What's New in v0.2.3

🔧 Bug Fixes

• Fix React hooks crash: ToolDisplay component crashed with "Rendered fewer hooks than expected" when rendering human_interact tool (login/captcha scenarios). Fixed hooks order to comply with React rules.
• Fix workflow [object Object] display: Watch nodes in workflow editor displayed [object Object] instead of description text. Added proper description field fallback.

🔌 Watch Trigger Improvements (jarvis-agent 0.3.1)

• Initial condition check: watch_trigger now checks if the target condition is already met on the page before entering the monitoring loop. If a "Retry" or "Publish" button already exists, it returns immediately instead of waiting.
• Text model support: For non-vision models (e.g. DeepSeek), watch_trigger uses extract_page_content text comparison instead of screenshot-based visual comparison, fixing the infinite loop issue where text models couldn't process image data.
• Vision model detection: Automatically detects whether the LLM supports vision and routes to the appropriate comparison method.

Downloads

What's New in v0.2.2

🧠 Cross-Session Memory System

• Full memory system with automatic extraction from Explore conversations and manual save in Chat mode
• Hybrid search combining keyword matching and vector embeddings
• Embedding model selector supporting multiple providers (OpenAI, Voyage, local, etc.)
• Memory management panel in Settings with search, add, delete, and clear
• memory_search and memory_save tools injected into ChatAgent for proactive memory access

🔌 MCP Enhancements

• JSON Import: Batch import MCP services by pasting standard mcpServers JSON configs from platforms like 51ifind
• Headers Support: Per-service HTTP headers (e.g. Authorization) with secure password input
• Streamable HTTP Protocol: Auto-detect and fallback between Streamable HTTP and SSE protocols
• Chat Mode MCP: Chat mode can now directly call MCP tools without going through sub-agents — configure in Agent settings under "Chat Agent"

🛠 Skills System

• Skills panel with card grid, detail modal, and file tree viewer
• /skill-name command support in home and main pages with autocomplete
• Skill context injection for planner
• Cross-platform zip import support

🔧 Improvements

• Default session mode changed to Explore
• Code preview uses CSS counter line numbers for proper word wrap

Downloads

What's New

Skills System (New!)

• Skill Service: Backend service with IPC handlers and cross-platform resource loading
• Settings Panel: Skills management with card grid, detail modal, and file tree view
• Chat Integration: /skill-name command support with SkillCommandPopover in chat input
• Planner Injection: Skill context extracted and injected into Explore mode planner

Cross-Session Memory System (New!)

• Memory Recall: Auto-recall relevant memories during conversations via hybrid search (BM25 + vector cosine similarity)
• Auto Extraction: LLM-powered memory extraction from both Chat and Explore mode conversations
• Memory Tools: LLM can proactively search and save memories via memory_search / memory_save in Chat mode
• Vector Embedding: Optional embedding support (OpenAI/Gemini) for semantic search
• Management UI: Full settings panel with search, add, delete, statistics, and embedding model selector

Improvements

• Auto-generate task title from user's first message
• CSS counter line numbers for proper word wrap in code preview
• Upgrade @jarvis-agent/core and electron to 0.2.5

Downloads

• macOS: DeepFundAIBrowser-0.2.1-universal.dmg (Apple Silicon + Intel)
• Windows: ai-browser-0.2.1-win-x64-setup.exe (installer) or .zip (portable)

What's New

Chat Mode (New!)

• Dual mode interaction: Chat mode (default) for lightweight Q&A, Deep Explore mode for complex multi-step workflows
• Streaming text response with Markdown rendering and inline tool calls
• Multi-turn conversation with built-in memory management
• Thinking process display with expandable detail panel
• Auto-generated task titles from user's first message

Web Search Integration (New!)

• Built-in web search via Tavily, Serper, or SearXNG (self-hosted)
• Webpage QA — extract and analyze content from open browser tabs
• Configurable search provider in Chat settings

Workflow Enhancement

• Editable workflow steps with user confirmation before execution
• Task pause/resume support
• Expert mode with thinking process display
• Workflow confirm merged into workflow card UI

Agent System

• Per-agent MCP service configuration
• Custom agent support with user-defined prompts
• Built-in tools visibility in agent settings
• MCP service management panel

Bug Fixes & Improvements

• Fix deepseek-reasoner missing reasoning_content in multi-turn chat
• Fix multi-window global service collision
• Fix taskMode persistence across page navigation
• Fix MCP 204 notification response
• Custom SVG icons replacing Ant Design built-in icons
• Upgraded @jarvis-agent/core and electron to 0.2.5

Note

If you're upgrading from an older version, please click "Reset Settings" first, then reconfigure your settings.

What's New

Features

• Dynamic model fetching for Google Gemini and Anthropic: Now you can fetch available models directly from Google Gemini and Anthropic APIs by clicking the "Fetch" button in provider settings.

Improvements

• Refactored model fetching logic to use IPC, bypassing CORS restrictions
• Updated Google Gemini base URL to use OpenAI-compatible endpoint

Note

If you're upgrading from an older version, please click "Reset Settings" first, then reconfigure your settings.

What's New

Features

• Multi-tab browser support - Full multi-tab browsing with tab bar and management
• Address bar - Editable URL input with navigation buttons (back/forward/refresh)
• History screenshot overlay - View historical screenshots with close button
• Playwright E2E testing - Added end-to-end testing framework

Improvements

• Improved multi-tab stability and cleanup
• Better history view positioning

Dependencies

• Upgraded @jarvis-agent/electron to 0.1.14

What's New

UI/UX Improvements

• Optimized light/dark theme support across all interfaces
• Added purple-themed settings panels with enhanced interaction states
• Improved history panel and scheduled task components
• Optimized Main Page chat interface and Home page interactions

Architecture Improvements

• Added reusable UI component library (SelectableCard, ActionButton, SettingsDivider)
• Centralized design tokens for consistent styling
• Migrated settings panels and selectors to use unified UI components

Bug Fixes

• Fixed border and divider colors for light/dark mode
• Replaced static antd message with App.useApp() hook

Documentation

• Improved UX documentation and added important notices

⚠️ Breaking Changes

This release includes major architectural changes. Settings from previous versions may not be compatible.

✨ Key Features

Unified Settings System

• All configuration now through Settings UI (no more .env files)
• 8 comprehensive settings panels: Providers, General, Chat, Agent, Scheduled Tasks, UI, Network, Memory
• Real-time settings sync across all windows
• Import/Export configuration support

Provider Management

• Provider validation before sending messages
• Visual warning when no provider configured
• Improved model selection UI

Production Fixes

• Fixed Settings window access in production builds
• Resolved dark mode styling issues
• Updated to v0.1.0

📦 Installation

Download DeepFundAIBrowser-0.1.0-universal.dmg for macOS (Universal)

📝 Full Changelog

• feat(settings): unified settings system with 8 panels
• feat(settings): provider validation and warnings
• fix(settings): production URL and dark mode fixes
• docs: updated configuration guide
• refactor(ui): enhanced light/dark theme support

For detailed changes, see commit history from v0.0.15 to v0.1.0

Release v0.0.15

Features & Improvements

• Enhanced file preview with syntax highlighting
• Custom modal window system with i18n support
• Tray icon improvements for macOS and Windows

Bug Fixes

• Fixed tray icon not showing on Windows
• Fixed app not exiting properly on Windows
• Fixed tray context menu on macOS

Downloads

• macOS (Universal): DeepFundAIBrowser-0.0.15-universal.dmg

For auto-update configuration, see latest-mac.yml.

What's New in v0.0.14

Features

• Add custom modal window system with i18n support
• Add file download and loading page drag support
• Add playback speed control (0.5x-50x) for history replay

Bug Fixes

• Upgrade react and next.js to patch CVE-2025-55182
• Update onHumanSelect signature for jarvis-agent 0.1.11 compatibility