Give company knowledge
a useful answer.
Build a company knowledge assistant with AgenticOS. Use RAG for document retrieval and a Project for maintained wiki notes, sources and shared instructions.
Build a document assistant
01 / Start with
A small handbook or research collection, a question it answers and a question it cannot answer.
02 / Build toward
A document assistant you can test, plus a Project where sources, linked notes and an index can be maintained across chats.
03 / Check the result
Inspect the source behind an answer. Ask about an absent policy, then change a source and check that obsolete material is no longer used.
From example to your data
Build a version
you can inspect.
AgenticOS runs on your infrastructure. Set up the model, tools and access for this task, then use the linked documentation for the exact configuration.
- 01
Choose retrieval, a wiki, or both
RAG finds material for a question; a maintained wiki organizes what the team has learned. In AgenticOS, knowledge collections provide indexed retrieval. Projects keep files and instructions across chats. One does not automatically create the other.
- 02
Start with one checkable document
Configure a compatible embedding model, create a collection and upload a small source. Wait for processing, then bind only that collection to the test agent. An available chat model alone is not an ingestion setup.
- 03
Give continuing research a home
For a wiki, agree a Project structure: sources/, inbox/, notes/, index.md and a change log. Ask Architect to plan it and inspect the proposal before approving. Creating a Project does not write all those files or schedule updates.
- 04
Test the answer and the missing answer
Ask who handles a request, what information is required and what the document leaves unspecified. Inspect retrieval and citations. Repeat after a source change; do not assume that uploading the same filename replaces old vectors.
Questions worth
settling first.
Is an LLM wiki the same as RAG?
No. A wiki is a maintained set of linked notes and source records. RAG retrieves relevant material at question time. You can use a Project to maintain the wiki and a knowledge collection to add retrieval.
Does self-hosting keep every document local?
Only when the full processing path is configured that way. External embedding models, parsers, chat models and tools may receive data. Choose providers and deployment settings for the actual document policy.
Does every chat know the other chats?
Project files and instructions are shared context. Other chat transcripts are not automatically supplied as model context, even when members can open them in the interface.