Add dual-model support with separate chat and agent models

Split the single model setting into `chatModel` and `agentModel` to allow
using different LLMs for conversational modes (Ask, Research) versus
agentic modes (Edit, Organize, Workflow, auto-organizer). Defaults are
`deepseek-v4-flash` for chat and `glm-5.1` for agents.

Includes backward compatibility migration from legacy `model` field,
updated settings UI, per-mode tool filtering via new `agent-modes.ts`
configs, and vault search scoring improvements (exact phrase, recency,
filename bonuses).

src/chat-view.ts

@@ -67,12 +67,11 @@ export class ChatView extends ItemView {
     this.listenersAttached = false;
     this.settings = settings;
     this.currentAgentMode = settings.agentMode ?? 'ask';
-    this.ollamaClient = new OllamaClient(
-      settings.ollamaUrl,
-      settings.model,
-      undefined,
-      settings.cacheConfig
-    );
+    this.ollamaClient = this.createOllamaClient(settings.chatModel ?? settings.model, settings);
+    this.agentOllamaClient =
+      (settings.agentModel ?? settings.model) === (settings.chatModel ?? settings.model)
+        ? this.ollamaClient
+        : this.createOllamaClient(settings.agentModel ?? settings.model, settings);
     this.vaultIndexer = new VaultIndexer(this.app.vault, undefined, vectorStore);
     this.toolExecutor = new ToolExecutor(this.app.vault, this.app, telemetryManager);
     this.actionPreviewBuilder = new ActionPreviewBuilder(this.app.vault, this.app);
@@ -84,7 +83,7 @@ export class ChatView extends ItemView {
       this.app.vault,
       this.app,
       settings.ollamaUrl,
-      settings.model,
+      settings.agentModel ?? settings.model,
       { cacheConfig: settings.cacheConfig }
     );
   }
@@ -95,20 +94,22 @@ export class ChatView extends ItemView {
     if (this.modeSelectorEl) {
       this.modeSelectorEl.value = this.currentAgentMode;
     }
-    this.ollamaClient = new OllamaClient(
-      newSettings.ollamaUrl,
-      newSettings.model,
-      undefined,
-      newSettings.cacheConfig
+    this.ollamaClient = this.createOllamaClient(
+      newSettings.chatModel ?? newSettings.model,
+      newSettings
     );
+    this.agentOllamaClient =
+      (newSettings.agentModel ?? newSettings.model) === (newSettings.chatModel ?? newSettings.model)
+        ? this.ollamaClient
+        : this.createOllamaClient(newSettings.agentModel ?? newSettings.model, newSettings);
     this.workflowEngine = new WorkflowEngine(
       this.app.vault,
       this.app,
       newSettings.ollamaUrl,
-      newSettings.model,
+      newSettings.agentModel ?? newSettings.model,
       { cacheConfig: newSettings.cacheConfig }
     );
-    void this.ollamaClient.initializeCache().catch(() => {
+    void this.initializeClientCaches().catch(() => {
       new Notice(
         'Semantic cache unavailable: could not connect to ChromaDB. Check the ChromaDB URL in settings.'
       );
@@ -122,6 +123,9 @@ export class ChatView extends ItemView {
 
   async clearCache(): Promise<void> {
     await this.ollamaClient.clearCache();
+    if (this.agentOllamaClient !== this.ollamaClient) {
+      await this.agentOllamaClient.clearCache();
+    }
   }
 
   getViewType(): string {
@@ -138,7 +142,7 @@ export class ChatView extends ItemView {
 
   async onOpen(): Promise<void> {
     try {
-      await this.ollamaClient.initializeCache();
+      await this.initializeClientCaches();
     } catch {
       new Notice(
         'Semantic cache unavailable: could not connect to ChromaDB. Check the ChromaDB URL in settings.'
@@ -744,7 +748,7 @@ export class ChatView extends ItemView {
     if (followUpMessages.length > 0) {
       const finalMessages = [...messages, followUp, ...followUpMessages];
       const followUpStartTime = Date.now();
-      const response = await this.ollamaClient.chat(finalMessages, tools);
+      const response = await this.getActiveOllamaClient().chat(finalMessages, tools);
       const followUpDurationMs = Date.now() - followUpStartTime;
       const finalResponse = response.content || fullResponse;
       this.updateMessageById(assistantMessageId, {
@@ -755,7 +759,7 @@ export class ChatView extends ItemView {
 
       // Record follow-up LLM call telemetry
       this.telemetryManager?.recordLlmCall({
-        model: this.settings.model,
+        model: this.getActiveModel(),
         promptTokens: Math.round(finalMessages.reduce((sum, m) => sum + m.content.length, 0) / 4),
         completionTokens: Math.round(finalResponse.length / 4),
         totalTokens: Math.round(
@@ -812,7 +816,7 @@ export class ChatView extends ItemView {
     if (followUpMessages.length > 0) {
       const finalMessages = [...messages, followUp, ...followUpMessages];
       const followUpStartTime = Date.now();
-      const response = await this.ollamaClient.chat(finalMessages, tools);
+      const response = await this.getActiveOllamaClient().chat(finalMessages, tools);
       const followUpDurationMs = Date.now() - followUpStartTime;
       const finalResponse = response.content || 'Actions applied successfully.';
       this.updateMessageById(assistantMessageId, {
@@ -823,7 +827,7 @@ export class ChatView extends ItemView {
 
       // Record follow-up LLM call telemetry
       this.telemetryManager?.recordLlmCall({
-        model: this.settings.model,
+        model: this.getActiveModel(),
         promptTokens: Math.round(finalMessages.reduce((sum, m) => sum + m.content.length, 0) / 4),
         completionTokens: Math.round(finalResponse.length / 4),
         totalTokens: Math.round(
@@ -1090,7 +1094,9 @@ export class ChatView extends ItemView {
       // Prepend structured memory as a system message if available
       const messagesWithMemory = this.buildMessagesWithMemory(completeMessages);
 
-      const stream = this.ollamaClient.streamChat(messagesWithMemory, tools);
+      const activeClient = this.getActiveOllamaClient();
+      const activeModel = this.getActiveModel();
+      const stream = activeClient.streamChat(messagesWithMemory, tools);
 
       let fullResponse = '';
       let toolCalls: OllamaToolCall[] = [];
@@ -1131,7 +1137,7 @@ export class ChatView extends ItemView {
         completionTokens > 0 ? completionTokens : fullResponse.length / 4;
 
       this.telemetryManager?.recordLlmCall({
-        model: this.settings.model,
+        model: activeModel,
         promptTokens: Math.round(estimatedPromptTokens),
         completionTokens: Math.round(estimatedCompletionTokens),
         totalTokens: Math.round(estimatedPromptTokens + estimatedCompletionTokens),
@@ -1232,6 +1238,7 @@ export class ChatView extends ItemView {
   private listenersAttached: boolean = false;
   private settings: PluginSettings;
   private ollamaClient: OllamaClient;
+  private agentOllamaClient: OllamaClient;
   private vaultIndexer: VaultIndexer;
   private toolExecutor: ToolExecutor;
   private actionPreviewBuilder: ActionPreviewBuilder;
@@ -1253,6 +1260,31 @@ export class ChatView extends ItemView {
     tools: OllamaTool[];
     assistantMessageId: string;
   } | null = null;
+
+  private createOllamaClient(model: string, settings: PluginSettings): OllamaClient {
+    return new OllamaClient(settings.ollamaUrl, model, undefined, settings.cacheConfig);
+  }
+
+  private async initializeClientCaches(): Promise<void> {
+    await this.ollamaClient.initializeCache();
+    if (this.agentOllamaClient !== this.ollamaClient) {
+      await this.agentOllamaClient.initializeCache();
+    }
+  }
+
+  private getActiveModel(): string {
+    return this.isAgenticMode(this.currentAgentMode)
+      ? (this.settings.agentModel ?? this.settings.model)
+      : (this.settings.chatModel ?? this.settings.model);
+  }
+
+  private getActiveOllamaClient(): OllamaClient {
+    return this.isAgenticMode(this.currentAgentMode) ? this.agentOllamaClient : this.ollamaClient;
+  }
+
+  private isAgenticMode(mode: AgentMode): boolean {
+    return mode === 'edit' || mode === 'organize' || mode === 'workflow';
+  }
 }
 
 const MAX_TOOL_CALLS = 5;