- Documentation
- AI support
- Configuring
Knowledge
The sources the agent reads, and how to keep them current.
Knowledge belongs to the server and attaches to agents. Several agents can share the same source.
Source types
| Type | Detail |
|---|---|
| File | PDF, Word, text, Markdown, CSV |
| Web page | A public URL, re-read regularly |
| Discord channel | For instance #faq or #rules, resynchronised automatically |
| Note | A text you write directly in the dashboard |
Editing a source
The note is the only source whose text lives at Bao: it is therefore the only one you rewrite from the dashboard. The others are corrected at the source, then reindexed.
A reindex button forces a source to be re-read at any time. Live sources (web pages, channels) are resynchronised automatically.
How your knowledge is used
Bao decides on its own, based on the size of your base:
- Small base: it is handed to the agent in full. That is more accurate than a search, since no passage can be missed.
- Large base: Bao performs a semantic search and only injects the passages relevant to the question asked.
There is nothing for you to tune: you drop in your documents, that is all.
What makes a good base
- Answers, not documents. A 40-page marketing PDF answers worse than 15 short notes each addressing a real question.
- One fact, one place. Two documents that contradict each other produce an unstable answer.
- Up to date. An out-of-date base is worse than an empty one: the agent confidently states something false.
Mind the cost: a knowledge base is not storage paid for once. As long as it fits in the agent's memory, it is re-read on every reply. Growing your base raises the price of every future reply. The dashboard shows that cost per reply.