Copilot
The Copilot is the most useful day-to-day tool for working with your AI flows. Instead of editing forms by hand, you talk to an assistant that can read every configured flow, explain how things work, make changes for you, and troubleshoot when something misbehaves.
It's under active development, so expect occasional rough edges — but it's the primary way the team builds and maintains everything from flows and data points to integrations, tickets, WhatsApp templates, and FAQs.
Features
The Copilot's capabilities group into a handful of areas. Each has its own page with sample prompts you can adapt.
| Area | What it handles |
|---|---|
| Understand & plan | Explain configuration, review a flow, or plan an approach — read-only. |
| Build & change flows | Flows, objectives, branching, data points, and child-ticket templates. |
| Connect external systems | Integrations, custom APIs, source APIs, and sample data. |
| Tickets & disposition | The disposition form, on-closure actions, and per-flow presets. |
| WhatsApp & outreach | WhatsApp Flows, broadcast templates, and template reply hooks. |
| Knowledge & replies | FAQs and agent-facing canned responses. |
| Testing | Draft test cases and review why a test passed or failed. |
| Debugging | Why a flow, transition, custom API, or ticket misbehaved. |
Business Opening Message
Ask Copilot to set the business opening sentence and configure its quick replies, including buttons that start specific tasks:
"Set our opening sentence to 'Welcome to Acme! How can we help?' and add a 'Track order' button that opens the Order tracking task."
You can also ask to view the current greeting and button mappings, rename or remove a button, change its task, or clear the greeting. Changing only the greeting leaves the buttons intact. Opening quick replies support up to 10 buttons with labels of up to 20 characters. Task mappings use saved, enabled tasks in your account; Copilot asks for clarification if a requested task is missing or ambiguous.
To create a task and a welcome button together, include both in your request. Copilot saves the task before configuring the button that opens it.
How a Request Is Handled
Working with the Copilot follows the same loop no matter what you're changing:
- You describe what you want in plain language — a question, a change, or a whole new flow.
- The Copilot reads your live configuration — every flow, data point, custom API, template, and FAQ in the account — so its answer or change is grounded in what you actually have, not a generic template.
- It plans the work. For anything non-trivial it breaks the request into ordered steps and streams its progress as it goes, so you can see what it's doing and why.
- It asks before guessing. If something is ambiguous or a required detail is missing, it stops and asks a short question (usually with suggested options) rather than inventing an answer.
- You review the result. Read-only requests just answer. Changes are made against your configuration for you to review — then confirm they behave with a Test flow run or an automated test.
The clearer your request, the fewer round-trips: name the customer intent, the specific flow or data point, and the outcome you want.
When a Request Spans Several Areas
You can ask for several things at once and the Copilot will break the work into ordered steps, respecting dependencies (for example, creating a data point before the flow that uses it).
"Add a data point for tracking and update the refund flow to use it." "Create a refund flow and add its disposition presets."
If you paste mixed content — say, some fixed policy facts plus flow-style instructions — it splits that into FAQs and flows so nothing gets dropped.
When It Asks You to Clarify
- If a request is too vague to act on confidently, the Copilot asks a short clarifying question with suggested options rather than guessing.
- If a request is outside what it can configure (e.g. product UI changes, analytics/report generation, billing, or editing source code), it tells you and points to the closest supported alternative.
Long Conversations
The Copilot keeps the messages of a chat in its context so it can refer back to earlier work. When a thread grows large, it automatically summarizes the older messages and keeps only the recent ones verbatim. A divider reading Earlier messages summarized marks where this happened in the transcript.
Nothing is deleted. The earlier messages stay visible, and the summary preserves the flows, data points, APIs, decisions, and open questions they contained, so you can keep saying "edit that flow" without repeating yourself. If you want a completely fresh context, clear the chat or start a new one; clearing a chat also discards its summary.
Tips
- Use it to learn an unfamiliar account before changing anything — ask how existing journeys work.
- Be specific about the customer, the trigger, and the desired outcome.
- Name exactly what you mean when it matters ("edit the data point, not the API") to avoid ambiguity.
- After the Copilot makes a change, validate it with the Test flow feature or an automated test run.
Related
- Workflows — the building blocks the Copilot edits.
- Flows — the full flow configuration reference.
- Data Points — built-in and custom-API-backed data points and execute actions.
- Custom APIs — endpoints and source APIs.
- Integrations — providers, the API library, and resources.
- Ticket Fields & Disposition — the disposition form the Copilot configures.
- Broadcasts — WhatsApp templates and reply hooks.
- Testing — confirm changes behave as expected.