Attachment nodes
Attachments change what an AI node knows. Flux ships three: GPT Memory, Gemini Memory and Knowledge Base.
Memory
Without a memory attachment, every message is the model's first. It cannot resolve "and what about tomorrow?", because it has never seen what came before.
Attaching memory carries the conversation through the session, so the agent follows references back to earlier turns.
A GPT Model node with a GPT Memory attachment and a Knowledge Base attachment connected below it
Screenshot to be addedThe cost of memory
Conversation history is sent to the model on every call, so a long conversation costs more per message than a short one. That is normal and worth knowing when you read token usage — cost per session grows with conversation length, not with how many nodes you drew.
Knowledge Base
Attach documents and the agent answers from your material rather than from what the model happens to know. This is what makes a support bot useful: it knows your product, your policies and your pricing.
Documents are uploaded as knowledge files on the attachment node, and the AI node consults them when answering.
Knowledge Base or QnA tool?
| Knowledge Base | QnA tool | |
|---|---|---|
| Shape | Documents | Defined question and answer pairs |
| Attached to | The attachment connector | The tool connector |
| Answers | Phrased by the model from your material | Close to what you wrote |
| Best for | Broad coverage — manuals, policies, guides | The handful that must be exact |
They combine well: a knowledge base for the long tail, plus a QnA tool for the answers that carry commercial or legal weight.
Grounding the agent
Attaching a knowledge base does not stop a model inventing things. Say so in the system prompt as well:
Answer only from the attached knowledge base. If it does not contain the answer, say you do not know and call the transfer tool.