Similar to how you can use variables in your LLM system prompt, you can also customize your input for the Vector Database, Tools, and Agent.

Let's use vector database as an example. Assuming we're setting up a simple RAG (vector database + LLM), we'll simply refer the "payload" value, and use it for retrieval:

You can also get creative and set up a template as well:

Of course if you have a couple of actions happening before the vector database, you can also refer those outputs as well:

All of the above logics apply to "Tools" and "Agent" actions as well.