Knowledge · 1 min read
Knowledge sources
Sources an assistant searches on demand rather than keeping loaded, and how embedding models constrain them.
A knowledge source is material an assistant searches when a question calls for it, rather than material it holds open at all times. That makes sources the right home for a body of reference material too large to attach.
What a knowledge source is
Section titled “What a knowledge source is”Rather than being loaded into every conversation, a knowledge source is searched on demand. The assistant looks into it only when a question seems to call for what it contains, then draws on whatever it finds.
Attachment or source?
Section titled “Attachment or source?”An attachment is small enough, and central enough, that the assistant should always have it loaded. A source is large, or used only sometimes, and searching it on demand keeps the assistant fast without giving up access to it. See Attachments for the other half of that choice.
Adding a source to an assistant
Section titled “Adding a source to an assistant”Attach an existing knowledge source to an assistant from the Knowledge step of the creation wizard, or later from the assistant’s settings. See Knowledge.
Embedding models and compatibility
Section titled “Embedding models and compatibility”Every knowledge source is indexed using an embedding model, and all the sources attached to one assistant need to share the same embedding model to be searched together. If a source was indexed with a different embedding model than the assistant’s other sources, it can’t be selected alongside them.
Why a source may be unavailable
Section titled “Why a source may be unavailable”A source can be missing from the list for a couple of reasons: its embedding model doesn’t match the assistant’s other sources, or the Space doesn’t have any compatible knowledge sources configured yet.