Executions
The Executions page shows the history of every workflow run — what triggered it, which nodes ran, how long each took, what they produced, and whether they succeeded or failed.
Execution List
Navigate to Executions in the left sidebar. The list shows all executions across all workflows in your project, sorted newest-first.
Each row shows:
| Column | Description |
|---|---|
| Status | ✅ Completed / ❌ Failed / 🟡 Paused (HiL) / ⏳ Running |
| Workflow | Name of the workflow that ran |
| Trigger | How it was triggered (Manual / Webhook / Schedule) |
| Started | Timestamp when execution began |
| Duration | Total time from start to completion |
| Initiated by | User who triggered (or "Scheduled" for cron runs) |
Filtering
Filter by:
- Workflow — Show executions for a specific workflow
- Status — Show only completed / failed / paused
- Date range — Executions from a specific period
Execution Detail View
Click any execution row to open the detail view. This shows:
Overview Panel
- Execution ID — Unique identifier
- Status — Current status
- Workflow — Which workflow ran
- Trigger input — What
{{trigger.output}}contained - Start/end time — Full timestamps
- Total duration — End-to-end time
Node Results
Expandable list of all nodes in the workflow, in execution order:
▶ [Trigger] — Completed in 0.01s
▶ [Agent: researcher] — Completed in 8.4s
▼ [Agent: writer] — Completed in 3.2s
Input:
Research findings:
[Quantum computing is the use of...]
Please format as a blog post.
Output:
## Quantum Computing: A Beginner's Guide
Quantum computing represents one of the most...
Tool Calls: none
▶ [Output] — Completed in 0.01s
Node Expansion Details
Expanding a node shows:
| Section | Content |
|---|---|
| Input | The fully-resolved input text sent to the node (expressions already substituted) |
| Output | The node's output text |
| Duration | Time taken for this node |
| Tool Calls | List of capability/MCP tool calls made (for agent nodes) |
| Error | Error message and stack trace (if failed) |
Execution States
| State | Indicator | Description |
|---|---|---|
| Running | Blue pulsing dot | Actively executing |
| Completed | Green checkmark | All nodes finished successfully |
| Failed | Red X | One or more nodes failed |
| Paused | Yellow clock | Waiting for Human-in-Loop approval |
| Rejected | Red stop | HiL decision was "reject" |
| Cancelled | Gray dash | Manually cancelled |
Re-Running an Execution
You can re-run any completed or failed execution:
- Open the execution detail.
- Click Re-run (top right).
- A modal appears — the original input is pre-filled.
- Optionally modify the input.
- Click Execute.
A new execution starts. The original execution record is not modified.
If you re-run without changing the input, you'll get a fresh execution with the same trigger input. For non-deterministic agents (temperature > 0), the output will likely differ slightly each time.
Cancelling a Running Execution
If an execution is taking too long or you need to stop it:
- Open the Executions list (or the workflow detail page).
- Find the running execution (blue indicator).
- Click Cancel.
Cancellation is immediate — TARX stops the execution as soon as the signal is received, interrupting the node that's mid-flight (including an in-progress LLM call), so you stop spending tokens right away. Already-completed nodes' outputs are preserved in the record, and the run is marked Cancelled.
Deleting a Run
You can permanently remove a finished execution — handy for clearing out cancelled runs or early errors, similar to deleting a pipeline run in a CI system.
- Open the workflow detail page and hover over a finished run.
- Click the trash icon on the right.
- Confirm in the dialog.
The execution record is deleted from the database and its stored logs are removed from blob storage. This is permanent and cannot be undone.
Workflows are a data source for Canvas charts. Deleting a run removes its data from any Canvas that aggregates this workflow — charts will have one fewer data point on their next refresh.
A run that is Running, Pending, or Paused (awaiting approval) cannot be deleted. Cancel it first, then delete.
Human-in-Loop Approvals
Paused executions (waiting for HiL approval) appear with a 🟡 yellow indicator. To approve or reject:
- Click the paused execution.
- Find the Human-in-Loop node in the node list.
- See the display content shown for review.
- Enter an optional comment in the Comment field.
- Click Approve (green) or Reject (red).
After your decision, the execution status updates and the workflow resumes (if approved) or ends (if rejected).
See Human-in-Loop for full details.
Execution Logs
For each execution, detailed logs are stored securely in the cloud. These include:
- Full LLM prompts and responses (for debugging)
- Tool call details (arguments and results)
- Timing information for each step
- Error tracebacks
Detailed logs are retained for 30 days. After that, the execution summary (status, duration, per-node statuses) remains available, but the full input/output text is no longer retained.
Execution Statistics
The Dashboard (home page of your project) shows aggregate execution stats:
| Metric | Description |
|---|---|
| Total executions | All-time execution count |
| Success rate | % of executions that completed without error |
| Average duration | Mean execution time |
| Executions today | Count for the current day |
| Failures today | Count of failed executions today |
These stats help you monitor workflow health and identify problematic workflows.
Debugging Failed Executions
When a workflow fails, the Executions page is your primary debugging tool.
Step 1: Identify the failing node
- In the node list, the failed node shows a ❌ indicator
- The error message appears at the top of the execution detail
Step 2: Check the failed node's details
- Expand the failing node
- Read the Error section — it shows the exception type and message
- Check the Input — was the node receiving the right data?
Step 3: Common failure causes
| Error | Likely Cause | Fix |
|---|---|---|
AgentNotFound | The agent referenced in the node was deleted | Re-select the agent in the canvas node inspector |
APIKeyNotFound | The agent's API key was deleted | Add a new key in Settings → Keys & Secrets, update the agent |
ProviderAuthError | API key is invalid or expired | Check your API key in the provider's dashboard |
ProviderRateLimitError | Hit provider rate limits | Reduce execution frequency or upgrade your provider plan |
ToolCallMaxRoundsExceeded | Agent used more than 10 tool rounds | Simplify the agent's task or increase the limit (contact support) |
ExpressionError | Expression references non-existent variable | Check the expression — does the variable name match the agent name exactly? |
HiLTimeoutError | HiL timed out waiting for approval | Increase timeout or approve within the window |
Step 4: Fix and re-run
- Make the necessary fix (update agent, fix expression, etc.)
- Click Re-run from the execution detail
- Verify the re-run succeeds
Execution Retention
Execution records are stored indefinitely. However:
- Detailed logs are retained for 30 days
- After 30 days, only the summary (status, duration, node statuses) is available — not the full input/output text
For compliance-sensitive workflows, consider exporting execution records before the 30-day log retention window.