Agent run monitoring

What you’ll build: a control room for the agents you run — every run logged, the success rate trending, and failures surfaced immediately. This is the core PulseDeck use case: your Hermes / CrewAI / LangGraph / n8n agents reporting on themselves.

pulsedeck.app / dashboards / agents

SCREENSHOT

Finished agent dashboard: a runs-over-time chart, a success-rate metric, and a feed of failed runs.

Drop an image into public/shots/ and pass src.

What you’ll need

PulseDeck running (Quickstart).
A registered source and its pd_… API key.
An agent or pipeline whose runs you want to track. Post once at the end of each run.

1. Plan the structure

Category: agents
Stream: one per agent, e.g. research-bot

2. Send the report

At the end of every run, the agent posts a report describing how it went: a metric for the success rate, a status block for the steps, an alert if the run failed, and an artifact linking to the full output or logs.

curl -X POST http://localhost:3001/api/v1/reports \
  -H "Authorization: Bearer pd_xxxxx" \
  -H "Idempotency-Key: research-bot-run-8412" \
  -H "Content-Type: application/json" \
  -d @run.json

{
  "version": "1.0",
  "source": { "id": "src_…" },
  "category": { "slug": "agents" },
  "stream": { "slug": "research-bot" },
  "report": {
    "title": "research-bot — run #8412",
    "summary": "Completed with 1 step retried.",
    "severity": "warning",
    "occurred_at": "2026-06-23T12:30:00Z",
    "tags": ["research-bot", "run"]
  },
  "blocks": [
    {
      "id": "rate", "type": "metric",
      "key": "success_rate", "label": "Step success rate", "value": 92.3,
      "unit": "%", "format": "percent", "precision": 1,
      "trend": "down", "sentiment": "negative", "delta": -3.1, "comparison_label": "vs last run"
    },
    {
      "id": "steps", "type": "status",
      "items": [
        { "key": "fetch", "label": "Fetch sources", "status": "healthy" },
        { "key": "summarize", "label": "Summarize", "status": "degraded" },
        { "key": "publish", "label": "Publish", "status": "healthy" }
      ]
    },
    {
      "id": "warn", "type": "alert",
      "title": "Summarize step retried", "severity": "warning",
      "message": "Rate-limited by the model API; succeeded on retry 2."
    },
    {
      "id": "output", "type": "artifact",
      "label": "Full run log", "url": "https://logs.example.com/runs/8412",
      "mime_type": "text/plain"
    }
  ]
}

Set severity to critical when the run fails outright, warning on retries, info on a clean run — so the failure feed only shows what needs attention.

3. Build the dashboard

A report_count widget — runs over time (cadence and gaps at a glance).
A metric widget — the latest success rate.
An alert_feed scoped to the agents category — failed and degraded runs surface here first.

{
  "version": 1,
  "widgets": [
    { "id": "cnt", "type": "report_count", "span": "half", "row": 0,
      "config": { "scope": "category", "targetId": "cat_agents", "bucket": "day", "range": "30d" } },
    { "id": "rate", "type": "metric", "span": "half", "row": 0,
      "config": { "streamId": "str_research_bot", "metricKey": "success_rate", "label": "Success rate" } },
    { "id": "fails", "type": "alert_feed", "span": "full", "row": 1,
      "config": { "categoryId": "cat_agents", "limit": 15 } }
  ]
}

pulsedeck.app / dashboards / agents / edit

SCREENSHOT

Builder with a runs-over-time count, a success-rate metric, and a full-width failure feed.

Drop an image into public/shots/ and pass src.

Make it your own

Many agents — one stream per agent under agents; the count widget can scope to the whole category or a single stream.
One workspace per client (agencies) — each client’s agents report into their own workspace; you get a portfolio view per client.
Link everything — the artifact block turns each run into one click to the full logs or output.