AI Copilot
Build and modify workflows using natural language commands
The AI Copilot is your in-editor assistant for building and modifying workflows. It understands your canvas context -- the blocks, connections, and data flow -- and can add blocks, configure settings, wire data references, and suggest improvements, all from natural language instructions.
What the Copilot Can Do
Build workflows from descriptions. Tell the copilot what you want to automate and it will create the workflow structure, add the right blocks, and wire the connections.
Modify existing workflows. Ask it to add a condition, change a trigger, add error handling, or restructure the flow. It operates on your current canvas.
Answer questions. Ask about block configuration, data flow, best practices, or how to achieve a specific automation pattern. The copilot has access to the full documentation.
Debug issues. Describe a problem with your workflow and the copilot will analyze the configuration, identify likely causes, and suggest fixes.
Build dashboards. Ask it to add a chart, metric, or table to a dashboard, or describe a whole multi-widget board -- filters, KPIs with period-over-period deltas, a waterfall, a cohort heatmap, a budget-vs-actual table, a pivot with subtotals -- and it will assemble one against your connected tables. It reads your table schemas first, so charts reference real columns rather than guessed ones. Charts the copilot draws in chat use the same components, so Save to dashboard adds exactly what you saw.
Change dashboard settings by chat. Each dashboard has one currency, locale, negative-number style, fiscal year start, and density. Ask for it in plain words, for example "show this board in EUR with brackets for negatives" or "our fiscal year starts in April". There is no settings panel; the copilot or the widget JSON sets them. See Dashboard components for the full list.
Example Prompts
Building
Create a workflow that triggers when a Stripe invoice is paid,
looks up the customer in Salesforce, and sends a Slack notification
to #finance with the invoice details.Add an approval step before the QuickBooks invoice creation.
Route to the #finance-approvals Slack channel for amounts over $5000.Build an open-invoices dashboard from my invoices table, with a
date range filter and a breakdown by customer.Add an MRR overview to this dashboard: MRR, net new MRR and churn
with deltas vs last month, and an MRR bridge as a waterfall.Configuring
Change the agent block to use Claude Opus 4.6 instead of Sonnet.
Set temperature to 0 and add an output schema for invoice extraction.Add retry logic to the Stripe block -- 3 attempts with exponential backoff.Debugging
The workflow is failing at the transform step. The Salesforce response
has the customer email nested under Account.PersonEmail but the
transform is looking for contact.email. Fix the mapping.Querying
What's the best way to handle partial payments in a Stripe workflow?How do I pass data from the webhook trigger to a downstream agent block?How It Works
The copilot operates within the workflow editor. It has access to:
- Your current canvas -- All blocks, their configurations, connections, and data flow
- Block registry -- All 29 available block types and their configuration options
- Integration schemas -- Available actions and parameters for each integration
- Documentation -- Full Loopfour Studio documentation
When you ask the copilot to make changes, it generates a set of canvas operations (add block, remove block, update config, add edge, remove edge) that are applied to your workflow. You can review and undo any changes.
Copilot Tools
The copilot has access to specialized tools for workflow manipulation:
| Tool | Description |
|---|---|
| Add Block | Place a new block on the canvas with initial configuration |
| Update Block | Modify a block's configuration (prompts, actions, conditions) |
| Remove Block | Delete a block and its connections |
| Add Edge | Connect two blocks with a data flow edge |
| Remove Edge | Disconnect two blocks |
| Read Canvas | Inspect the current workflow structure |
| Search Docs | Look up documentation for blocks, integrations, and patterns |
| Generate Visualization | Draw a chart, KPI, or table in the chat from your connected tables, using the same components as dashboards |
| Dashboard Widget | Create, update, or remove a dashboard widget -- a chart, metric, table, or a full multi-part board with its own filters -- and change dashboard settings (currency, locale, negatives, fiscal year) |
Tips for Effective Prompting
-
Be specific about what you want. Instead of "make it better," say "add error handling to the Stripe block with 3 retries and exponential backoff."
-
Reference blocks by name or type. "Change the agent block" is clearer than "change the AI thing."
-
Describe the data flow. When building from scratch, mention where data comes from and where it should go: "when a Stripe payment is received, look up the customer in Salesforce by email, then send a Slack message."
-
Ask one thing at a time. Complex requests work better as a sequence of smaller instructions.
-
Use the copilot for exploration. Not sure which block to use? Ask "what's the best way to handle X?" before asking it to build.
The copilot operates on your draft canvas. Changes are not saved until you explicitly save or deploy the workflow.
Related
Agent Block
Use AI models within your workflows for processing and extraction
Blocks Overview
All 29 block types available for workflow building
Quickstart
Create your first workflow step by step
Loopfour