Every workflow execution is recorded. History (Workflow module sidebar) is where you review what ran, what it produced, and what went wrong.
What a run record contains
Select a run to inspect:
| Element | What it tells you |
|---|---|
| Status | Completed, running, awaiting approval, or failed. |
| Timestamps | Start time and duration — useful for spotting slow steps. |
| Node-by-node progress | Each node's status (pending / running / success / error) on the workflow view. |
| Inputs | The file or values the run started with. |
| Outputs and artifacts | Results, including files written by Output File nodes for download. |
| Approval state | Whether a User Approve step was approved, rejected, or is still waiting. |
| Telemetry / event log | A chronological log of execution events — tool calls, model responses, errors. |
Common review tasks
- Verify an output — open the run, check the final node output or download the produced file.
- Find a bottleneck — compare node timings in the event log; LLM and tool nodes are the usual costs.
- Audit a decision — expand the relevant node to see the exact prompt/instructions and returned data, including the agent's reasoning trail in agentic runs.
- Resume or approve — runs waiting on a User Approve step are actioned from here.
Troubleshooting failures
| Symptom | First checks |
|---|---|
| A node failed with an error | Open the event log entry for that node — the message names the cause (invalid input, tool unavailable, format check failed). |
| Format Check failed | The previous node didn't produce the expected structure. Loosen or correct the check, or strengthen the instructions in the producing node. |
| Tool node failed | Confirm the tool server is connected and you have permission. See Tools. |
| LLM node slow or empty | Check node timeouts/settings, and whether the model is available under load (GPU Monitoring). |
| Run stuck "awaiting approval" | Someone with the right permission needs to action the approval. |
| Token/limit errors | Split large inputs, reduce map iterations, or ask your administrator about capacity. See Usage & Credits. |
Safe re-run: fix the offending node's configuration, then run the workflow again. History preserves earlier attempts, so comparisons are possible.
Housekeeping
Runs accumulate over time. Where your permissions allow, tidy old records so the list stays navigable — and export anything you need for compliance from the run detail before removing it.
Related
- Workflow Builder · Workflow Nodes
- Agentic Workflows — auditing autonomous runs
- Usage & Credits — cost of automation