Fix the model config issue, OpenRouter Oauth issue and model refresh … (#45)

- New Features
  - Per-task model-size options (mini/nano) across agents, tools, and evaluations.
  - New LLM context support for tool executions and a new ObjectiveDrivenAction tool.
  - Automatic fetch of OpenRouter models after OAuth with UI update and event dispatch.
  - OpenRouter model caching with TTL, auto-refresh, and programmatic update hook.

- Changes
  - Many agent/tool flows now prefer or require mini/nano models; evaluation/UI accept explicit main/mini/nano.
  - Tool execute signatures extended to accept LLM context; reduced verbose tool logging.
  - Centralized model/provider sentinels and placeholders; stricter JSON parsing for extractors.

- Tests
  - Added unit tests covering OpenRouter model cache and refresh behavior.

Author: tysonthomas9

front_end/panels/ai_chat/BUILD.gn

@@ -337,6 +337,7 @@ ts_library("unittests") {
     "ui/__tests__/ChatViewAgentSessionsOrder.test.ts",
     "ui/__tests__/ChatViewSequentialSessionsTransition.test.ts",
     "ui/__tests__/ChatViewInputClear.test.ts",
+    "ui/__tests__/SettingsDialogOpenRouterCache.test.ts",
     "ui/input/__tests__/InputBarClear.test.ts",
     "ui/message/__tests__/MessageCombiner.test.ts",
     "ui/message/__tests__/StructuredResponseController.test.ts",

front_end/panels/ai_chat/agent_framework/AgentRunner.ts

@@ -18,6 +18,7 @@ import { sanitizeMessagesForModel } from '../LLM/MessageSanitizer.js';
 const logger = createLogger('AgentRunner');
 
 import { ConfigurableAgentTool, ToolRegistry, type ConfigurableAgentArgs, type ConfigurableAgentResult, type AgentRunTerminationReason, type HandoffConfig /* , HandoffContextTransform, ContextFilterRegistry*/ } from './ConfigurableAgentTool.js';
+import { MODEL_SENTINELS } from '../core/Constants.js';
 
 /**
  * Configuration for the AgentRunner
@@ -33,6 +34,10 @@ export interface AgentRunnerConfig {
   provider: LLMProvider;
   /** Optional vision capability check. Defaults to false (no vision). */
   getVisionCapability?: (modelName: string) => Promise<boolean> | boolean;
+  /** Mini model for smaller/faster operations */
+  miniModel?: string;
+  /** Nano model for smallest/fastest operations */
+  nanoModel?: string;
 }
 
 /**
@@ -218,6 +223,8 @@ export class AgentRunner {
     parentSession?: AgentSession, // For natural nesting
     defaultProvider?: LLMProvider,
     defaultGetVisionCapability?: (modelName: string) => Promise<boolean> | boolean,
+    miniModel?: string, // Mini model for smaller/faster operations
+    nanoModel?: string, // Nano model for smallest/fastest operations
     overrides?: { sessionId?: string; parentSessionId?: string; traceId?: string }
   ): Promise<ConfigurableAgentResult & { agentSession: AgentSession }> {
     const targetAgentName = handoffConfig.targetAgentName;
@@ -286,12 +293,28 @@ export class AgentRunner {
     // Enhance the target agent's system prompt with page context
     const enhancedSystemPrompt = await enhancePromptWithPageContext(targetConfig.systemPrompt);
 
+    // Resolve model name for the target agent
+    let resolvedModelName: string;
+    if (typeof targetConfig.modelName === 'function') {
+      resolvedModelName = targetConfig.modelName();
+    } else if (targetConfig.modelName === MODEL_SENTINELS.USE_MINI) {
+      if (!miniModel) {
+        throw new Error(`Mini model not provided for handoff to agent '${targetAgentName}'. Ensure miniModel is passed in context.`);
+      }
+      resolvedModelName = miniModel;
+    } else if (targetConfig.modelName === MODEL_SENTINELS.USE_NANO) {
+      if (!nanoModel) {
+        throw new Error(`Nano model not provided for handoff to agent '${targetAgentName}'. Ensure nanoModel is passed in context.`);
+      }
+      resolvedModelName = nanoModel;
+    } else {
+      resolvedModelName = targetConfig.modelName || defaultModelName;
+    }
+
     // Construct Runner Config & Hooks for the target agent
     const targetRunnerConfig: AgentRunnerConfig = {
       apiKey,
-      modelName: typeof targetConfig.modelName === 'function'
-        ? targetConfig.modelName()
-        : (targetConfig.modelName || defaultModelName),
+      modelName: resolvedModelName,
       systemPrompt: enhancedSystemPrompt,
       tools: targetConfig.tools
               .map(toolName => ToolRegistry.getRegisteredTool(toolName))
@@ -300,6 +323,8 @@ export class AgentRunner {
       temperature: targetConfig.temperature ?? defaultTemperature,
       provider: defaultProvider as LLMProvider,
       getVisionCapability: defaultGetVisionCapability,
+      miniModel,
+      nanoModel,
     };
     const targetRunnerHooks: AgentRunnerHooks = {
       prepareInitialMessages: undefined, // History already formed by transform or passthrough
@@ -845,6 +870,8 @@ export class AgentRunner {
                   currentSession, // Pass current session for natural nesting
                   config.provider,
                   config.getVisionCapability,
+                  config.miniModel,
+                  config.nanoModel,
                   { sessionId: nestedSessionId, parentSessionId: currentSession.sessionId, traceId: getCurrentTracingContext()?.traceId }
               );
 
@@ -947,11 +974,13 @@ export class AgentRunner {
             }
 
             try {
-              logger.info(`${agentName} Executing tool: ${toolToExecute.name} with args:`, toolArgs);
+              logger.info(`${agentName} Executing tool: ${toolToExecute.name}`);
               const execTracingContext = getCurrentTracingContext();
               toolResultData = await toolToExecute.execute(toolArgs as any, ({
                 provider: config.provider,
                 model: modelName,
+                miniModel: config.miniModel,
+                nanoModel: config.nanoModel,
                 getVisionCapability: config.getVisionCapability,
                 overrideSessionId: preallocatedChildId,
                 overrideParentSessionId: currentSession.sessionId,
@@ -1210,7 +1239,9 @@ export class AgentRunner {
                 undefined, // No llmToolArgs for max iterations handoff
                 currentSession, // Pass current session for natural nesting
                 config.provider,
-                config.getVisionCapability
+                config.getVisionCapability,
+                config.miniModel,
+                config.nanoModel
             );
             // Extract the result and session
             const { agentSession: childSession, ...actualResult } = handoffResult;

front_end/panels/ai_chat/agent_framework/ConfigurableAgentTool.ts

@@ -4,16 +4,30 @@
 
 import { AgentService } from '../core/AgentService.js';
 import type { Tool } from '../tools/Tools.js';
-import { AIChatPanel } from '../ui/AIChatPanel.js';
 import { ChatMessageEntity, type ChatMessage } from '../models/ChatTypes.js';
 import { createLogger } from '../core/Logger.js';
 import { getCurrentTracingContext } from '../tracing/TracingConfig.js';
+import { MODEL_SENTINELS } from '../core/Constants.js';
 import type { AgentSession } from './AgentSessionTypes.js';
+import type { LLMProvider } from '../LLM/LLMTypes.js';
 
 const logger = createLogger('ConfigurableAgentTool');
 
 import { AgentRunner, type AgentRunnerConfig, type AgentRunnerHooks } from './AgentRunner.js';
 
+// Context passed along with agent/tool calls
+export interface CallCtx {
+  provider?: LLMProvider,
+  model?: string,
+  miniModel?: string,
+  nanoModel?: string,
+  mainModel?: string,
+  getVisionCapability?: (modelName: string) => Promise<boolean> | boolean,
+  overrideSessionId?: string,
+  overrideParentSessionId?: string,
+  overrideTraceId?: string,
+}
+
 /**
  * Defines the possible reasons an agent run might terminate.
  */
@@ -412,27 +426,62 @@ export class ConfigurableAgentTool implements Tool<ConfigurableAgentArgs, Config
 
     // Initialize
     const maxIterations = this.config.maxIterations || 10;
-    const modelName = typeof this.config.modelName === 'function'
-      ? this.config.modelName()
-      : (this.config.modelName || AIChatPanel.instance().getSelectedModel());
+    
+    // Parse execution context first
+    const callCtx = (_ctx || {}) as CallCtx;
+    
+    // Resolve model name from context or configuration
+    let modelName: string;
+    if (this.config.modelName === MODEL_SENTINELS.USE_MINI) {
+      if (!callCtx.miniModel) {
+        throw new Error(`Mini model not provided in context for agent '${this.name}'. Ensure context includes miniModel.`);
+      }
+      modelName = callCtx.miniModel;
+    } else if (this.config.modelName === MODEL_SENTINELS.USE_NANO) {
+      if (!callCtx.nanoModel) {
+        throw new Error(`Nano model not provided in context for agent '${this.name}'. Ensure context includes nanoModel.`);
+      }
+      modelName = callCtx.nanoModel;
+    } else if (typeof this.config.modelName === 'function') {
+      modelName = this.config.modelName();
+    } else if (this.config.modelName) {
+      modelName = this.config.modelName;
+    } else {
+      // Use main model from context, or fallback to context model
+      const contextModel = callCtx.mainModel || callCtx.model;
+      if (!contextModel) {
+        throw new Error(`No model provided for agent '${this.name}'. Ensure context includes model or mainModel.`);
+      }
+      modelName = contextModel;
+    }
+    
+    // Override with context model only if agent doesn't have its own model configuration
+    if (callCtx.model && !this.config.modelName) {
+      modelName = callCtx.model;
+    }
+    
+    // Validate required context
+    if (!callCtx.provider) {
+      throw new Error(`Provider not provided in context for agent '${this.name}'. Ensure context includes provider.`);
+    }
+    
     const temperature = this.config.temperature ?? 0;
-
     const systemPrompt = this.config.systemPrompt;
     const tools = this.getToolInstances();
-
+    
     // Prepare initial messages
     const internalMessages = this.prepareInitialMessages(args);
-
-    // Prepare runner config and hooks
     const runnerConfig: AgentRunnerConfig = {
       apiKey,
       modelName,
       systemPrompt,
       tools,
       maxIterations,
       temperature,
-      provider: AIChatPanel.getProviderForModel(modelName),
-      getVisionCapability: (m: string) => AIChatPanel.isVisionCapable(m),
+      provider: callCtx.provider,
+      getVisionCapability: callCtx.getVisionCapability ?? (() => false),
+      miniModel: callCtx.miniModel,
+      nanoModel: callCtx.nanoModel,
     };
 
     const runnerHooks: AgentRunnerHooks = {
@@ -446,7 +495,7 @@ export class ConfigurableAgentTool implements Tool<ConfigurableAgentArgs, Config
     };
 
     // Run the agent
-    const ctx: any = _ctx || {};
+    const ctx: any = callCtx || {};
     const result = await AgentRunner.run(
       internalMessages,
       args,

front_end/panels/ai_chat/agent_framework/implementation/ConfiguredAgents.ts

@@ -10,13 +10,13 @@ import { BookmarkStoreTool } from '../../tools/BookmarkStoreTool.js';
 import { DocumentSearchTool } from '../../tools/DocumentSearchTool.js';
 import { NavigateURLTool, PerformActionTool, GetAccessibilityTreeTool, SearchContentTool, NavigateBackTool, NodeIDsToURLsTool, TakeScreenshotTool, ScrollPageTool } from '../../tools/Tools.js';
 import { HTMLToMarkdownTool } from '../../tools/HTMLToMarkdownTool.js';
-import { AIChatPanel } from '../../ui/AIChatPanel.js';
 import { ChatMessageEntity, type ChatMessage } from '../../models/ChatTypes.js';
 import {
   ConfigurableAgentTool,
   ToolRegistry, type AgentToolConfig, type ConfigurableAgentArgs
 } from '../ConfigurableAgentTool.js';
 import { WaitTool } from '../../tools/Tools.js';
+import { MODEL_SENTINELS } from '../../core/Constants.js';
 import { ThinkingTool } from '../../tools/ThinkingTool.js';
 import type { Tool } from '../../tools/Tools.js';
 
@@ -70,7 +70,6 @@ If the page does not match the expected content, retry with a different URL patt
 Remember: Always use navigate_url to actually go to the constructed URLs. Return easy-to-read markdown reports.`,
     tools: ['navigate_url', 'get_page_content'],
     maxIterations: 5,
-    modelName: () => AIChatPanel.instance().getSelectedModel(),
     temperature: 0.1,
     schema: {
       type: 'object',
@@ -322,7 +321,7 @@ Remember: You gather data, content_writer_agent writes the report. Always hand o
       'document_search'
     ],
     maxIterations: 15,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0,
     schema: {
       type: 'object',
@@ -423,7 +422,7 @@ Your process should follow these steps:
 The final output should be in markdown format, and it should be lengthy and detailed. Aim for 5-10 pages of content, at least 1000 words.`,
     tools: [],
     maxIterations: 3,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.3,
     schema: {
       type: 'object',
@@ -531,7 +530,7 @@ Conclusion: Fix the args format and retry with proper syntax: { "method": "fill"
       'take_screenshot',
     ],
     maxIterations: 10,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.5,
     schema: {
       type: 'object',
@@ -640,7 +639,7 @@ Remember that verification is time-sensitive - the page state might change durin
       'take_screenshot'
     ],
     maxIterations: 3,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.2,
     schema: {
       type: 'object',
@@ -725,7 +724,7 @@ When selecting an element to click, prioritize:
       'node_ids_to_urls',
     ],
     maxIterations: 5,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.7,
     schema: {
       type: 'object',
@@ -805,7 +804,7 @@ When selecting a form field to fill, prioritize:
       'schema_based_extractor',
     ],
     maxIterations: 5,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.7,
     schema: {
       type: 'object',
@@ -881,7 +880,7 @@ When selecting an element for keyboard input, prioritize:
       'schema_based_extractor',
     ],
     maxIterations: 5,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.7,
     schema: {
       type: 'object',
@@ -966,7 +965,7 @@ When selecting an element to hover over, prioritize:
       'schema_based_extractor',
     ],
     maxIterations: 5,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.7,
     schema: {
       type: 'object',
@@ -1048,7 +1047,7 @@ The accessibility tree includes information about scrollable containers. Look fo
       'schema_based_extractor',
     ],
     maxIterations: 5,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.7,
     schema: {
       type: 'object',
@@ -1283,7 +1282,6 @@ Remember: **Plan adaptively, execute systematically, validate continuously, and
       'thinking',
     ],
     maxIterations: 15,
-    modelName: () => AIChatPanel.instance().getSelectedModel(),
     temperature: 0.3,
     schema: {
       type: 'object',
@@ -1422,7 +1420,7 @@ Remember to adapt your analysis based on the product category - different attrib
       'get_page_content',
     ],
     maxIterations: 5,
-    modelName: () => AIChatPanel.getMiniModel(),
+    modelName: MODEL_SENTINELS.USE_MINI,
     temperature: 0.2,
     schema: {
       type: 'object',