AI root cause analysis · read-only · your model
Every incident is a chain. Intellira reads it backwards.
Commit → build → deploy → failing pod. Intellira correlates Bitbucket, Jenkins, ArgoCD and Kubernetes into one evidence trail and names the change that broke production — in about a minute.
- ✓ Never changes your running infrastructure
- ✓ Every claim cites evidence
- ✓ ~10-min setup
Why is payments-gateway in CrashLoopBackOff?
Reads your stackKubernetesArgoCDJenkinsBitbucketPrometheusJiraResponds inSlackTeamsPagerDuty
How it works
Most tools stop at the symptom. Intellira walks back to the cause.
13:44:26
Bitbucket — the change
Commit a1f9c2e raises the in-process cache to 2 GB. The memory limit is not touched.
13:51:03
Jenkins — the build
Build #4821 succeeds and pushes image 2.14.0. Nothing looks wrong yet.
13:58:47
ArgoCD — the deploy
Auto-sync rolls the image out. Health flips Synced → Degraded within a minute.
14:02:11
Kubernetes — the symptom
payments-gateway OOMKilled ×7. This is where every other tool starts — and stops.
Beyond the answer
The RCA is the start. Intellira closes the loop.
GUIDED REMEDIATION
Every RCA ends with the fix
Not "investigate further" — the file, the line, the value: a Quick Fix plus ordered next actions, matched against your runbooks. Intellira never touches your running infrastructure; the most it will do is open a draft pull request in your repo, and only if you turn that on.
deploy/payments-gateway.yaml:31 → limits.memory: 2Gi
INVESTIGATION MEMORY
A knowledge base that investigates with you
Runbooks, postmortems and past RCAs live per tenant and feed every new investigation. The second OOMKill looks like the first, Intellira says so — and cites your own postmortem.
"Matches postmortem: checkout OOMKill, May 12"
INCIDENT RESPONSE
Respond per incident, not per alert
Alertmanager and Grafana alerts are correlated and deduped into one incident — and can auto-trigger the RCA before anyone is paged. Acknowledge, investigate and act from the incident timeline or straight from Slack, every step kept in the evidence chain for the post-mortem.
17 alerts → 1 incident · RCA auto-started · resolved in 19m
VISIBILITY
The whole estate, one pane
Live service maps from cluster and GitOps state, incident timelines, and DevOps Health scans that surface misconfigurations before they page anyone.
160 services · 221 dependencies mapped
The actual product
This is what your on-call sees
The Investigate workspace: ask about a real incident, watch the evidence gather, get a cited root cause. Live on your own stack about ten minutes after the read-only connector is running.
demo environment · fictional data
⌕∿2△1⬡▤⊹⌂⚙#662 – payments-gateway is in Cra…
just now
#661 – checkout-api OutOfSync af…
2h ago
#659 – The pipeline failed becau…
6h ago
Quick fix
Set resources.limits.memory: 2Gi in deploy/payments-gateway.yaml:31
Root cause
Commit a1f9c2e raised the in-process cache to 2 GB; the pod memory limit stayed at 1Gi — every warm-up ends in OOMKilled.
working set 1.4Gi > limit 1Gi
Synced → Degraded 13:58:47
build #4821 → image 2.14.0
a1f9c2e cache 512Mi → 2GiNext actions
1. Set resources.limits.memory: 2Gi and redeploy
2. Add a working-set alert on cache-heavy services
Service topology
See the blast radius, then click into it
Inside the product
Beyond the investigation
The RCA chat is where incidents end. These are the pages your team lives in between them. All frames show fabricated demo data; the layouts are the real product's.
Command Center
Your system at a glance
3
Active
2
Agents
1.8h
MTTR
42s
RCA P95
0.8%
RCA Errors
Config Intelligence
AI-analyzed configuration findings across your integrations
Showing 3 of 114 · Load more
Customer Outcomes
The numbers you take into the QBR
MTTR trend
1.8h ▾ 57%
4.2h → 1.8h over 6 weeks
Savings rollup
96h
est. manual investigation time saved · 214 investigations this quarter
Cost per investigation
$0.55 avg
p95 $1.82 · $118 total LLM spend · your own model
Plan usage
214 / 500
investigations included · no surprise overage
Per-seat pricing. No per-host bill shock.
Full pricing →Free
$0
Judge the evidence quality yourself — 20 investigations a month, one environment.
Team
Popularfrom $20
/ user / mo · billed annually
500 investigations, service maps and ChatOps for the whole on-call rotation.
Business
from $36
/ user / mo · billed annually
SSO/SAML, RBAC and longer retention for platform teams at scale.
Enterprise
Custom
Self-hosted, unlimited investigations, dedicated support.
Built for a security review
READ-ONLY
No write scopes requested, ever. Intellira cannot scale, delete or apply — enforced at the connector layer, not by policy.
EVIDENCE CHAIN
Every conclusion cites the log line, event or commit it came from. Auditable end to end, exportable to your post-mortem.
BRING YOUR OWN MODEL
Run analysis on your own Claude, OpenAI or Gemini model — or one you host. Your data path stays under your control.
SECRET REDACTION
Tokens, keys and credentials are stripped from every model input and every rendered output.
TENANT ISOLATION
Isolation enforced at the data layer, not the UI. Per-tenant encryption keys on Business and above.
OUTBOUND-ONLY CONNECTOR
One read-only connector agent in your network, outbound connections only — no inbound firewall holes, nothing else deployed. Speaks the open Model Context Protocol, so new tools are additive.
Questions, answered
What is AI root cause analysis?
Is it safe to connect to production?
Which tools does it connect to?
Can I use my own model?
Stop tool-hopping. Read the root cause.
Connect read-only, bring your own model, and get an evidence-backed RCA on your next incident.