Google Cloud Agentic AI x Elastic

Elastic On-Call Agent:
Agentic Ops with Google Cloud

Alert-triggered triage using Cloud Run, Elastic evidence, and GitHub Issues to move from signal to root cause to agent-executed remediation.

Agentic flow

Elastic MCP CHECKING Vertex AI Gemini 2.5 Pro CHECKING Cloud Run CHECKING GitHub Issues CHECKING

Elastic alert IDLE Cloud Run endpoint IDLE Evidence collection IDLE Gemini reasoning IDLE Root cause IDLE Remediation plan IDLE Apply fix IDLE Verify health IDLE GitHub issue IDLE

Demo controls

Incident status IDLE

Service checkout-api

Severity high

Root cause -

Confidence -

Live workload health

checkout-api CHECKING

Waiting for live checkout-api status.

HTTP: - | mode: -

Open service Health Checkout flow

payment-api CHECKING

Waiting for live payment-api status.

HTTP: -

Open service Health Status

customer checkout flow CHECKING

Waiting for live end-to-end checkout flow status.

HTTP: -

Open flow Checkout status

Agent resolved the incident

The agent repaired checkout-api runtime mode and verified the live customer checkout flow.

Verification

checkout-api /healthz returned 200 after remediation.
checkout-api /checkout returned 200 after remediation.
payment-api /healthz remained healthy.
Customer checkout flow returned to healthy.

GitHub issue created

[resolved by agent] checkout-api failure after redis_timeout runtime mode

incident agent-remediated checkout-api elastic-evidence

Waiting for remediation result.

Open GitHub issue

Issue body preview

Root cause: checkout-api entered redis_timeout failure mode.
Action taken: agent called checkout-api /admin/repair.
Verification: live workload checks returned healthy.
Follow-up: review checkout-api runtime config guardrails.

Incident output

Click Health check, then Simulate incident.

Ask Agent

Agent answer

Ask a question after simulating an incident.

AGENT REMEDIATION PLAN

Repair checkout-api runtime mode

The agent will call the real checkout-api /admin/repair endpoint, then verify /healthz, /checkout, and payment-api /healthz before closing the loop.

Why this is safe

payment-api is still healthy, so the blast radius is isolated to checkout-api.
The failure mode is runtime-level redis_timeout and can be repaired without redeploying.
The agent verifies the customer checkout flow before marking the incident solved.
GitHub issue creation keeps an engineering follow-up trail.

What the agent will do

Read live checkout-api and payment-api status.
Call checkout-api /admin/repair.
Verify checkout-api /healthz and /checkout.
Create a GitHub issue for post-incident review.

Elastic On-Call Agent: Agentic Ops with Google Cloud