Skip to main content
The editor is where you review and finish building the agent you generated. You arrive with the instructions written and the tools chosen, but there are decisions only you can make: which agent to hand off to, which variable to use, which document to send. Agent editor with the instructions on the left and the tools in the middle The editor has three areas:
  • Instructions, on the left: the agent’s complete prompt.
  • Tools and Knowledge base, in the middle: what the agent can do and what it knows.
  • Playground, on the right: the chat to test it without leaving the screen.

​
Adjust the instructions

The instructions are editable text. You can rewrite whatever you want, but it’s worth respecting the structure Treble generated: identity and voice, goal and limits, steps to follow, edge cases and exit rules. Inside the text you have two shortcuts:
  • Type / to open the quick commands menu and insert a tool at the exact point where you want it used.
  • Type {{ to insert a saved variable and use its value inside the conversation.
When the agent doesn’t behave as you expect, the fix is almost always to make the instructions more explicit: add the case that failed as a concrete example instead of rewriting the whole prompt.

​
Complete the pending tools

Tools appear as pills inside the instructions. The ones that need data from you are shown in yellow, and above the editor you’ll see a notice like “There are 4 tools pending configuration”. Notice of tools pending configuration in the editor Use the arrows in the notice to go through them one by one: each click takes you to the next pending pill and opens its configuration. For example, Transfer to an agent needs you to pick the destination tag, and Save response needs the name of the variable to store the data in.
You can’t publish the agent while tools are still pending. If you try to save too early, Treble takes you to the ones that are missing.
The available tools are grouped by category —General, Treble, HubSpot, Advanced— and you can add the ones the generator didn’t include. You’ll find the detail of each one in the AI Agent introduction.

​
Review the knowledge base

The Knowledge base tab holds the files you attached during generation and the agent’s context text. You can add more documents, replace the ones that went out of date, or write information directly. Before adding material, review the best practices for building a knowledge base: the quality of these documents defines the quality of the answers.

​
Define what happens if the contact doesn’t reply

Above the instructions you’ll find the question “What should the AI Agent do if the contact does not respond?”. There you define a timeout and the action that runs when it’s reached:
  • Close conversation: ends the interaction and marks it as resolved.
  • Send conversation: sends the contact to another Treble conversation.
  • Transfer to an agent: hands the chat off to a human agent.
Always configure this. Without this rule, a conversation the contact abandons stays open indefinitely.

​
Save your agent

When no tools are left pending, save the agent with the button in the top right corner. From there you can create a test conversation and send it to yourself on WhatsApp.
If you close the editor with unsaved changes, Treble keeps them. The next time you come in you’ll see the notice “You have unsaved changes from a previous session” with the option to recover them.

​
Next steps

Test your agent

Validate the behavior in the playground and send yourself a test conversation on WhatsApp.

AI Agent

See the detail of every tool and how to write instructions from scratch.