Salt AI Docs
Running Pipelines

Monitoring Execution

Track pipeline progress in real time and debug errors.

Salt provides real-time feedback while your pipeline runs so you always know what's happening and where.

Real-Time Progress

When you run a pipeline, the canvas becomes a live dashboard:

  • Pending nodes remain in their default state, waiting for upstream data.
  • Running nodes are highlighted to show they are actively processing.
  • Completed nodes update their visual state to indicate success. You can click them to inspect output.
  • Failed nodes are marked with an error indicator so you can immediately see where something went wrong.

Progress updates arrive through a WebSocket connection, so there's no need to refresh -- the canvas updates automatically as each node transitions between states.

Inspecting Node Output

After a node completes (or while the pipeline is still running), click on it to open the Properties panel. The panel shows:

  • The node's configured inputs
  • The output data produced by this execution
  • Any warnings or messages from the node

This lets you verify intermediate results at each step of your pipeline, not just the final output.

Handling Errors

When a node encounters an error during execution:

  1. The node is marked with an error indicator on the canvas.
  2. Click the node to see the error message in the Properties panel.
  3. Other branches of the pipeline that don't depend on the failed node may still complete normally.

Common Causes of Errors

IssueWhat to Check
Missing required inputMake sure all required fields have a value or an incoming edge.
Incompatible dataVerify that upstream nodes are producing the expected data type.
Service timeoutThe operation took too long. Try simplifying the input or check your connector configuration.
Credential errorYour API key or credential may be expired or invalid. Check the Connectors settings.

After fixing an issue, you can re-run the pipeline by pressing Play again. Salt re-executes the entire pipeline from the beginning.

Agent Iterations

Some pipelines use agent nodes that run iteratively -- the node executes multiple cycles, refining its output or gathering additional information with each pass. During agent execution:

  • The node shows iteration progress (e.g., "Iteration 3 of 10").
  • You can monitor each iteration's output as it completes.
  • The node finishes when it reaches its goal or hits the maximum iteration count.

Agent iterations are visible in real time, just like regular node execution.

Next Steps

On this page