Deployment
Real deployment close to the customer environment.
OpsDiag is positioned as an RCA system that works near the infrastructure, services, and operational knowledge responders already use.
Operator architecture
Customer environment to RCA output.
The deployment path is shown as an architecture route, not a generic process list.
- 01
Customer side
Customer environment
Incident context starts near the services, clusters, provider data, and runbooks involved in the response.
- 02
Install
Agentic RCA deployment
Agentic RCA is installed through Helm and configured around approved model providers and evidence paths.
- 03
Access route
MCP Server Proxy
RCA and CLI questions enter the proxy before reaching scoped MCP Servers for service or provider evidence.
- 04
Evidence
MCP Servers and provider APIs
MCP Servers query relevant systems such as Kubernetes, cloud APIs, observability, deploy history, edge controls, and alerts.
- 05
Output
RCA output
The result is a reviewable brief with likely causes, affected scope, validation checks, rollback candidates, and mitigation steps.
Data boundary
Close to the customer, limited in the output.
Stays close
Operational context
- Runbooks, topology notes, service ownership, and policy context.
- Deploy context, provider signals, and incident-specific assumptions.
- Scoped MCP Server evidence paths operated around the customer workflow.
RCA output
Decision context
- What changed and what is affected.
- Signals that support or weaken a proposed cause.
- The safest next validation, rollback, mitigation, or handoff step.
Install
Helm-first setup
The install path is intentionally shown as an operator command because the page is about a real deployment model.
Models
Provider-aware operation
The deployment can be positioned around Anthropic, Gemini, and OpenAI model providers without turning the page into model marketing.
Access
Scoped evidence paths
Agent access is described as scoped around the incident path and the MCP evidence sources the customer is prepared to operate.