Copilot is built into the workflow editor. It reads the workflow you have open, understands the blocks and connections that already exist, and can explain, edit, debug, and improve that workflow from natural language.
For workspace-wide work across workflows, tables, files, knowledge bases, and scheduled jobs, use Mothership.
Copilot is included with hosted Recall plans. Self-hosted deployments need a Copilot API key configured in COPILOT_API_KEY.
What Copilot Can Do
Copilot can work directly against the current workflow:
- Add, configure, connect, and delete workflow blocks
- Edit existing block settings without rebuilding the whole workflow
- Explain what the workflow does and how data flows through it
- Debug failed runs by reading execution logs and current block state
- Suggest leaner block choices, better models, safer credentials, and cleaner data flow
Modes
Use the mode selector at the bottom of the Copilot input to choose how much agency Copilot has.
Ask
Read-only help. Copilot can inspect and explain, but it cannot create, edit, or delete workflow resources.
Build
Workflow editing mode. Copilot can add blocks, wire connections, update settings, and debug issues.
Plan
Planning mode. Copilot drafts the exact implementation steps without making changes.
Models
Use the model selector at the bottom of the Copilot input to choose the model for the next workflow-scoped response. Faster models are better for quick explanations and small edits; stronger models are better for deep debugging and complex workflow design.
The available model list is loaded dynamically from Recall's Copilot backend, so it may change as new models are rolled out.
Context
Copilot automatically receives the current workflow as context. You can add more context from the input:
- Type
@to reference workflows, workflow blocks, logs, knowledge bases, tables, files, docs, templates, or prior chats - Use the
+menu to attach workspace resources or files - Drag workspace resources into the input when you want Copilot to use them
When you ask Copilot to edit an existing workflow, it reads what is already there before applying changes. You do not need to restate the current blocks, connections, or settings unless you want to override them.
Chat History
Click History in the Copilot header to see past conversations for the current workflow. You can resume any chat, start a new one, or delete old chats you no longer need.
File Attachments
Click the attachment icon or drag files into the input to include files with your message. Copilot can use images, PDFs, text files, JSON, XML, and other document formats as context.
Checkpoints
When Copilot modifies a workflow, it saves a checkpoint of the previous workflow state.
To revert, hover over the Copilot message that made the change, open the checkpoint menu, and choose the state you want to restore. Reverting cannot be undone, so review the checkpoint before confirming.
Thinking Blocks
For complex requests, Copilot may show an expandable thinking block while it works. These blocks make long-running edits easier to follow without mixing hidden reasoning into the final answer.
Usage
Copilot usage is billed per token and counts toward your plan's credit usage. If you reach your limit, enable on-demand billing from Settings → Subscription.
See the Cost Calculation page for billing and plan details.
Copilot MCP
You can use Copilot as an MCP server from external editors and AI clients. This lets you build, test, deploy, and manage Recall workflows from tools like Cursor, Claude Code, Claude Desktop, and VS Code.
Generating a Copilot API Key
- Sign in to tryrecall.com
- Navigate to Settings → Copilot
- Click Create
- Copy the key immediately — it is only shown once
The key will look like sk-recall-....
Cursor
Add the following to .cursor/mcp.json:
{
"mcpServers": {
"recall-copilot": {
"url": "https://dev.tryrecall.com/api/mcp/copilot",
"headers": {
"X-API-Key": "YOUR_COPILOT_API_KEY"
}
}
}
}Claude Code
claude mcp add recall-copilot \
--transport http \
https://dev.tryrecall.com/api/mcp/copilot \
--header "X-API-Key: YOUR_COPILOT_API_KEY"Claude Desktop
Claude Desktop requires mcp-remote. Add this to ~/Library/Application Support/Claude/claude_desktop_config.json on macOS:
{
"mcpServers": {
"recall-copilot": {
"command": "bunx",
"args": [
"mcp-remote",
"https://dev.tryrecall.com/api/mcp/copilot",
"--header",
"X-API-Key: YOUR_COPILOT_API_KEY"
]
}
}
}VS Code
Add this to settings.json or .vscode/settings.json:
{
"mcp": {
"servers": {
"recall-copilot": {
"type": "http",
"url": "https://dev.tryrecall.com/api/mcp/copilot",
"headers": {
"X-API-Key": "YOUR_COPILOT_API_KEY"
}
}
}
}
}For self-hosted deployments, replace {{APP_URL}} with your self-hosted Recall URL.