AI SRE

The AI SRE investigates the way a careful engineer does: pulls the events, reads the logs, checks the spec, walks the timeline, and says what it thinks happened — except it does the boring part in seconds, at 3am, without being paged. It runs on your keys against the provider you choose, and it's entirely optional: unconfigured, it's a page that does nothing and sends nothing.

Three rules that don't bend

• Read-only tools. The agent works through Moonway's own read-only toolset — logs, events, specs, metrics, the journal. There is no tool in its hands that can mutate a cluster.
• Every check on the record. Each tool call the agent makes is shown in the conversation as it happens — expand any answer and audit exactly what it looked at before believing it.
• It proposes, you run. When action is warranted the agent writes the kubectl command and hands it to you. The human stays between the idea and the cluster.

Starting an investigation

Ask anything on the AI SRE page — conversations stream, persist, and can be picked back up or deleted. The faster path: Explain on any Triage finding opens a fresh conversation seeded with the incident brief, so the agent starts already knowing what broke and where.

Providers

Configure once in Settings → AI assist; the credential is stored by the daemon on your machine and goes nowhere else.

The provider is the only outbound connection Moonway ever makes, and only while you're actually asking. A local model through the OpenAI-compatible route keeps even that on your machine — the air-gapped configuration is simply an endpoint that never leaves it.