Control what the assistant can see

Conversation Context is where you decide what the assistant has in front of it. It is per-conversation, so one thread can be tightly grounded while another stays open-ended.

Open it from the conversation’s options — Manage the system prompt, knowledge bases, and documents for this conversation.

System prompt

Free text: Custom instructions that guide the AI behavior for this conversation. This is the most direct lever you have. Use it for standing instructions — the audience, format, things to avoid, and level of detail.

A persona sets this too. A conversation-level system prompt is the right place for instructions specific to this piece of work; a persona is the right place for anything you would otherwise retype. See Personas.

Knowledge collections

Knowledge bases linked to this conversation for RAG context. Link one and the assistant retrieves relevant passages from it and cites them, rather than being handed the whole corpus.

Manage Knowledge Bases links and unlinks them. With none linked you will see No knowledge collections linked.

Documents

Individual documents attached to this conversation. Use this when the material is a handful of specific files rather than a body of reference material — the distinction between this and a knowledge base is scale and reuse, not kind.

Memory

Include saved memories controls whether what AODex has remembered about you is carried into this conversation. Turning it off for one thread does not delete anything; it just keeps that thread clean.

See Memories for what is remembered and how to remove it.

Saving

Save Changes applies them — AODex confirms with Context settings saved. There is a reset that returns the conversation to defaults, confirmed with Context reset to defaults.

Context is not retroactive. Changing it affects what happens next in the thread. Replies that are already there were produced under the old settings.