Highlights 🎉
v0.6.9 delivers two highly requested features: image support for multimodal models and stable MCP (no more experimental flags!). Plus intelligent model capability detection, enhanced tool calling, and major UX improvements.
🖼️ Multimodal AI is here
Image upload support — Finally! Upload images directly in your chats and get visual understanding from compatible models:
- Local models: Automatic support for Gemma3, Qwen3, and other vision-capable models
- Cloud models: Manual toggle for GPT-4V, Claude 3.5 Sonnet, Gemini Pro Vision
- Smart detection: Jan automatically identifies which models support images
- Privacy first: Local models process images entirely offline
The team has been working hard to bring this long-awaited capability to Jan, and it's been worth the wait.
🔧 MCP is stable
Model Context Protocol graduated from experimental to stable:
- No more experimental flags — MCP tools work out of the box
- Better error handling for smoother MCP server connections
- Cancel tool calls mid-execution when needed
- Enhanced reliability with improved server status tracking
MCP unlocks powerful workflows: web search, code execution, productivity integrations, and more.
🎯 Intelligent model management
- Auto-detect capabilities: Jan identifies tool calling and vision support for local models
- Model compatibility checker: Hub shows hardware requirements before download
- Favorite models: Mark frequently used models for quick access
- Universal GGUF import: Import any valid .gguf file regardless of extension
- Hardware-aware suggestions: Get model recommendations based on your system specs
🚀 Enhanced tool calling
- GPT-OSS models: Upgraded llama.cpp brings tool calling to more open-source models
- Improved performance: Better tool execution with upgraded engine
- Remote provider control: Manual toggle for cloud model capabilities
- Streamlined workflows: Cancel operations, better error handling
Major Features
🖼️ Multimodal support
- Image uploads work with compatible local and cloud models
- Automatic detection for local model vision capabilities
- Manual toggle for remote providers (GPT-4V, Claude, Gemini)
- Privacy-preserving local image processing
🔧 Stable MCP
- MCP Server stable release — no more experimental flags needed
- Enhanced error handling for MCP connections
- Tool cancellation support during execution
- Improved server status synchronization
🎯 Smart model features
- Favorite models — bookmark your go-to models
- Auto-capability detection for local models
- Hardware compatibility checks in Hub
- Universal GGUF import regardless of file extension
⚡ Enhanced engine
- Tool calling support for GPT-OSS models
- Upgraded llama.cpp version with stability improvements
- Better performance across model types
Improvements
🔄 API & automation
- Auto-start API server on Jan startup (optional)
- Model auto-loading when API server starts
- Ollama endpoint support restored for custom configurations
🎨 User experience
- Thinking windows for OpenRouter models render correctly
- Better error messages across MCP operations
- Improved import UX with retired model cleanup
- CPU architecture detection at runtime
🔧 Technical enhancements
- Vulkan backend re-enabled for integrated GPUs with sufficient memory
- Enhanced stability with better error handling
- Performance optimizations across the board
Thanks to our incredible team
The engineering team delivered major features that users have been requesting for months. Image support required deep multimodal AI integration, while stabilizing MCP involved extensive testing and refinement. The auto-detection features showcase thoughtful UX design that makes AI more accessible.
Special recognition to the contributors who made v0.6.9 possible through their dedication to bringing powerful, privacy-focused AI capabilities to everyone.
Update your Jan or download the latest (opens in a new tab).
For the complete list of changes, see the GitHub release notes (opens in a new tab).