Heimdall
See everything, everywhere, all the time.
AI-native observability for modern systems. Watches every signal across every realm of your stack, hears anomalies in the noise, and explains the cause before you ask.
No credit card. Self-hosted available. Cancel any time.
PROVEN ON THE WORKLOADS THAT CAN’T GO DOWN
FEATURES
One watcher, every realm.
Most observability tools surface raw data and leave the humans to find the story. Heimdall is built around an LLM that investigates anomalies the moment they fire, stitches signals from every source, and presents one explanation per incident.
Anomaly-only alerts
No static thresholds, no rule sprawl. Statistical anomaly detection on every RED metric every 30 s; Claude investigates each one over MCP and produces one grouped incident, not ten symptom alerts.
Cross-source correlation
Spans, logs, host metrics, k8s events, federated Splunk, and your own custom sources reduced to one timeline. AI groups signals from different layers when they describe the same failure.
AI-generated dashboards
Type what you want to see. Heimdall picks the panels, queries the right services, mixes native + federated data, and ships a working dashboard you can save, share, and group.
Live service flow
A particle-stream view of every request through your business apps as it happens. Bottlenecks light up. Failures pulse. The diagram is the dashboard.
Cloud cost visibility
Per-resource AWS / GCP / Azure spend in the same surface as the telemetry. See what the bill actually pays for, not just the total at month end.
Built on open standards
OpenTelemetry end-to-end. ClickHouse for storage. MCP for the AI tooling. No proprietary agents, no lock-in, no consultancy required to get started.
HOW IT WORKS
From signal to story in one hop.
-
1
Send your signals
Point any OTel collector at
otlp.heimdallsystems.ai. Traces, metrics, logs — all three streams, one endpoint. Splunk, Datadog, custom Prometheus exporters all fan in via the standard exporters. -
2
Detect anomalies
The detector loop scans every service every 30 s for RED-metric outliers using an EWMA + z-score model. No rules to maintain, no thresholds to tune.
-
3
Investigate with AI
When an anomaly fires, Claude is given access to your entire telemetry estate via MCP and asked to investigate. It pulls traces, follows the topology, reads the logs, and writes a structured report.
-
4
One incident, not ten
Related signals get grouped automatically by business app + topology. You see one incident card with services involved, blast radius, and suggested actions — not a flood.
PRICING
Pricing that scales with you.
Start free. Pay only when the platform is mission-critical. Self-hosted available on every tier.
Free
$0/month
For evaluating + small teams.
- Up to 3 users
- 10 GB telemetry / month
- 14-day retention
- Anomaly detection + AI investigation
- Single-cloud cost view
- Community support
Pro
$0.30/GB ingested
For production teams.
- Unlimited users
- 30-day retention (configurable)
- Federated data sources (Splunk, etc.)
- Multi-cloud cost view
- SSO + SCIM provisioning
- Email support, 1-business-day SLA
Enterprise
Custom
For regulated & high-throughput estates.
- Self-hosted on your VPC
- Custom retention & throughput
- Dedicated MCP tool development
- Private LLM connector (Bedrock, Vertex)
- SOC 2 + ISO 27001 evidence
- Named TAM, 24×7 phone
RESOURCES
Built for builders.
DOCS
Quickstart
Point your OTel collector, see your first incident inside 5 minutes.
REFERENCE
MCP tool catalog
Every tool the AI uses to investigate. Bring your own.
GUIDE
Federate Splunk
Mix Heimdall’s native data with your existing SIEM in one dashboard.
BLOG
Why anomaly-only alerting works
The case against threshold rules, with actual numbers.
CHANGELOG
What shipped this week
Cloud Costs page, k8s metric flow, Alerts feed.
STATUS
Platform status
All systems normal. Subscribe for incident updates.
ABOUT
Built by operators, for operators.
The team behind Heimdall has spent careers buried in the wrong dashboard at 3 a.m. We built the watcher we always wanted: one place to look, one explanation per incident, no rule maintenance, and an AI that does the boring detective work.
Heimdall is named after the Norse god who watches all nine realms from the foot of the rainbow bridge. He sees everything, hears everything, sleeps less than a bird, and only blows the horn when something genuinely warrants your attention. That’s the product.
14k
spans / 5min
on the LSE simulator
30s
detection cycle
9
realms watched
1
incident, not ten
Stop chasing alerts.
Start watching like Heimdall.
14-day free trial. Live data in 5 minutes. No credit card.