Skip to content

v0.12.9 — Sub-Agent Model Mapping & Settings Cache

Choose a tag to compare

@xcl1989 xcl1989 released this 18 Jul 02:11

[0.12.9] — Sub-Agent Model Mapping & Settings Cache

Added

  • Sub-agent model mapping — New cost-control mechanism in Settings → General: when a parent agent using model M (from provider P) delegates a sub-task, the sub-agent automatically switches to a mapped model M' (on provider P'), possibly cross-provider. Stored as {"P|M": "P'|M'"} JSON in app_settings. Resolution priority: map hit → sub-agent profile default → parent model fallback.
  • Sub-Agent Model Mapping editor — New SubAgentModelMapEditor component with two ModelSelector dropdowns per row (parent → child) plus add/remove controls.

Fixed

  • Settings page reload flash on tab switchSettingsPage is conditionally rendered, so every tab switch re-fetched /settings, /providers, and one /providers/:name/models per provider. Added a module-level useSettingsData() cache that renders instantly on re-mount and only refetches on explicit provider changes or after a save.
  • Charts/KPI stay in pending state during streaming — Moved the isStreaming guard into the catch block of ChartBlock/KpiBlock so a visualization renders as soon as its JSON payload parses, instead of waiting for the entire assistant response to finish.

Changed

  • Delegation uses resolved model — Sub-agent event reporting and lesson reflection now use the sub-agent's actual resolved model/provider rather than the parent's.