Canvas
See the whole pipeline. Change any part of it.
Every source, join, transform, check, and output sits on one canvas. The agent adds nodes where you can watch them land, and every node opens to its schema, preview, and last run.
Chat changes the canvas, not a hidden plan
The graph is the state
Chat changes the canvas, not a hidden plan
When the agent adds a node, you see the family, the instruction, and the edges it connects before it exists. Accept it and the card appears on the graph with its status.
- Three CSV sources joined in two steps, built from one sentence
- Each card shows its family, its file or instruction, and its status
- Status moves from ready to running to done on the node itself
Canvas in detail
01
Every family on one surface
Source, profile, clean, transform, mapping, join, validate, and output nodes share one canvas.
02
Schema flows downstream
Each node carries the columns it receives and the columns it emits, so the next step is not a guess.
03
Preview in place
Open a node to read sample rows from its last output without leaving the graph.
04
Status on the card
Ready, running, done, or needs a run shows on the node that owns it.
05
Add from anywhere
Add a node from the toolbar, from a node's outgoing handle, or by describing it in chat.
06
One click to tidy
Auto-layout lines the graph up left to right when a build leaves it crowded.
Related capabilities
Transformations
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 analysisRuns
A node, a branch, or the full pipeline
Run a node, a downstream branch, or the whole pipeline in isolated, observable jobs.
Explore runs
Ready to build a pipeline you can inspect?
Describe the outcome, inspect what the agent proposes, and run it when you trust it.