Copilot

Watch Nathaniel create and revise a report in the video above. The recording shows an earlier interface.

Copilot lets your agent create a document, interactive tool, diagram, or code snippet in a dedicated viewer within the conversation. Users can ask for changes, inspect the result, and download it for use elsewhere.

A useful place to start is a document your users can refine through conversation: a client proposal, a weekly action plan, or a report. The agent drafts it, the user asks for changes, and the finished document can be downloaded. Copilot also supports interactive tools, diagrams, and code when the task calls for them.

Illustrative example using Pickaxe’s document viewer with sample content: ask for a change in chat, review the updated proposal, and choose a download format.

Try the same workflow: ask your agent to create a one-page client proposal with goals, scope, timeline, and budget. Then ask, “Keep the scope, but shorten the timeline to four weeks and update the milestones.” Review the revised document and download it as DOCX or PDF.

Here, editing means asking for revisions in chat. To make manual edits in a word processor, download the DOCX file.

Add Copilot to your agent

1. Open your agent in the Agent Builder and select Editor.
2. Go to Configure → Capabilities. Next to Add copilot to help your users create content, tools, or snippets, click Copilot.
3. This opens the Actions library filtered to Copilot. Select the Copilot action and connect it to your agent. You can also find it directly through the library's Copilot filter.
4. Set the action's Trigger prompt to explain when it should create or revise content. Add matching instructions to your agent's System prompt.
5. Save your agent and test it in Preview. Request a complete deliverable, then ask for a revision to confirm the workflow.

The Capabilities link opens the library; clicking it alone does not connect the action. If the shortcut is missing, check that your selected model supports actions.

Tell your agent when to use it

Copilot works best for substantial, self-contained work that a user wants to reuse or refine. Short answers and ordinary conversation can stay in the chat.

For a coaching agent, try this system prompt instruction:

When the user asks for a weekly plan, use Copilot to create a structured document with their goal, daily tasks, and a progress checklist. Ask for any missing details first. When they request changes, update the existing document. Keep brief coaching questions in the chat.

Then use a trigger prompt such as:

Use Copilot when the user asks to create or revise their weekly action plan.

For an interactive tool, describe the inputs and behavior you want:

Create an interactive pricing calculator in Copilot. Let the user enter their hourly rate and estimated hours, then show the total. Include a reset button and explain the calculation.

What users can create

The available download options depend on the output type. PDF export uses your browser's print dialog: choose Save as PDF to create the file.

HTML and React outputs can run in the preview. A general code snippet is displayed as source code, so asking for a Python script does not run that script in the viewer. Creating a page preview also does not publish a website or add a page to your portal. See Pages for portal publishing.

Refine and use the result

Open or expand the generated content to inspect it. Where supported, switch between the rendered result and its source. Use the copy control to copy the content or the download menu to export it.

To make changes, continue the conversation:

Keep the same plan, but move the longer tasks to the weekend.

Add a discount field to the calculator and show the savings separately.

Turn the process into a diagram and include the approval step.

Copilot can update or rewrite generated content as the conversation continues. Describe what should change and what should stay the same.

Signed-in users may also see Save to Knowledge Base in the download menu. This saves the output as a user document; it does not automatically add it to the shared workspace knowledge base.

Troubleshooting

The agent replies in chat without opening Copilot. Check that the Copilot action is connected. Make the system and trigger prompts explicit, and test with a request for a complete document or tool rather than a short answer.

The output stops before it is complete. Check the agent's output token allocation in Configure. Long documents and interactive tools need more output space. Try a smaller first version, then request additions.

An interactive preview does not work. Ask the agent to simplify the output and fix the specific error or behavior. Generated HTML and React run within Pickaxe's preview environment; not every external library or standalone application will work there. Test the controls before sharing the result.

Copilot, AI Helper, and Wingman

Related guides

• Agent Builder: configure and test your agent.
• Actions: connect tools and write trigger prompts.
• Prompting Best Practices: give your agent clear instructions.
• Knowledge Base: provide source material for the content your agent creates.