Real-Time Monitoring
Watch workflows execute node-by-node with WebSocket-powered updates. Track execution timelines, monitor AI token costs, and stay informed as automations run.
Live Status Updates
A91I uses WebSocket connections to push execution updates to your browser in real time. You do not need to refresh the page — nodes light up, progress bars advance, and status badges update automatically.
Connection Indicator
A small badge in the header shows the WebSocket connection status:
| Status | Meaning |
|---|---|
| Connected | Live updates are flowing. You see changes as they happen. |
| Connecting | The client is attempting to establish a connection. Updates may be delayed. |
| Disconnected | No live connection. The page falls back to periodic polling (every 5 seconds). |
If disconnected, the client reconnects automatically with exponential backoff. Click the indicator to force a reconnection attempt.
Event Types
Two primary event types drive the live UI:
- execution:status — fires when the overall execution state changes (Pending → Running → Completed).
- node:status — fires when an individual node starts, completes, or fails. Includes the node ID, status, and output preview.
Canvas During Execution
When you start a test run from the editor, the canvas becomes a live dashboard. Each node shows its current state through visual cues:
- Blue pulsing border — the node is currently executing.
- Green border — the node completed successfully.
- Red border — the node failed.
- Grey border — the node is pending or was skipped.
Hover over any completed node to see a tooltip with a preview of its output (first 200 characters). Click the node for the full output in the configuration panel.
Execution Timeline
The execution detail page includes a Gantt-chart-style timeline that visualizes when each node started and how long it took.
Reading the Timeline
- Each row represents a node in the workflow.
- The horizontal bar shows the node's execution duration relative to the total execution time.
- Bar width is percentage-based — a node that took 50% of the total time fills half the row.
- Colors match the node status: green for completed, red for failed, blue for running.
- Hover over a bar to see the exact start time, end time, and duration in seconds.
Toggling the Timeline
The timeline panel is collapsible. Click the toggle to show or hide it. When hidden, more screen space is available for the node-by-node details.
Cost Tracking
Every execution that uses AI nodes tracks token usage and estimated cost. This information is available in the execution detail page.
What Is Tracked
| Metric | Description |
|---|---|
| Total tokens | Sum of input + output tokens across all AI nodes in the execution. |
| Input tokens | Tokens consumed by the prompts sent to the model. |
| Output tokens | Tokens generated by the model's response. |
| Estimated cost | Calculated from the model's per-token pricing. |
Cost Distribution
The cost breakdown chart shows how tokens are distributed across AI nodes in the execution. This helps identify which prompts are most expensive and where optimization would have the biggest impact.
Reducing AI costs
Live Log Streaming
The execution detail page includes a log panel that streams output as nodes execute. Logs appear in real time via WebSocket — you see each node's output the moment it finishes processing.
- Auto-scroll — the log panel automatically scrolls to the latest entry as new logs arrive.
- Manual scroll lock — scroll up in the log panel to pause auto-scrolling. New entries continue to accumulate but the view stays where you left it.
- Resume — scroll back to the bottom to re-enable auto-scrolling.
Dashboard Monitoring
The main dashboard provides an organizational overview of execution health:
- Executions This Month — total count and success rate percentage.
- Execution status chart — stacked bar showing completed, failed, running, and cancelled distributions.
- Recent activity feed — the latest execution events across all workflows.
- Token usage — monthly AI consumption and estimated cost.
Use the dashboard for a bird's-eye view, then drill into specific execution details when you need to investigate an issue.