Skip to main content

​
Monitor agent health and performance in Nora

View platform metrics, agent-level CPU and memory stats, activity events, cost tracking, and historical performance data for your agents in Nora.
Nora collects runtime metrics, activity events, and cost data for every agent you deploy. You can view a live snapshot of an agent’s resource usage, pull historical stats over configurable time windows, browse the activity event log, and check accumulated cost data — all from the dashboard or the monitoring API. This page explains what data is collected, how to read it, and the available time ranges for historical queries.

​
Platform metrics overview

Platform monitoring — agent counts, queue depth, and recent activity at a glance The platform-level monitoring endpoint returns aggregate statistics across all agents you can access, including directly owned agents and agents shared with you through workspaces. It includes totals for running agents, queued jobs, and recent activity.
Use this endpoint to get a high-level picture of your deployment before drilling into individual agents.

​
Fleet needs-attention

The operator dashboard (/app) shows a Needs attention strip at the top that lists the agents you should look at right now and why. It is derived server-side at read time from existing state — no extra configuration — and stays hidden when the fleet is healthy, so it never adds noise to an all-clear deployment. It surfaces these reasons: Errors sort before warnings. The strip covers the agents you can access — directly owned plus workspace-shared. It is backed by the fleet needs-attention roll-up API (GET /monitoring/fleet-status, requires the monitoring:read scope), and the nora doctor Fleet-health check is built on the same roll-up.

​
Agent-level metrics

Each agent exposes a live stats snapshot and a historical stats series. Stats are collected every 5 seconds and retained for 7 days.

​
Live stats

To fetch the current resource usage for a running agent:
The response includes the following fields:
If the agent is stopped or has no container, the endpoint still returns 200 with current.running: false and an error field (e.g. “No container assigned”) rather than live metrics.

​
Historical stats

To fetch a time series of stats for an agent, use the history endpoint with a time range:
Supported time ranges: You can also query a custom window using ISO timestamps:
Each data point in the history response contains the same metric fields as the live stats endpoint, plus a recorded_at timestamp. Results are ordered by recorded_at ascending and capped at 5000 rows for short windows (≤ 1 hour, returned un-aggregated) and 4000 rows for larger windows, which are time-bucketed/aggregated server-side. Historical stat fields:

​
Activity events

The events log records actions taken on agents and by agents — deployments, redeployments, errors, and other lifecycle events.
To filter events to a specific agent:
You can also control the number of results returned:
The default limit is 50 events, ordered by created_at descending (newest first).

​
Agent metrics summary

For a summarized view of an agent’s metrics over the default 24-hour window:
This returns aggregated highs, lows, and averages rather than individual data points — useful for a quick health check. For the full time-series data with optional type and date filters:

​
Agent cost tracking

Workspace cost panel — daily provider spend stacked across the past week Nora tracks token usage for each agent and estimates spend from model rates. The operator cost dashboard groups token spend by workspace, links each row back to the related agent, and reports a unique fleet total so an agent assigned to multiple workspaces is not double-counted. Each agent row can show token usage grouped by model. To view the all-workspaces cost dashboard through the API:
Use period_start=YYYY-MM-DD&period_end=YYYY-MM-DD for a custom UTC date range. The dashboard also supports searching by workspace, agent, model, or provider. The response includes workspaces[], unassigned.perAgent[], workspaceTotalUsd, uniqueFleetTotalUsd, and each agent’s cost_details.tokens. Use uniqueFleetTotalUsd for fleet-wide totals and workspace totals for team-level budgeting. Token cost uses COST_MODEL_RATES_JSON when the metric row includes model metadata. Rows without model metadata are grouped as Unknown model and priced with COST_PER_1K_TOKENS. To view the accumulated cost data for one agent:
The response includes token totals and usage breakdowns. If the agent exists but has no recorded token usage in the selected window, totals are zero. To view one workspace’s cost breakdown:

​
Per-agent budget caps

You can cap an agent’s LLM spend so it pauses itself before it runs away. Open an agent’s Settings tab and set one budget per period — daily, weekly, or monthly — each with a USD hard limit and a soft-threshold percent (default 80). Nora enforces budgets automatically:
  • When spend crosses the soft threshold, Nora emits an agent.budget_soft_exceeded event as an early warning.
  • When spend reaches 100% of a period’s limit, Nora pauses the runtime automatically, emits agent.budget_exceeded, and sets paused_reason = budget_exceeded.
  • Manually starting a still-over-cap agent clears paused_reason, but the budget sweep re-pauses it on its next cycle (within roughly a minute) while it remains over cap.
Manage budgets through the API with GET /api/agents/:id/budget, PUT /api/agents/:id/budget, and DELETE /api/agents/:id/budget/:budgetId (see the agents API reference).

​
Performance data

Platform-level API performance metrics are available for the last 24 hours by default:
You can narrow the window with a since parameter:
Use the 24-hour historical stats window combined with the activity events log to correlate performance dips with specific deployment or lifecycle events.