Approvals
The agent proposes. You decide what is trusted.
Node changes and new scripts arrive as proposals. You read the code, the lineage, and the last run, then approve or reject with a reason.
Every node change asks first
Proposals in chat
Every node change asks first
When the agent wants to add, change, or delete a node, the chat shows the family, the instruction, and the edges it will connect, then waits. Approve it and the node lands on the canvas.
- Add, update, and delete node actions ask first
- The proposal shows the instruction and its connections
- Approved nodes appear on the canvas with their status
Approvals in detail
01
Proposals, not surprises
Adding, changing, or deleting a node shows up as a card you approve or reject.
02
Script review
A generated script pauses its run until someone reads it and approves it.
03
One inbox
Everything waiting on you across a project sits in one approvals list that refreshes on its own.
04
Reasons on record
Attach a reason to an approval or a rejection so the next person knows why.
05
Cached scripts
Skip the prompt when a script has not changed since it was approved. New and edited scripts always ask.
06
Auto-approve when you choose
YOLO mode lets the agent apply its own actions without asking. Turn it off to review each one again.
Related capabilities
Canvas
The pipeline as one editable graph
Sources, joins, transforms, validation, and outputs on one graph you can see and change.
Explore canvasTransformations
Generated Python and SQL on each node
Describe the step. The agent writes Python and SQL from the upstream schema and attaches it to the node.
Explore transformationsAnalysis
Questions answered from saved outputs
Ask in plain language. Answers come from persisted node outputs, and new work becomes a node you can rerun.
Explore analysis
Ready to build a pipeline you can inspect?
Describe the outcome, inspect what the agent proposes, and run it when you trust it.