> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pinkfish.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Monitor Tab

> Inspect every past run of an Agent Mode workflow — execution timing, node-level inputs and outputs, streaming logs, and agent reasoning traces.

The **Monitor tab** is where you audit and debug runs after the fact. It's the second of the two view modes in the Agentic Editor (alongside **Editor**), and switching into it replaces the chat panel with a list of runs and the right pane with a read-only execution view.

## Switching to Monitor

At the top of the editor, use the **Editor / Monitor** toggle. Flipping to **Monitor**:

* Replaces the chat panel with a **Runs Panel** (list of past executions).
* Swaps the Workflow tab for a read-only **Monitor canvas** with execution-state overlays.
* Opens a **Node side-panel** on demand with inputs/outputs/logs/timing for any node you click.

The Workflow canvas here is a **viewer**, not an editor — you can't drag nodes or change wiring while in Monitor mode. Flip back to **Editor** to make changes.

## The runs list

The left panel shows every run of this workflow, most-recent first. Each row displays:

* **Status** — success (green), error (red), running (pulsing yellow), paused (purple), cancelled (gray). A **paused** run is suspended at a durable wait — a delay, a timer, or an approval — and resumes on its own.
* **Trigger** — what kicked it off (Manual, API, Schedule, Email, App, Event).
* **Timestamp** — relative ("3 minutes ago") and absolute on hover.
* **Duration** — total elapsed wall-clock time.

### Filtering

Controls at the top of the runs list let you narrow down:

* **Date range** — restrict to runs from a specific time window.
* **Status** — show only successes, only errors, etc.
* **Trigger type** — restrict to runs from a specific kind of trigger.

### Exporting

Click the **Download CSV** button to export the currently-filtered run list as a CSV file. Useful for post-mortem analysis in external tools or reporting.

## The Monitor canvas

Selecting a run loads its execution state onto the main canvas. Every node is colored and badged to reflect what happened to it during that specific run:

* **Gray** — not executed (typically because it was on a skipped branch).
* **Yellow pulse** — running (for an in-progress run).
* **Green check** — completed successfully.
* **Red X** — failed.
* **Dimmed slash** — skipped explicitly.

## Node side-panel

Click any node on the Monitor canvas to open the **Node side-panel** with four sub-tabs:

<CardGroup cols={2}>
  <Card title="Inputs" icon="arrow-down-to-bracket">
    The exact arguments the node received, rendered as JSON.
  </Card>

  <Card title="Outputs" icon="arrow-up-from-bracket">
    The node's return value. The same view modes as the Result tab (JSON, table, markdown, HTML).
  </Card>

  <Card title="Logs" icon="scroll">
    Streaming output from this specific node — stdout, tool-call traces, errors.
  </Card>

  <Card title="Timing" icon="stopwatch">
    Duration, started-at, ended-at, and any retries.
  </Card>
</CardGroup>

## Agent Debugger Panel

For runs that involve an Agent node or any LLM-driven step, the **Agent Debugger Panel** exposes a detailed trace of the agent's reasoning:

* **Messages** — the full conversation history the agent saw.
* **Tool calls** — every tool the agent invoked, with arguments and results.
* **Thoughts / reasoning** — the agent's internal monologue, where exposed by the model.

This view is essential when an agent-driven workflow produces an unexpected result — you can see exactly why the agent made the decisions it did.

## Bottom detail panel

Below the main panel, the **Monitor Bottom Panel** aggregates additional detail about the selected run — metadata, batch-run siblings (if this is part of a batch), triggering event payload, and any run-level artifacts.

## Streaming live runs

Runs that are currently in progress stream their state into the Monitor view in real-time:

* Node statuses update as each node completes.
* Log entries append live.
* The **Streaming** badge in the run list row indicates a live run.

You can kick off a run from the Editor, switch to Monitor, and watch the execution progress without refreshing.

## Run artifacts

If the workflow produced files (a generated report, a scraped dataset, a rendered image), they surface as **run state files** — downloadable artifacts attached to the run. Access them from the run's detail view.

## Related: global Monitor

The Monitor tab documented here is **per-workflow**. Pinkfish also has a **global Monitor** — a dashboard of runs across every workflow in your org. See [Monitor](/orchestration/monitor) for the org-wide view. The [Agent Monitor](/orchestration/agent-monitor) and [Guardian Agent](/orchestration/guardian-agent) offer AI-driven observability on top.

## What's next

<CardGroup cols={2}>
  <Card title="Running a workflow" icon="play" href="/agentic-editor/running">
    The execution flow whose runs end up in Monitor.
  </Card>

  <Card title="Canvas" icon="diagram-project" href="/agentic-editor/canvas">
    The Editor counterpart to the Monitor canvas.
  </Card>

  <Card title="Global Monitor" icon="chart-column" href="/orchestration/monitor">
    Org-wide run observability.
  </Card>

  <Card title="Guardian Agent" icon="shield" href="/orchestration/guardian-agent">
    AI-driven analysis of runs and failures.
  </Card>
</CardGroup>