Generate Text Output

Output of the Agent to the user is primarily created by the Core Loop and LLM Request blocks. In both of these blocks, the LLM ultimately generates text that can be sent to the user.

However, you may want more control on what exactly the Agent responds to the user. For that, you can use the Generate Text Output block.

The Generate Text Output block contains a CEL or Jinja template, that is expanded and evaluated using the current state and scratchpad.

This can be used, for example:

• At the end of the turn to provide an answer with predictable formatting

• Before a long operation, to indicate to the user that this will take some time.

The output of this block can also be saved to the state, scratchpad, or conversation history, as well as being streamed to the user.