Hephaisto demo Docs Site GitHub
These are recordings, not a live agent. Ten investigations the agent actually ran against a k3s cluster full of seeded faults, replayed and rendered as static pages. Nothing here is connected to anything: no cluster, no database, no model. Each incident's timeline says which cassette it came from, which model investigated it, and how it was graded — including the one that got it wrong.

Ten investigations

Every row below links to the full step trace: what the model asked for, what it was shown, the diagnosis it wrote, and every evidence excerpt resolving back to the untruncated tool output it came from.

graded correct
9 of 10
total cost
$0.05
investigating model
gpt-oss:120b
tool traces from
gemini-3.7-flash
statesevkindtargetincident stepscostgraded
! Escalated Critical CrashLoopBackOff hephaisto-chaos/c1-oomkill-6c8dd474dd-48sz5 CrashLoopBackOff on c1-oomkill (hephaisto-chaos) The container "balloon" is being OOMKilled during init because its memory limit (64Mi) is too low for the proc… 7 $0.0026 Correct
! Escalated Critical CrashLoopBackOff hephaisto-chaos/c2-crashloop-56b6585fd9-4hj9x CrashLoopBackOff on c2-crashloop (hephaisto-chaos) The container exits with code 1 because its startup script cannot connect to the MongoDB service (mongo.infra-… 5 $0.0019 Correct
! Escalated Warning Unschedulable hephaisto-chaos/c3-unschedulable-fd55d4b7c-tpkg7 Unschedulable on c3-unschedulable (hephaisto-chaos) The Deployment c3-unschedulable requests 500Gi of memory, which exceeds the node's allocatable memory, causing… 9 $0.0034 Correct
! Escalated Warning ImagePullBackOff hephaisto-chaos/c4-imagepull-7fd88fdb68-5n8sf ImagePullBackOff on c4-imagepull (hephaisto-chaos) Deployment hephaisto-chaos/c4-imagepull references a non-existent image tag busybox:this-tag-does-not-exist, c… 6 $0.0024 Correct
! Escalated Warning JobFailed hephaisto-chaos/c5-badjob JobFailed on c5-badjob (hephaisto-chaos) The Job's container exits with a non‑zero status (exit code 3) causing the pod to terminate with reason Error;… 22 $0.0095 Correct
! Escalated Warning ConfigError hephaisto-chaos/c7-configerror-7cd96c86f9-4dlvl ConfigError on c7-configerror-7cd96c86f9-4dlvl (hephaisto-chaos) The pod c7-configerror cannot start because the referenced Secret "c7-database-credentials" does not exist in … 5 $0.0016 Correct
! Escalated Warning ReadinessFlapping hephaisto-chaos/c8-readiness-flap-dd589d77c-6mxdb ReadinessFlapping on c8-readiness-flap (hephaisto-chaos) The container's /healthz endpoint intentionally alternates between HTTP 200 and 404 each minute, causing the c… 13 $0.0055 Correct
! Escalated Warning HighErrorRate /faulty-service HighErrorRate on faulty-service () No finding. 22 $0.0069 NoFinding
! Escalated Critical CrashLoopBackOff hephaisto-chaos/c11-transient-6f59fc7fcc-27h6g CrashLoopBackOff on c11-transient (hephaisto-chaos) The container exits with code 1 because its startup script detects a stale lock file (generation 1) on the att… 12 $0.0052 Correct
! Escalated Warning CrashLoopBackOff hephaisto-chaos/c12-stale-lease-5b894c6649-k22gg CrashLoopBackOff on c12-stale-lease-5b894c6649-k22gg (hephaisto-chaos) The container exits because its startup script detects a stale lease file on the persisted volume that claims … 12 $0.01 Correct

What this cannot show you

None of these ten shows the agent acting. Eight correctly declined to propose an action, one proposed nothing at all, and in two the grader judged an action was missed. That is a measured property of the model these were replayed against, not a limitation of this page — gpt-oss:120b proposed an action in 0 of 18 runs on a fixture where deepseek-v4-flash proposed one in 4 of 8.

The honest version of the claim is therefore: this is what the agent's diagnosis looks like. The evidence page has the denominators.