AI Copilot

The Copilot does the work on the canvas for you. Ask in your own words, and it creates the connectors and syncs, lays out the steps, deploys the pipeline and runs it, then tells you what it did and anything left for you to do.

What it can do

Say something like "copy new rows from our Postgres orders table into the warehouse every hour, then post to Slack when it's done". The Copilot works through the same steps you would:

• Looks at your workspace first, and reuses the connectors, syncs and components already there instead of creating duplicates.
• Creates connectors from the details you give it. It never makes up a host, database, user, password or key. If something required is missing, it asks you and stops.
• Sets up syncs: finds the source's streams and copies the ones you want, choosing incremental where a stream supports it.
• Builds the pipeline on the canvas: source and destination steps, the sync between them, and whatever should run after the data lands, including your custom components.
• Sets schedules, triggers and notifications, including near real-time.
• Deploys and runs the pipeline, but only when you asked it to.
• Fixes what it can. If a step fails, it reads the error and tries again. If only you can fix it (a wrong password, a firewall), it tells you what's wrong and stops.

It follows the same rules you do. For example, if a sync already runs in another pipeline, the Copilot reacts to it with a Sync from another pipeline step rather than copying the data a second time. See Orchestration.

What it changes shows up on the canvas like anything you built by hand, and you can edit it the same way.

What it asks you first

The Copilot asks you to confirm before it:

• deletes a connector,
• deletes a sync, or
• clears the data a sync has copied into its destination, so the next run copies everything again.

Nothing happens until you confirm. Once you do, it checks again that the action is still safe before carrying it out: for example, a connector a pipeline has started using since isn't deleted.

It acts as you

The Copilot uses your sign-in for every action, so it can do exactly what you can and nothing more. It goes through the same checks as the Deploy button.

It can't create secrets. If a notification needs a webhook URL and no suitable secret exists, it asks you to add one under Settings and stops.

Connector passwords and keys you type into the chat are used to set up the connector, then removed from the saved conversation.

Keep working while you're away

A long Copilot task normally stops and waits for you when your sign-in runs out, for example when you close the tab or your session ends.

To let it carry on, go to Settings → Copilot → Let Copilot keep working while I'm away and choose Allow. The Copilot then renews your sign-in on its own until the task is done. It still acts only as you, with your permissions, and only in that organization.

You can stop this in the same place at any time. An organization admin can stop it for anyone, and leaving the organization ends it.

Authoring a custom component

For custom Python components, the Copilot writes a draft rather than saving anything. Describe what the component should do (its inputs, what it computes, what it outputs) and the Copilot fills in the new-component form with parameters and Python source. Nothing is created until you review the code and save it yourself, because this is code your pipelines will run.

Current limits

• The Copilot can set up Postgres connectors only for now. Set up other connectors yourself; the Copilot can then use them.
• Check what it builds. It gets things wrong now and then, so look over the canvas before relying on a pipeline.