Monitoring & usage
This page explains where to find inference volume, latency, and warehouse activity in Heimdall by RejiCo — without a separate observability stack.
- At least one deployed API key (ML/Forecast/Loop) or unstructured key (Read/Vision) for meaningful charts
- Signed in at heimdallapp.org
Where to look
| Surface | How to open | What it shows |
|---|---|---|
| Usage | Sidebar → Usage (/usage) | Account intelligence + Lake/Lab tabs |
| Per-asset detail | Sidebar → ML / Forecast / Loop → click asset | Request log, drift (ML), endpoint breakdown |
| Dashboard | Sidebar → Dashboard | KPIs, shortcuts to Lake and Usage |
| Lake catalog | Sidebar → Lake | Bronze/silver/gold inventory, ingest jobs |
| Platform status | Navigate to /health or sidebar footer health indicator | Backend service availability |
Account monitoring (/usage)
Open Usage from the sidebar. Two tabs:
Data Intelligence
Account-wide monitoring for every inference product:
- Daily volume — stacked bars by product (ML, Forecast, Loop, Read, Vision) or total line
- Channel mix — Heimdall UI (in-app tests) vs API (your integration)
- Response times — average latency per day
- Summary cards — total inferences, API requests, avg response time, peak day (7 / 30 / 90 days)
- Products panel — all-time totals with links to ML, Forecast, Loop, or Settings (Read/Vision)
- Top assets — models, forecasts, loops ranked by call volume
- Endpoint breakdown — which REST routes received traffic
Legacy URL aliases: ?tab=intelligence, ?tab=monitoring, ?tab=overview open this tab.
Data Warehouse
Lake and Lab activity:
- Lake inventory — dataset counts and storage by bronze, silver, gold
- Lab activity — notebook execution volume and recent work
Per-asset monitoring
After ML, Forecast, or Loop deploy, open the asset at /ml/{id}, /forecast/{id}, or /loop/{id}.
The Usage section (visible once an API key exists) includes:
| Metric | Description |
|---|---|
| Total inferences | Predictions or forecast points in the period |
| Total requests | HTTP calls recorded |
| Avg response time | Mean latency (ms) |
| Endpoints used | REST paths that received traffic |
| Request log | Timestamp, endpoint, inference count, latency, channel, user agent |
| Drift % (ML only) | Performance drift when available |
Filter the log by endpoint and range (7 / 30 / 90 days). Expand rows for full metadata.
Read and Vision have no per-asset page — volume appears on Usage → Data Intelligence only.
→ ML monitoring · Forecast monitoring · Loop monitoring
Platform status (/health)
Platform status polls backend health and shows operational vs unavailable for auth, predict, ingestion, loop, read, vision, and related services.
- Open directly:
/health - Quick check: green indicator in the sidebar footer (expands to full panel on the health page)
Use during incidents or before launch windows. This is customer-facing availability — not a replacement for your app logs.
Recommended workflow
- Deploy with an API key
- Integrate — send production traffic (API integration)
- Monitor trends — Usage → Data Intelligence
- Debug one model — asset page → Usage → request log
- Check platform —
/healthif failures span multiple products
Common mistakes
| Mistake | Reality |
|---|---|
| Expecting Read/Vision per-request detail | Only account-level charts |
| Zero usage after deploy | In-app tests count — look for Heimdall UI channel |
| Charts empty on day one | Normal until first API or UI inference |
Next steps
- API integration
- Best practices
- User journeys — Lake → gold → train → monitor