Tuning advice and AI availability
Treat recommendations as review inputs and verify their effect on your own server.
Available local advice
The console wizard builds a deterministic plan from your profile, player estimate, hardware, and available metrics. It can suggest worker counts, snapshot limits, and cache sizing. Read the reasons and sample quality. Estimates are not observations of every world or plugin workload.
Advice does not change JVM heap or server.properties, and does not promise TPS. Saved changes go through console review, backup, and restart.
External AI status
Source still includes optional analysis components and feature-flags.external-ai-advisor, default false. The embedded UI that previously used them was removed. This repository does not establish a working analysis-to-apply flow in the current cloud service. Enabling the flag alone does not prove a button/service is available.
Public docs therefore omit retired routes, provider/model names, service keys, and unverified click instructions. Use local advice. If customer access offers analysis, follow availability information for that service release.
If analysis is offered to customers
Read the data-transfer explanation and give informed consent before sending. Analysis consent is separate from consent to save a configuration. Check applicable costs and limits; do not assume it is free.
Local component design limits context to planning inputs, aggregate hardware/performance, and permitted tuning changes. That does not describe all cloud-bridge traffic, which has a broader scope. Do not manually send full configuration, secrets, chat, or player data to an analysis service.
AI advice can be inaccurate. Compare suggestions against the configuration reference, test on staging, and measure before/after. When analysis is unavailable, console setup still works. Do not execute arbitrary generated commands or YAML without review.