[vishr]

Summary

Hide the remaining chat entry points when agent_available is false and explain how to enable chat or use an MCP client in Settings.

agentAvailable = false
  Rail: no chat, AI create, service investigation, or dashboard source-chat link
  Panels: no Explain or Fix, including fullscreen
  Home: default dashboard
  Settings: provider/key setup and MCP connection guide

Implementation choices:

• Reuse agentAvailable throughout; gate Explain/Fix callbacks in the grid as well as the existing controls in PanelCard.
• Hide dashboard source-chat links and service investigations because both open chat. Disable service catalogue requests for these investigations too.
• Preserve the existing home-to-dashboards routing for both modes. Extend the integration test through actual default-dashboard selection, with the default deliberately listed second.
• Keep the existing chat/create controls and enabled behavior. Do not add a non-AI dashboard creation flow, routes, backend changes, or another renderer.
• Use the existing Settings components and MCP guide. Verify the provider/key names against the configuration schema; no credential inputs or storage.
• Regenerate committed host assets as required by the embedded UI contract.

Evidence

• Before: new rail tests exposed chat-origin links and service investigations while chat was disabled; the new settings test failed because setup guidance was absent.
  After: disabled-state coverage passes for rail chat/create entries, source-chat links, service investigations, panel Explain/Fix in grid and fullscreen, default-dashboard navigation, and Settings. Enabled-state coverage remains green.
• Full UI suite: 101 files, 1,466 tests passed with bun run test --maxWorkers=2. The default-worker run had two timing failures in unchanged tests: the ECharts panel inspection exceeded 5 seconds and the 400-service layout took 57 ms against a 50 ms limit. No test thresholds or configuration changed.
• Passed: just test (all Go packages), just lint, just ui (TypeScript checks and production builds), just ui-check, just notices-check, just docs-generate-check, and just site-build.
• just check passed every preceding gate, then failed at the pre-existing ui-audit failure: two moderate advisories in unchanged seroval@1.6.8, GHSA-f86p-h4j2-7qgj and GHSA-9wx9-5pv9-3349. The standalone audit reproduces it. Dependencies were left unchanged; checks after the audit gate were run separately as listed above.
• The pre-push hook reproduced the same audit failure. Publication excluded only its already-run check job with LEFTHOOK_EXCLUDE=check.

Merge Danger

Door: two-way

Blast Radius: UI

Only conditional AI affordances, Settings guidance, tests, and generated host assets change. No schema, API, or dependency changes.

[vishr]

Folded into #410.