Workflow Analytics

Monitor runs, measure performance, and track token cost for your workflows.

Analytics are available at two scopes:

Scope	Who can access	Where
Per-workflow	agent_developer role or higher within the workflow's space	API + Editor Runs tab
Platform-wide	Admin only	Admin > Analytics > Agents

---

Per-workflow analytics

Every workflow developer has access to run history and aggregate metrics for their own workflows — no admin access required.

Run history

GET /api/v1/agent-analytics/workflows/{workflow_id}/runs
Authorization: Bearer <token>

Query parameters:

Parameter	Type	Description
status	string	Filter by run status: running, completed, failed, waiting_decision
limit	int	Page size (1–200, default: 25)
offset	int	Pagination offset (default: 0)
start_ts	int	Unix timestamp — only runs created after this time
end_ts	int	Unix timestamp — only runs created before this time

Response:

{
  "items": [
    {
      "id": "run_abc123",
      "workflow_id": "wf_xyz",
      "space_id": "space_def",
      "catalog_id": "cat_ghi",
      "status": "completed",
      "triggered_by": "user_id_or_webhook_token",
      "started_at": 1700001000,
      "ended_at": 1700001003,
      "duration_ms": 3186,
      "error_message": null,
      "prompt_tokens": 512,
      "completion_tokens": 128,
      "total_tokens": 640,
      "cost_usd": 0.0012,
      "provider": "anthropic",
      "model_id": "claude-sonnet-4-6"
    }
  ],
  "total": 42
}

Aggregate summary

GET /api/v1/agent-analytics/workflows/{workflow_id}/summary
Authorization: Bearer <token>

Query parameters: start_ts, end_ts (same as run history).

Response:

{
  "total_runs": 42,
  "completed_runs": 38,
  "failed_runs": 3,
  "paused_runs": 1,
  "success_rate": 0.9048,
  "avg_duration_ms": 3186.0,
  "p95_duration_ms": 7842.0,
  "total_prompt_tokens": 21504,
  "total_completion_tokens": 5376,
  "total_cost_usd": 0.0504
}

success_rate is completed_runs / total_runs (0–1).

---

Run states

Status	Meaning
running	Actively executing nodes
waiting_decision	Paused at a user_approval node; awaiting inbox decision
ready_to_resume	Decision made; waiting for the next execution slot
completed	All paths reached an end node successfully
failed	An error occurred; run terminated
cancelled	Manually cancelled via the API

---

Editor: Runs tab

Inside the Agent Builder editor, click the Runs tab in the right panel to see the most recent runs for the current workflow without leaving the editor. Each row shows status, elapsed time, and total token cost.

---

Evaluate panel

The editor's Evaluate panel (test cases) shows per-run metrics after each evaluation:

• Latency — end-to-end duration in milliseconds
• Tokens — total tokens consumed (prompt + completion)
• Cost — estimated USD cost based on the model's pricing

Use the Evaluate panel to compare different versions of a workflow before publishing.

---

Admin-level analytics

Platform admins can see analytics across all workflows, catalogs, and node types at Admin > Analytics > Agents.

Platform summary

GET /api/v1/agent-analytics/summary
Authorization: Bearer <admin-token>
  ?catalog_id=...
  &space_id=...
  &workflow_id=...
  &start_ts=...
  &end_ts=...

All runs (paginated)

GET /api/v1/agent-analytics/runs
Authorization: Bearer <admin-token>
  ?status=failed
  &limit=50

The admin runs table displays workflow names alongside run IDs — no need to look up workflow IDs separately.

Node-type performance

GET /api/v1/agent-analytics/node-stats
Authorization: Bearer <admin-token>
  ?catalog_id=...
  &workflow_id=...

Response:

[
  {
    "node_type": "agent",
    "count": 1240,
    "avg_latency_ms": 1823.5,
    "p50_latency_ms": 1600.0,
    "p95_latency_ms": 4200.0,
    "error_count": 12
  },
  {
    "node_type": "classify",
    "count": 620,
    "avg_latency_ms": 412.1,
    "p50_latency_ms": 380.0,
    "p95_latency_ms": 920.0,
    "error_count": 2
  }
]

Results are sorted by avg_latency_ms descending — the slowest node types appear first.

Run timeline (per run)

GET /api/v1/agent-analytics/runs/{run_id}/timeline
Authorization: Bearer <admin-token>

Returns the ordered list of SSE events stored for a specific run — useful for replaying what happened during a failed run or debugging unexpected output.

---

Related

• Run Infrastructure — SSE event types, pause/resume, rate limiting
• Triggers — run_id returned immediately when a webhook fires a workflow
• Workflow Lifecycle — publishing gates that affect which runs appear in analytics