PodPulse detects production incidents, explains the root cause, and proposes a fix — directly as a pull request.
demo / demo — connected to a real k3d cluster
| Incident | Signal | Confidence | Detected | PR |
|---|---|---|---|---|
|
crash-test DEFAULT
k3d-bayou-local-server-0
CRASHLOOPBACKOFF
|
QUEUED
Restarts 3
|
Apr 8, 2:11 PM CLUSTER 31CD39D1 |
NO PR | |
|
oom-test DEFAULT
k3d-bayou-local-server-0
OOMKILLED
|
PR OPENED
Restarts 1
|
82%
MODERATE CONFIDENCE
|
Apr 8, 1:26 PM CLUSTER 31CD39D1 |
↗ View PR |
Three steps. Zero kubectl required.
The in-cluster agent watches pod lifecycle events, container restarts, and Kubernetes events in real time — with zero cluster-write permissions.
Incident context is sent to the AI engine (powered by Claude). Root cause, confidence score, and proposed fix are returned as structured output.
When confidence exceeds the threshold, PodPulse opens a GitHub PR with detection, root cause hypothesis, and proposed fix. You merge. Nothing is applied without review.
PodPulse was built for SREs who need to trust the tools they run in their clusters.
The agent holds get, list, and watch permissions only. It never modifies any cluster resource — ever.
The GitHub App holds PR-creation permissions only. Every fix is a PR. You stay in control of what merges.
Kubernetes Secrets and environment variable values are never collected, never included in incident context, never transmitted.
The in-cluster component is open source. Audit the code, run it yourself, know exactly what runs inside your cluster.
PodPulse is invitation-only while we work with our first design partners. Tell us about your setup.
No spam. We'll reply within 48 hours.
We'll be in touch within 48 hours. In the meantime, check the live demo at app.podpulse.io — login: demo / demo.