Every node, every connection, every config. It already has the full picture. Ask how something works, where the data comes from, or what a node cost, and get a straight answer grounded in your actual workflow, not a generic guess.
No setup and no copy-paste. The Copilot pulls everything it needs the moment you open it.
It gets the full definition automatically, every node, every edge, every config. Secrets and keys are stripped out before it ever sees them.
Ask about a node and it pulls the config, the model, the prompt, and the node's most recent run, including what that run cost.
Hybrid vector and keyword search over the documents attached to the workflow, with the source named so you can check it.
Looks through the files your past runs wrote, with filtering and sorting by relevance or recency.
Trace how a value moves from one node to the next, or ask why a step did what it did. The answer points at real nodes and real numbers.
When the Copilot names a node, the name is a chip. Click it and the canvas highlights that exact node, even after you rename it.
The same brain as your agents. The Copilot searches your knowledge bases and storage with the exact tools your AI nodes use, so what it tells you matches what your agents actually see. It caches your workflow context too, so the second question comes back faster and cheaper than the first.