV, .txt, .md, .rtf**; the cap is 20 MiB per file. The server extracts
text and reports the number of chunks produced per file.
- **+ Text** — paste text directly: title, source and body.
- **+ URL** — point at a link; the page content is fetched.
## The list
Each document shows its title/source, chunk count and time added. There is search
and sorting by name or date, and a total chunk counter in the header.
## How to use it
1. Upload documents (clean, structured text works best).
2. Check that the chunk count is non-zero — that means the text was extracted.
3. Make sure the agent is allowed to use knowledge (the `capabilities` section)
and that the chosen [embedder](/en/v1.0/concepts/knowledge-rag) fits your
language.
## Configuration
Extraction, embedding and chunk-size settings belong to the RAG engine — see
[Knowledge & RAG](/en/v1.0/concepts/knowledge-rag). The default embedding model is
multilingual; for large volumes it can be moved to a GPU.
## Tips
- Split large documents by meaning — retrieval is more precise that way.
- Give a meaningful "source": it helps both people and the agent cite where a fact
came from.
## Next
- [Knowledge & RAG](/en/v1.0/concepts/knowledge-rag) — how semantic search works.
- [Einstein memory](/en/v1.0/webui/memory) — how facts differ from documents.
- [Agents](/en/v1.0/webui/agents) — where to enable knowledge use.
