Offline runtime profile and diagnostics for local OpenCode in an isolated lab

[Spriithy]

I use the installed t3 command in a network-isolated lab, with OpenCode connected to local LLMs. Dependencies are available through an internal npm mirror. On my machine, t3 exits silently after boot.

A Docker check with T3 0.0.45, OpenCode 1.18.34, and --network none did not reproduce that exit: T3 started, served HTTP 200, and discovered the configured local model. Inference was not tested. The observed failures were repeated telemetry errors and OpenCode model-catalog refresh errors.

Could we have a supported t3 serve --offline profile that:

• Skips public telemetry, update checks, model-catalog and pricing refreshes, using bundled/cached data.
• Keeps loopback/LAN connections to local OpenCode and inference servers working.
• Persists across service restarts and reports actionable startup failures, perhaps through t3 doctor --offline.

Existing controls (T3CODE_TELEMETRY_ENABLED=false, enableProviderUpdateChecks=false, and OpenCode's OPENCODE_DISABLE_MODELS_FETCH=1) cover parts of this, but there is no unified profile; T3's pricing refresh is separate.

Related: #13455 covers offline SSH installation. This request concerns runtime behavior after installation. I'd be interested in a small PR if maintainers agree on the scope.