Changelog

Two bugs caused the app to load everything at startup even when disabled:

1. PluginRegistry initialized ALL discovered plugins (vault unlock, host

creation, InitializeAsync with SDK queries) before checking disabled

state in Phase 3. Now the disabled check runs before Phase 1 — disabled

plugins only get lightweight schema registration (needed for backlinks

and sync data integrity) and skip vault unlock, host creation, and

InitializeAsync entirely.

2. RagIndexService unconditionally loaded the ONNX embedding model and ran

a full RAG index 5 seconds after startup, regardless of AiEnabled

setting. Now gated at two levels: App.axaml.cs only resolves the service

when AiEnabled is true, and the service's auto-init task checks the

setting as defense-in-depth.

With everything disabled: plugins go from 16 initialized to only core

plugins, and the ~800MB embedding model stays unloaded.

Root cause: DiscoverAndInitialize() (sync) parallelized plugin init

via Task.WhenAll but blocked the calling thread pool thread with

.GetAwaiter().GetResult(). All call sites wrapped this in Task.Run(),

so one thread pool thread was blocked waiting for ~15 async plugin

init tasks that themselves needed thread pool threads to run their

continuations. This starved the pool (default min = CPU count ≈ 8-10),

causing unrelated HttpClient calls (update check, plugin registry) to

time out at 30s despite the server responding in <400ms.

Fix: Switch all call sites from Task.Run(() => sync DiscoverAndInitialize)

to the properly async DiscoverAndInitializeAsync(). Also added the

parallel Task.WhenAll init pattern to the async version (it previously

initialized plugins sequentially) so startup time is not regressed.

The sync DiscoverAndInitialize() is retained for Reinitialize() (called

from the UI thread during workspace switches) but ReinitializeAsync()

now also uses the async path.

Add a DispatcherTimer that refreshes memory usage (WorkingSet64,

GC heap), data storage totals (DuckDB file sizes), and detailed

storage breakdown every 5 seconds while the Dashboard tab is active.

The timer starts on OnNavigatedToAsync and stops on OnNavigatedFrom,

so it only polls while the Dashboard is actually visible. Uses the

existing SystemMetricsService for memory and GetDataMetricsAsync for

storage — both are lightweight local-only operations.

Bump Dashboard plugin version 1.3.0 → 1.4.0.

When the user toggles AI off in settings, the ONNX embedding model

(InferenceSession + tokenizer) and local LLM weights (LLamaWeights +

LLamaContext) were staying resident in memory because nothing called

Dispose or released the native resources. This caused ~1.7GB+ of

memory to persist even after AI was disabled.

Changes:

• Add EmbeddingService.UnloadAsync() — disposes the ONNX session and

tokenizer without permanently disposing the service, so it can be

re-initialized later via InitializeAsync()

• Add LocalLlamaProvider.UnloadModelAsync() — same pattern for the

LLama weights and context

• Add IEmbeddingService.UnloadAsync() to the interface contract

(+ HeadlessEmbeddingService no-op stub)

• Wire OnAiEnabledChanged(false) to call UnloadAiModelsAsync() which

releases both models and nudges GC to reclaim native buffers

• Add RagIndexService + EmbeddingService disposal to

MainWindowViewModel.Cleanup() for clean app shutdown