Knowledge
Knowledge is what your agents and assistants answer from: collections of documents, the vector databases that hold them, and the pipelines that load them. Builders create and fill it; agents and assistants search it when they answer.
Who can open it
Section titled “Who can open it”| Role | What they can do |
|---|---|
| Admin | Everything on this page and its three views: create, load, search, change and delete. |
| Builder | The same as an Admin. |
| Auditor | Open every view and read everything in it. Cannot create, change or delete, and cannot ask or search a collection. |
| Reviewer | Cannot open Knowledge. |
| End user | Cannot open Knowledge. |
| Platform admin | Everything an Admin can, for any tenant. |
Asking and searching a collection spend the tenant’s tokens, which is why they belong to the roles that can try things (Admin and Builder) and not to the Auditor.
What you see
Section titled “What you see”Build › Knowledge is one row in the rail with three tabs. Each tab keeps its own address, so a link or a bookmark to one of them still opens it.
| Tab | What it lists | Read more |
|---|---|---|
| Collections | Every collection in the tenant, across all of its vector databases. Opens first. | Collections |
| Vector databases | The tenant’s vector databases, their status, size and placement. | Vector databases |
| Pipelines | The ingestion pipelines that load documents into collections, their source, schedule and last run. | Pipelines and connectors |
How the three fit together
Section titled “How the three fit together”- A vector database is the store. It belongs to one tenant. Depending on how Olympusx AI was installed, each one is a database server of its own on a cluster you choose, or a database on the install’s shared vector store.
- A collection lives in one vector database. It is a searchable set of documents, embedded by one embedding model that you choose when you create it and cannot change later.
- A pipeline loads documents into one collection: files you upload, or a URL, an S3 bucket, an NFS share or an SFTP server, on demand or on a schedule. It parses each document, cuts it into chunks, embeds the chunks with the collection’s model and stores them.
- Each collection also gets a graph of its documents, their sections and the people, organizations and places they mention. See The collection graph.
- An agent or an assistant names the collections it answers from. See Create and edit an agent.
The usual order is: create a vector database once, create a collection in it, then load documents into the collection, either by dropping files on it or with a pipeline. The collection wizard can create the collection and its first pipeline in one pass.
Code collections
Section titled “Code collections”The Code Assistant keeps the code it indexes in collections too. Those collections appear in the Collections tab, but their pipelines are managed from the Code Assistant, and their graphs are on each repository’s page. See Code sources.