Skip to content

AgentMaker

AgentMaker is the copilot that appears in the agent editor as a left side panel next to the prompt. It understands the current prompt, the connected tools, and the organization context, and helps you structure, improve, or expand the prompt through natural language conversation.

AgentMaker open in the prompt editor

Use AgentMaker when you want to:

  • turn a vague idea (“I want a sales agent for my CRM”) into a structured prompt;
  • improve an existing prompt — clarity, tone, rules, structure;
  • add behavior rules through conversation;
  • discover which tools would make sense for the agent;
  • ask for help solving a bad behavior you observed in tests.

It talks in plain, simple language — you don’t need to be technical or know anything about “prompt engineering”. And when the agent is still starting from scratch (no tools and no knowledge base), AgentMaker keeps it basic, focusing on the agent’s role, mission, and tone of voice, without overwhelming you with options.

In the Prompt tab of the agent editor, click Edit Prompt in the top right. This single button does two things at once: it opens the prompt text editor (manual editing) and the AgentMaker panel beside it. AgentMaker lives inside “Edit Prompt” — it’s the assisted way to edit.

To go back, click Preview: the prompt returns to read mode and the AgentMaker panel closes. If you’d rather close just the panel and keep editing the text by hand, use the X at the top of the panel — that closes AgentMaker without leaving edit mode.

The History button next to it is separate and opens the previous prompt versions.

  1. You describe what you want in the panel’s text field.
  2. AgentMaker analyzes the current prompt and proposes changes — could be rewriting a section, adding a new one, removing something, or rewriting the whole prompt.
  3. Changes appear as staged edits — you review before applying.
  4. You apply all, edit manually, or discard.
  5. AgentMaker can also suggest enabling or disabling tools, and recommend configuration changes by pointing you to where to make them.

Your conversation with AgentMaker stays saved while you work on that agent: if you reload the page or come back later, the history is still there — you don’t lose the context of what you were building. It’s one conversation per agent, in this browser.

To start over, use the Reset chat button (restart icon) at the top of the panel. It asks for confirmation before clearing — starting over does not change the agent’s saved prompt, it only clears the conversation.

At the top of the panel there’s a help button (question-mark icon, “What AgentMaker does”). It opens a quick summary of AgentMaker’s current capabilities and a link to this documentation.

  • Be specific in your request. “Improve the prompt” produces generic changes; “make the tone more formal and add a rule about not offering discounts” produces focused changes.
  • Apply one improvement at a time. Easier to test and revert.
  • Always test the agent after. AgentMaker is an assistant, not the owner of the result — responsibility for the final behavior is yours.
  • Combine with the direct editor. AgentMaker is great for structure and proposing big changes; fine-tuning is usually faster in the text editor.