v3.1.13

Release 3.1.13

• Adiciona flag supportsEffort ao AIModel para gate do output_config.effort

• Evita enviar output_config.effort a modelos Anthropic que não suportam (evita 400 errors)

• Bump VERSION 3.1.12 → 3.1.13

v3.1.12

Release 3.1.12

• Adiciona output_config.effort à API Anthropic para controlar o effort real do modelo (Claude Opus 4.7+)

• Estende ReasoningLevel com .xhigh e .max; ícone passa a mostrar todos os 6 níveis (Off, Low, Medium, High, X-High, Max)

• Atualiza budget_tokens e maxOutputTokens em Anthropic, NVIDIA e OpenAI para os novos níveis

• Torna toda a área do ContextIndicatorView clicável (.contentShape(Circle))

• Bump VERSION 3.1.11 → 3.1.12

v3.1.11

Release 3.1.11

• Expande SystemPromptAdapted.md a partir do prompt Fable 5 (1354 linhas, adaptado para Nvidia AI Studio / macOS / Swift)

• Adiciona AgentErrorHandler, ToolCallParser e ToolExecutor para infraestrutura de agentes e tool calling

• Refatora ChatViewModel e simplifica ModelFetcher / NVIDIAAPIService / AnthropicAPIService

• Ajustes em AppState, SwiftDataStore, KnowledgeManager, PTYManager, SwarmOrchestrator, LiveCanvasView, EnvParser, GitHubService, ImageGenerationSkill

• Bump VERSION 3.1.10 → 3.1.11

v3.1.10

Bump version to 3.1.10

v3.1.9

🐛 Bug Fix: Messages Disappearing (Deeper Fix)

Problem: Previous fix didn't resolve the issue. Messages still disappeared during thinking/tool-calling mode.

Root Cause: The LazyVStack with equatable() modifier was causing rendering issues. The lazy rendering combined with the equatable comparison was preventing messages from being rendered properly during streaming updates.

Fix:

• Replaced LazyVStack with regular VStack to force all messages to render
• Removed equatable() modifier that was preventing re-renders
• Removed visibleMessageCount limit that was causing additional issues
• Simplified the message list structure

Changes:

• Views/Chat/ChatView.swift — Simplified message rendering
• Bump version to 3.1.9

v3.1.8

🐛 Bug Fix: Nested ScrollView Caused Messages to Disappear

Problem: When the model entered thinking mode, the chat interface would appear empty. Users had to scroll up to see the messages.

Root Cause: The ReasoningView had a ScrollView nested inside the main chat ScrollView. This is a known SwiftUI issue — nested ScrollViews can cause the outer ScrollView to collapse, making all messages appear empty.

Fix: Replaced the nested ScrollView in ReasoningView with a regular Text view that uses lineLimit(6) during streaming to cap the visible content. This prevents the layout collapse while still showing the reasoning content.

Changes:

• Views/Chat/MessageBubbleView.swift — Removed nested ScrollView in ReasoningView
• Bump version to 3.1.8

v3.1.7

🐛 Bug Fix: Messages Disappearing During Thinking/Tool-Calling

Problem: When the conversation had several messages and the model entered thinking mode or tool-calling, the chat interface would appear empty. Users had to scroll up to see the messages.

Root Cause: The LazyVStack was limiting visible messages to 20. When the model entered thinking mode, a new message with expanded reasoning would push previous messages out of view, making the chat appear empty.

Fix:

• Increased initial visibleMessageCount from 20 to 100
• "Load earlier messages" button now loads 50 at a time (was 20)
• Added retry mechanism to scrollToBottom (100ms delay) to handle async layout changes during reasoning expansion and tool results
• Reset visibleMessageCount to 100 on thread switch (was 20)

Changes:

• Views/Chat/ChatView.swift — Better message visibility and scroll handling
• Bump version to 3.1.7

v3.1.6

🐛 Bug Fix: Message Layout

Problem: When the app window was resized smaller, some words/text in messages would get hidden or overflow because the message bubbles didn't have proper width constraints.

Fix:

• Added frame(maxWidth: .infinity) to message bubbles so they fill available width
• Added fixedSize(horizontal: false, vertical: true) to text content so it wraps properly
• Reduced user message Spacer from 80 to 40 to give more room for content on narrow windows

Changes:

• Views/Chat/MessageBubbleView.swift — Proper width constraints for bubbles
• Bump version to 3.1.6

v3.1.5

🐛 Bug Fix: Better Fake Image Detection

Problem: Some models hallucinate image generation by describing an image in text without calling the generate_image tool. The previous fix only detected placeholder URLs.

Fix:

• Now also detects when the user asked for an image but no tool was called
• Checks for image-related keywords in the user's message (English + Portuguese)
• Strengthened system prompt to explicitly forbid placeholder URLs
• Clearer error message explaining the correct flow:
  1. Selected model should call generate_image tool with a detailed prompt
  2. Tool internally calls the image model (Flux.1-Dev, etc.)
  3. Generated image is returned and displayed

Changes:

• ViewModels/ChatViewModel.swift — Detect user image requests + missing tool calls
• Models/SystemPrompt.swift — Stronger anti-hallucination instructions
• Bump version to 3.1.5

v3.1.4

🐛 Bug Fix: Detect Fake Image Generation

Problem: Some models (especially those with weak tool calling support) would generate text that looks like an image result (e.g., data:image/png;base64,PLACEHOLDER) without actually calling the generate_image tool. This resulted in the user seeing a fake image placeholder instead of a real image or a clear error.

Fix: Added detection in ChatViewModel that recognizes when the model output contains image-like placeholders but no tool was actually called. When detected, the app now shows a clear error message explaining that the selected model doesn't support the generate_image tool and suggests alternatives (Claude, GPT-4o, or NVIDIA models with tool support).

Changes:

• ViewModels/ChatViewModel.swift — Detect fake image generation patterns
• Bump version to 3.1.4