@jupyterlite/ai 0.15.0 → 0.17.0

package/lib/tokens.d.ts

@@ -4,7 +4,7 @@ import { IRenderMimeRegistry } from '@jupyterlab/rendermime';
 import { Token } from '@lumino/coreutils';
 import type { IDisposable } from '@lumino/disposable';
 import { ISignal } from '@lumino/signaling';
-import type { Tool, LanguageModel } from 'ai';
+import type { Tool, LanguageModel, UserContent, ModelMessage } from 'ai';
 import { ISecretsManager } from 'jupyter-secrets-manager';
 import type { IModelOptions } from './providers/models';
 import { AIChatModel } from './chat-model';
@@ -17,6 +17,7 @@ export declare namespace CommandIds {
     const openSettings = "@jupyterlite/ai:open-settings";
     const reposition = "@jupyterlite/ai:reposition";
     const openChat = "@jupyterlite/ai:open-chat";
+    const openOrRevealChat = "@jupyterlite/ai:open-or-reveal-chat";
     const moveChat = "@jupyterlite/ai:move-chat";
     const refreshSkills = "@jupyterlite/ai:refresh-skills";
     const saveChat = "@jupyterlite/ai:save-chat";
@@ -153,6 +154,12 @@ export interface IProviderToolCapabilities {
 /**
  * Provider information
  */
+export interface IProviderModelInfo {
+    /**
+     * Default context window for the model in tokens.
+     */
+    contextWindow?: number;
+}
 export interface IProviderInfo {
     /**
      * Unique identifier for the provider
@@ -173,6 +180,10 @@ export interface IProviderInfo {
      * Default model names for this provider
      */
     defaultModels: string[];
+    /**
+     * Optional per-model metadata keyed by model ID.
+     */
+    modelInfo?: Record<string, IProviderModelInfo>;
     /**
      * Whether this provider supports custom base URLs
      */
@@ -246,6 +257,7 @@ export interface IProviderParameters {
     temperature?: number;
     maxOutputTokens?: number;
     maxTurns?: number;
+    contextWindow?: number;
     supportsFillInMiddle?: boolean;
     useFilterText?: boolean;
 }
@@ -282,6 +294,7 @@ export interface IAIConfig {
     toolsEnabled: boolean;
     sendWithShiftEnter: boolean;
     showTokenUsage: boolean;
+    showContextUsage: boolean;
     commandsRequiringApproval: string[];
     commandsAutoRenderMimeBundles: string[];
     trustedMimeTypesForAutoRender: string[];
@@ -453,7 +466,7 @@ export interface IAgentManager {
     /**
      * Clears conversation history and resets agent state.
      */
-    clearHistory(): void;
+    clearHistory(): Promise<void>;
     /**
      * Sets the conversation history with a list of messages from the chat.
      * @param messages The chat messages to set as history
@@ -480,7 +493,12 @@ export interface IAgentManager {
      * Handles the complete execution cycle including tool calls.
      * @param message The user message to respond to (may include processed attachment content)
      */
-    generateResponse(message: string): Promise<void>;
+    generateResponse(message: UserContent): Promise<void>;
+    /**
+     * Create a transient language model to request a text response, which won't be added to history.
+     * @param messages - the messages sequence to send to the model.
+     */
+    textResponse(messages: ModelMessage[]): Promise<string>;
     /**
      * Initializes the AI agent with current settings and tools.
      * Sets up the agent with model configuration, tools, and MCP tools.
@@ -549,6 +567,10 @@ export interface ICreateChatOptions {
      * Whether the chat is autosaved or not.
      */
     autosave?: boolean;
+    /**
+     * An optional title to the chat.
+     */
+    title?: string | null;
 }
 /**
  * Token for the chat model handler.
@@ -633,6 +655,15 @@ export interface ITokenUsage {
      * Number of output tokens generated (completion tokens)
      */
     outputTokens: number;
+    /**
+     * Estimated prompt tokens used by the most recent model request.
+     * This is based on the final step of the latest request.
+     */
+    lastRequestInputTokens?: number;
+    /**
+     * Configured context window size for the active provider/model.
+     */
+    contextWindow?: number;
 }
 /**
  * The string that replaces a secret key in settings.

package/lib/tokens.js

@@ -7,6 +7,7 @@ export var CommandIds;
     CommandIds.openSettings = '@jupyterlite/ai:open-settings';
     CommandIds.reposition = '@jupyterlite/ai:reposition';
     CommandIds.openChat = '@jupyterlite/ai:open-chat';
+    CommandIds.openOrRevealChat = '@jupyterlite/ai:open-or-reveal-chat';
     CommandIds.moveChat = '@jupyterlite/ai:move-chat';
     CommandIds.refreshSkills = '@jupyterlite/ai:refresh-skills';
     CommandIds.saveChat = '@jupyterlite/ai:save-chat';
@@ -15,15 +16,15 @@ export var CommandIds;
 /**
  * The tool registry token.
  */
-export const IToolRegistry = new Token('@jupyterlite/ai:tool-registry', 'Tool registry for AI agent functionality');
+export const IToolRegistry = new Token('@jupyterlite/ai:IToolRegistry', 'Tool registry for AI agent functionality');
 /**
  * The skill registry token.
  */
-export const ISkillRegistry = new Token('@jupyterlite/ai:skill-registry', 'Skill registry for AI agent functionality');
+export const ISkillRegistry = new Token('@jupyterlite/ai:ISkillRegistry', 'Skill registry for AI agent functionality');
 /**
  * Token for the provider registry.
  */
-export const IProviderRegistry = new Token('@jupyterlite/ai:provider-registry', 'Registry for AI providers');
+export const IProviderRegistry = new Token('@jupyterlite/ai:IProviderRegistry', 'Registry for AI providers');
 /**
  * Token for the AI settings model.
  */
@@ -31,19 +32,19 @@ export const IAISettingsModel = new Token('@jupyterlite/ai:IAISettingsModel');
 /**
  * Token for the agent manager.
  */
-export const IAgentManager = new Token('@jupyterlite/ai:agent-manager');
+export const IAgentManager = new Token('@jupyterlite/ai:IAgentManager');
 /*
  * Token for the agent manager factory.
  */
-export const IAgentManagerFactory = new Token('@jupyterlite/ai:agent-manager-factory');
+export const IAgentManagerFactory = new Token('@jupyterlite/ai:IAgentManagerFactory');
 /**
  * Token for the chat model handler.
  */
-export const IChatModelHandler = new Token('@jupyterlite/ai:chat-model-handler');
+export const IChatModelHandler = new Token('@jupyterlite/ai:IChatModelHandler');
 /**
  * Token for the diff manager.
  */
-export const IDiffManager = new Token('@jupyterlite/ai:diff-manager');
+export const IDiffManager = new Token('@jupyterlite/ai:IDiffManager');
 /**
  * The string that replaces a secret key in settings.
  */

package/lib/widgets/ai-settings.js

@@ -13,6 +13,7 @@ import MoreVert from '@mui/icons-material/MoreVert';
 import Settings from '@mui/icons-material/Settings';
 import { Alert, Box, Button, Card, CardContent, Chip, Dialog, DialogActions, DialogContent, DialogTitle, Divider, FormControl, FormControlLabel, IconButton, InputLabel, List, ListItem, ListItemText, Menu, MenuItem, Select, Switch, Tab, Tabs, TextField, ThemeProvider, Tooltip, Typography, createTheme } from '@mui/material';
 import React, { useEffect, useMemo, useState } from 'react';
+import { getEffectiveContextWindow } from '../providers/model-info';
 import { SECRETS_REPLACEMENT } from '../tokens';
 import { ProviderConfigDialog } from './provider-config-dialog';
 /**
@@ -409,6 +410,7 @@ const AISettingsComponent = ({ model, agentManagerFactory, themeManager, provide
                             const providerInfo = providerRegistry.getProviderInfo(provider.provider);
                             const providerToolCapabilities = providerInfo?.providerToolCapabilities;
                             const params = provider.parameters;
+                            const effectiveContextWindow = getEffectiveContextWindow(provider, providerRegistry);
                             const webSearchEnabled = !!providerToolCapabilities?.webSearch &&
                                 provider.customSettings?.webSearch?.enabled === true;
                             const webFetchEnabled = !!providerToolCapabilities?.webFetch &&
@@ -444,6 +446,7 @@ const AISettingsComponent = ({ model, agentManagerFactory, themeManager, provide
                                         (params?.temperature !== undefined ||
                                             params?.maxOutputTokens !== undefined ||
                                             params?.maxTurns !== undefined ||
+                                            effectiveContextWindow !== undefined ||
                                             webSearchEnabled ||
                                             webFetchEnabled) && (React.createElement(Box, { sx: {
                                                 display: 'flex',
@@ -454,6 +457,7 @@ const AISettingsComponent = ({ model, agentManagerFactory, themeManager, provide
                                             params?.temperature !== undefined && (React.createElement(Chip, { label: trans.__('Temp: %1', params.temperature), size: "small", variant: "outlined" })),
                                             params?.maxOutputTokens !== undefined && (React.createElement(Chip, { label: trans.__('Tokens: %1', params.maxOutputTokens), size: "small", variant: "outlined" })),
                                             params?.maxTurns !== undefined && (React.createElement(Chip, { label: trans.__('Turns: %1', params.maxTurns), size: "small", variant: "outlined" })),
+                                            effectiveContextWindow !== undefined && (React.createElement(Chip, { label: trans.__('Context: %1', effectiveContextWindow), size: "small", variant: "outlined" })),
                                             webSearchEnabled && (React.createElement(Chip, { label: trans.__('Web Search'), size: "small", variant: "outlined", color: "info" })),
                                             webFetchEnabled && (React.createElement(Chip, { label: trans.__('Web Fetch'), size: "small", variant: "outlined", color: "info" }))))),
                                     React.createElement(IconButton, { onClick: e => handleMenuClick(e, provider.id), size: "small" },
@@ -483,6 +487,11 @@ const AISettingsComponent = ({ model, agentManagerFactory, themeManager, provide
                                 }), color: "primary" }), label: React.createElement(Box, null,
                                 React.createElement(Typography, { variant: "body1" }, trans.__('Show Token Usage')),
                                 React.createElement(Typography, { variant: "caption", color: "text.secondary" }, trans.__('Display token usage information in the chat toolbar'))) }),
+                        React.createElement(FormControlLabel, { control: React.createElement(Switch, { checked: config.showContextUsage, onChange: e => handleConfigUpdate({
+                                    showContextUsage: e.target.checked
+                                }), color: "primary" }), label: React.createElement(Box, null,
+                                React.createElement(Typography, { variant: "body1" }, trans.__('Show Context Usage')),
+                                React.createElement(Typography, { variant: "caption", color: "text.secondary" }, trans.__('Display estimated context usage in the chat toolbar'))) }),
                         React.createElement(FormControlLabel, { control: React.createElement(Switch, { checked: config.showCellDiff, onChange: e => handleConfigUpdate({
                                     showCellDiff: e.target.checked
                                 }), color: "primary" }), label: React.createElement(Box, null,

package/lib/widgets/main-area-chat.d.ts

@@ -26,5 +26,6 @@ export declare class MainAreaChat extends MainAreaWidget<ChatWidget> {
      */
     get area(): string | undefined;
     private _writersChanged;
+    private _titleChanged;
     private _outputAreaCompat;
 }

package/lib/widgets/main-area-chat.js

@@ -1,7 +1,7 @@
 import { CommandToolbarButton, MainAreaWidget } from '@jupyterlab/apputils';
 import { launchIcon } from '@jupyterlab/ui-components';
 import { SaveComponentWidget } from '../components/save-button';
-import { TokenUsageWidget } from '../components/token-usage-display';
+import { UsageWidget } from '../components/usage-display';
 import { RenderedMessageOutputAreaCompat } from '../rendered-message-outputarea';
 import { CommandIds } from '../tokens';
 /**
@@ -10,7 +10,8 @@ import { CommandIds } from '../tokens';
 export class MainAreaChat extends MainAreaWidget {
     constructor(options) {
         super(options);
-        this.title.label = this.content.model.name;
+        this.title.label = this.model.name;
+        this.title.caption = this.model.title ?? this.model.name;
         const { trans } = options;
         // Move to side button.
         this.toolbar.addItem('moveToSide', new CommandToolbarButton({
@@ -30,25 +31,27 @@ export class MainAreaChat extends MainAreaWidget {
             }));
         }
         // Add the token usage button.
-        const tokenUsageWidget = new TokenUsageWidget({
+        const usageWidget = new UsageWidget({
             tokenUsageChanged: this.model.tokenUsageChanged,
             settingsModel: options.settingsModel,
             initialTokenUsage: this.model.agentManager.tokenUsage,
             translator: trans
         });
-        this.toolbar.addItem('token-usage', tokenUsageWidget);
+        this.toolbar.addItem('usage', usageWidget);
         // Temporary compat: keep output-area CSS context for MIME renderers
         // until jupyter-chat provides it natively.
         this._outputAreaCompat = new RenderedMessageOutputAreaCompat({
             chatPanel: this.content
         });
         this.model.writersChanged.connect(this._writersChanged);
+        this.model.titleChanged.connect(this._titleChanged);
     }
     dispose() {
         super.dispose();
         // Dispose of the approval buttons widget when the chat is disposed.
         this._outputAreaCompat.dispose();
         this.model.writersChanged.disconnect(this._writersChanged);
+        this.model.titleChanged.disconnect(this._titleChanged);
     }
     /**
      * Get the model of the chat.
@@ -74,5 +77,8 @@ export class MainAreaChat extends MainAreaWidget {
             this.content.inputToolbarRegistry?.show('send');
         }
     };
+    _titleChanged = () => {
+        this.title.caption = this.model.title ?? this.model.name;
+    };
     _outputAreaCompat;
 }

package/lib/widgets/provider-config-dialog.js

@@ -4,6 +4,7 @@ import Visibility from '@mui/icons-material/Visibility';
 import VisibilityOff from '@mui/icons-material/VisibilityOff';
 import { Accordion, AccordionDetails, AccordionSummary, Autocomplete, Box, Button, Chip, Dialog, DialogActions, DialogContent, DialogTitle, FormControl, FormControlLabel, IconButton, InputAdornment, InputLabel, List, ListItem, ListItemText, MenuItem, Select, Slider, Switch, TextField, Typography } from '@mui/material';
 import React from 'react';
+import { getProviderModelInfo } from '../providers/model-info';
 /**
  * Default parameter values for provider configuration
  */
@@ -80,6 +81,7 @@ export const ProviderConfigDialog = ({ open, onClose, onSave, initialConfig, mod
     const [expandedAdvanced, setExpandedAdvanced] = React.useState(false);
     const selectedProviderInfo = React.useMemo(() => providerRegistry.getProviderInfo(provider), [providerRegistry, provider]);
     const providerToolCapabilities = selectedProviderInfo?.providerToolCapabilities;
+    const selectedModelInfo = React.useMemo(() => getProviderModelInfo(selectedProviderInfo, model), [selectedProviderInfo, model]);
     const webSearchImplementation = providerToolCapabilities?.webSearch?.implementation;
     const supportsWebSearch = !!providerToolCapabilities?.webSearch;
     const supportsWebFetch = !!providerToolCapabilities?.webFetch;
@@ -300,13 +302,24 @@ export const ProviderConfigDialog = ({ open, onClose, onSave, initialConfig, mod
                                     maxOutputTokens: e.target.value
                                         ? Number(e.target.value)
                                         : undefined
-                                }), placeholder: trans.__('Leave empty for provider default'), helperText: trans.__('Maximum length of AI responses'), inputProps: { min: 1 } }),
+                                }), placeholder: trans.__('Leave empty for provider default'), helperText: trans.__('Maximum length of AI responses'), slotProps: { htmlInput: { min: 1 } } }),
                             React.createElement(TextField, { fullWidth: true, label: trans.__('Max Turns (Optional)'), type: "number", value: parameters.maxTurns ?? '', onChange: e => setParameters({
                                     ...parameters,
                                     maxTurns: e.target.value
                                         ? Number(e.target.value)
                                         : undefined
-                                }), placeholder: trans.__('Default: %1', DEFAULT_MAX_TURNS), helperText: trans.__('Maximum number of tool execution turns'), inputProps: { min: 1, max: 100 } }),
+                                }), placeholder: trans.__('Default: %1', DEFAULT_MAX_TURNS), helperText: trans.__('Maximum number of tool execution turns'), slotProps: { htmlInput: { min: 1, max: 100 } } }),
+                            React.createElement(TextField, { fullWidth: true, label: trans.__('Context Window (Optional)'), type: "number", value: parameters.contextWindow ?? '', onChange: e => setParameters({
+                                    ...parameters,
+                                    contextWindow: e.target.value
+                                        ? Number(e.target.value)
+                                        : undefined
+                                }), placeholder: selectedModelInfo?.contextWindow !== undefined
+                                    ? trans.__('Default: %1', selectedModelInfo.contextWindow.toLocaleString())
+                                    : trans.__('e.g., 128000'), helperText: selectedModelInfo?.contextWindow !== undefined &&
+                                    parameters.contextWindow === undefined
+                                    ? trans.__('Using provider metadata default of %1 tokens for this model unless you override it here.', selectedModelInfo.contextWindow.toLocaleString())
+                                    : trans.__('Model context window size in tokens (used for context usage estimation)'), slotProps: { htmlInput: { min: 1 } } }),
                             React.createElement(Typography, { variant: "body2", color: "text.secondary", sx: { mt: 2, mb: 1 } }, trans.__('Completion Options')),
                             React.createElement(FormControlLabel, { control: React.createElement(Switch, { checked: parameters.supportsFillInMiddle ?? false, onChange: e => setParameters({
                                         ...parameters,
@@ -344,7 +357,7 @@ export const ProviderConfigDialog = ({ open, onClose, onSave, initialConfig, mod
                                         webSearchImplementation === 'anthropic' && (React.createElement(React.Fragment, null,
                                             React.createElement(TextField, { fullWidth: true, label: trans.__('Web Search Max Uses'), type: "number", value: webSearchSettings.maxUses ?? '', onChange: e => updateCustomSetting('webSearch', 'maxUses', e.target.value
                                                     ? Number(e.target.value)
-                                                    : undefined), inputProps: { min: 1 } }),
+                                                    : undefined), slotProps: { htmlInput: { min: 1 } } }),
                                             renderDomainList('webSearch.blockedDomains', trans.__('Blocked Domains'), trans.__('spam.example.com'), webSearchSettings.blockedDomains))))))),
                                 supportsWebFetch && (React.createElement(React.Fragment, null,
                                     React.createElement(FormControlLabel, { control: React.createElement(Switch, { checked: webFetchSettings.enabled === true, onChange: e => updateCustomSetting('webFetch', 'enabled', e.target.checked) }), label: trans.__('Enable Web Fetch') }),
@@ -358,10 +371,10 @@ export const ProviderConfigDialog = ({ open, onClose, onSave, initialConfig, mod
                                         } },
                                         React.createElement(TextField, { fullWidth: true, label: trans.__('Web Fetch Max Uses'), type: "number", value: webFetchSettings.maxUses ?? '', onChange: e => updateCustomSetting('webFetch', 'maxUses', e.target.value
                                                 ? Number(e.target.value)
-                                                : undefined), inputProps: { min: 1 } }),
+                                                : undefined), slotProps: { htmlInput: { min: 1 } } }),
                                         React.createElement(TextField, { fullWidth: true, label: trans.__('Web Fetch Max Content Tokens'), type: "number", value: webFetchSettings.maxContentTokens ?? '', onChange: e => updateCustomSetting('webFetch', 'maxContentTokens', e.target.value
                                                 ? Number(e.target.value)
-                                                : undefined), inputProps: { min: 1 } }),
+                                                : undefined), slotProps: { htmlInput: { min: 1 } } }),
                                         renderDomainList('webFetch.allowedDomains', trans.__('Allowed Domains'), trans.__('docs.example.com'), webFetchSettings.allowedDomains),
                                         renderDomainList('webFetch.blockedDomains', trans.__('Blocked Domains'), trans.__('spam.example.com'), webFetchSettings.blockedDomains),
                                         React.createElement(FormControlLabel, { control: React.createElement(Switch, { checked: webFetchSettings.citationsEnabled === true, onChange: e => updateCustomSetting('webFetch', 'citationsEnabled', e.target.checked) }), label: trans.__('Enable Citations') })))))))))))),

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@jupyterlite/ai",
-    "version": "0.15.0",
+    "version": "0.17.0",
     "description": "AI code completions and chat for JupyterLite",
     "keywords": [
         "jupyter",
@@ -53,7 +53,8 @@
         "watch:src": "tsc -w --sourceMap",
         "watch:labextension": "jupyter labextension watch .",
         "docs": "jupyter book start",
-        "docs:build": "sed -e 's/\\[@/[/g' -e 's/@/\\@/g' CHANGELOG.md > docs/_changelog_content.md && jupyter book build --html"
+        "docs:build": "sed -e 's/\\[@/[/g' -e 's/@/\\@/g' CHANGELOG.md > docs/_changelog_content.md && jupyter book build --html",
+        "sync:model-context-windows": "node scripts/sync-model-context-windows.mjs && prettier --write src/providers/generated-context-windows.ts && eslint --fix src/providers/generated-context-windows.ts"
     },
     "dependencies": {
         "@ai-sdk/anthropic": "^3.0.58",

package/schema/settings-model.json

@@ -54,6 +54,11 @@
                 "maximum": 100,
                 "default": 25
               },
+              "contextWindow": {
+                "type": "number",
+                "description": "Model context window size in tokens (used for context usage estimation)",
+                "minimum": 1
+              },
               "supportsFillInMiddle": {
                 "type": "boolean",
                 "description": "Whether the model supports fill-in-middle completion"
@@ -211,6 +216,12 @@
       "type": "boolean",
       "default": false
     },
+    "showContextUsage": {
+      "title": "Show Context Usage",
+      "description": "Display estimated context usage percentage in the chat toolbar",
+      "type": "boolean",
+      "default": false
+    },
     "commandsRequiringApproval": {
       "title": "Commands Requiring Approval",
       "description": "List of commands that require user approval before AI can execute them",

package/src/agent.ts

@@ -4,6 +4,7 @@ import { IRenderMimeRegistry } from '@jupyterlab/rendermime';
 import { PromiseDelegate } from '@lumino/coreutils';
 import { ISignal, Signal } from '@lumino/signaling';
 import {
+  generateText,
   ToolLoopAgent,
   type ModelMessage,
   type LanguageModel,
@@ -13,11 +14,14 @@ import {
   type TypedToolError,
   type TypedToolOutputDenied,
   type TypedToolResult,
-  type AssistantModelMessage
+  type UserContent,
+  type AssistantModelMessage,
+  APICallError
 } from 'ai';
 import { ISecretsManager } from 'jupyter-secrets-manager';
 
 import { createModel } from './providers/models';
+import { getEffectiveContextWindow } from './providers/model-info';
 import {
   createProviderTools,
   type IProviderCustomSettings
@@ -53,6 +57,10 @@ interface IStreamProcessResult {
    * Whether an approval request was encountered and processed.
    */
   approvalProcessed: boolean;
+  /**
+   * Whether the stream was aborted before completion.
+   */
+  aborted: boolean;
   /**
    * The approval response message to add to history (if approval was processed).
    */
@@ -387,7 +395,17 @@ export class AgentManager implements IAgentManager {
     return this._activeProvider;
   }
   set activeProvider(value: string) {
+    const previousProvider = this._activeProvider;
     this._activeProvider = value;
+
+    // Reset request-level context estimate only when switching between providers.
+    if (previousProvider && previousProvider !== value) {
+      this._tokenUsage.lastRequestInputTokens = undefined;
+    }
+
+    this._tokenUsage.contextWindow = this._getActiveContextWindow();
+
+    this._tokenUsageChanged.emit(this._tokenUsage);
     this.initializeAgent();
     this._activeProviderChanged.emit(this._activeProvider);
   }
@@ -463,7 +481,11 @@ export class AgentManager implements IAgentManager {
 
     // Clear history and token usage
     this._history = [];
-    this._tokenUsage = { inputTokens: 0, outputTokens: 0 };
+    this._tokenUsage = {
+      inputTokens: 0,
+      outputTokens: 0,
+      contextWindow: this._getActiveContextWindow()
+    };
     this._tokenUsageChanged.emit(this._tokenUsage);
   }
 
@@ -485,12 +507,13 @@ export class AgentManager implements IAgentManager {
     this._pendingApprovals.clear();
 
     // Convert chat messages to model messages
-    const modelMessages = messages.map(msg => {
-      const isAIMessage = msg.sender.username === 'ai-assistant';
+    const modelMessages: ModelMessage[] = messages.map(msg => {
+      const role =
+        msg.sender.username === 'ai-assistant' ? 'assistant' : 'user';
       return {
-        role: isAIMessage ? 'assistant' : 'user',
+        role,
         content: msg.body
-      } as ModelMessage;
+      };
     });
     this._history = Private.sanitizeModelMessages(modelMessages);
   }
@@ -552,10 +575,17 @@ export class AgentManager implements IAgentManager {
    * Handles the complete execution cycle including tool calls.
    * @param message The user message to respond to (may include processed attachment content)
    */
-  async generateResponse(message: string): Promise<void> {
+  async generateResponse(message: UserContent): Promise<void> {
     this._streaming = new PromiseDelegate();
     this._controller = new AbortController();
     const responseHistory: ModelMessage[] = [];
+
+    // Add user message to history
+    responseHistory.push({
+      role: 'user',
+      content: message
+    });
+
     try {
       // Ensure we have an agent
       if (!this._agent) {
@@ -566,12 +596,6 @@ export class AgentManager implements IAgentManager {
         throw new Error('Failed to initialize agent');
       }
 
-      // Add user message to history
-      responseHistory.push({
-        role: 'user',
-        content: message
-      });
-
       let continueLoop = true;
       while (continueLoop) {
         const result = await this._agent.stream({
@@ -581,9 +605,22 @@ export class AgentManager implements IAgentManager {
 
         const streamResult = await this._processStreamResult(result);
 
-        // Get response messages and update token usage
+        if (streamResult.aborted) {
+          try {
+            const responseMessages = await result.response;
+            if (responseMessages.messages?.length) {
+              this._history.push(
+                ...Private.sanitizeModelMessages(responseMessages.messages)
+              );
+            }
+          } catch {
+            // Aborting before a step finishes leaves no completed response to persist.
+          }
+          break;
+        }
+
+        // Get response messages for completed steps.
         const responseMessages = await result.response;
-        this._updateTokenUsage(await result.usage);
 
         // Add response messages to history
         if (responseMessages.messages?.length) {
@@ -615,9 +652,41 @@ export class AgentManager implements IAgentManager {
       this._history.push(...Private.sanitizeModelMessages(responseHistory));
     } catch (error) {
       if ((error as Error).name !== 'AbortError') {
+        let helpMessage = `${(error as Error).message}`;
+
+        // Remove attachments from history on payload rejection errors
+        if (
+          APICallError.isInstance(error) &&
+          (error.statusCode === 400 ||
+            error.statusCode === 404 ||
+            error.statusCode === 413 ||
+            error.statusCode === 415 ||
+            error.statusCode === 422)
+        ) {
+          for (const msg of [...this._history, ...responseHistory]) {
+            if (msg.role === 'user' && Array.isArray(msg.content)) {
+              const hasMedia = msg.content.some(p => p.type !== 'text');
+              if (hasMedia) {
+                const textContent = msg.content
+                  .filter(p => p.type === 'text')
+                  .map(p => (p as { text: string }).text)
+                  .join('\n');
+                msg.content =
+                  textContent || '_Attachment removed due to error_';
+              }
+            }
+          }
+          helpMessage +=
+            '\n\nAttachments have been removed from history. Please send your prompt again.';
+        }
         this._agentEvent.emit({
           type: 'error',
-          data: { error: error as Error }
+          data: { error: new Error(helpMessage) }
+        });
+        this._history.push(...Private.sanitizeModelMessages(responseHistory));
+        this._history.push({
+          role: 'assistant',
+          content: helpMessage
         });
       }
     } finally {
@@ -627,16 +696,56 @@ export class AgentManager implements IAgentManager {
   }
 
   /**
-   * Updates token usage statistics.
+   * Create a transient language model to request a text response which won't be added to history.
+   * @param messages - the messages sequence to send to the model.
+   */
+  async textResponse(messages: ModelMessage[]): Promise<string> {
+    try {
+      const model = await this._createModel();
+      const result = await generateText({
+        model,
+        messages
+      });
+      this._updateTokenUsage(result.totalUsage, result.totalUsage.inputTokens);
+      return result.text;
+    } catch (e) {
+      throw `Error while getting the topic of the chat\n${e}`;
+    }
+  }
+
+  /**
+   * Updates cumulative token usage statistics from a completed model step.
    */
   private _updateTokenUsage(
-    usage: { inputTokens?: number; outputTokens?: number } | undefined
+    usage: { inputTokens?: number; outputTokens?: number } | undefined,
+    lastRequestInputTokens?: number
   ): void {
+    const contextWindow = this._getActiveContextWindow();
+    const estimatedRequestInputTokens =
+      lastRequestInputTokens ?? usage?.inputTokens;
+
     if (usage) {
       this._tokenUsage.inputTokens += usage.inputTokens ?? 0;
       this._tokenUsage.outputTokens += usage.outputTokens ?? 0;
-      this._tokenUsageChanged.emit(this._tokenUsage);
     }
+
+    this._tokenUsage.lastRequestInputTokens = estimatedRequestInputTokens;
+    this._tokenUsage.contextWindow = contextWindow;
+
+    this._tokenUsageChanged.emit(this._tokenUsage);
+  }
+
+  /**
+   * Gets the configured context window for the active provider.
+   */
+  private _getActiveContextWindow(): number | undefined {
+    const activeProviderConfig = this._settingsModel.getProvider(
+      this._activeProvider
+    );
+    return getEffectiveContextWindow(
+      activeProviderConfig,
+      this._providerRegistry
+    );
   }
 
   /**
@@ -699,6 +808,13 @@ export class AgentManager implements IAgentManager {
       activeProviderConfig && this._providerRegistry
         ? this._providerRegistry.getProviderInfo(activeProviderConfig.provider)
         : null;
+    const contextWindow = getEffectiveContextWindow(
+      activeProviderConfig,
+      this._providerRegistry
+    );
+
+    this._tokenUsage.contextWindow = contextWindow;
+    this._tokenUsageChanged.emit(this._tokenUsage);
 
     const temperature =
       activeProviderConfig?.parameters?.temperature ?? DEFAULT_TEMPERATURE;
@@ -806,7 +922,10 @@ ${richOutputWorkflowInstruction}`;
   ): Promise<IStreamProcessResult> {
     let fullResponse = '';
     let currentMessageId: string | null = null;
-    const processResult: IStreamProcessResult = { approvalProcessed: false };
+    const processResult: IStreamProcessResult = {
+      approvalProcessed: false,
+      aborted: false
+    };
 
     for await (const part of result.fullStream) {
       switch (part.type) {
@@ -868,7 +987,18 @@ ${richOutputWorkflowInstruction}`;
           await this._handleApprovalRequest(part, processResult);
           break;
 
-        // Ignore: text-start, text-end, finish, error, and others
+        case 'error':
+          throw part.error;
+
+        case 'finish-step':
+          this._updateTokenUsage(part.usage, part.usage.inputTokens);
+          break;
+
+        case 'abort':
+          processResult.aborted = true;
+          break;
+
+        // Ignore: text-start, text-end, finish, and others
         default:
           break;
       }