Skip to main content
v0.12.0

V12: Offline AI, New Models, and Smarter Editing

Offline AI. New models. Smarter editing. V12 is here.

- Vysakh Sreenivasan

Twelve releases. Twelve weeks. V12 brings fully offline AI operation with bundled models, new model support including Gemma 3, SmolLM3, Qwen3 Coder, and DeepSeek R1, smarter markdown editing with automatic list continuation, and improved PowerPoint exports.

What's New in V12

Local AI Goes Fully Offline

The local AI now works completely offline with a bundled model catalog. No internet required. New models added: Gemma 3 (multimodal with image support), SmolLM3 3B, Qwen3 Coder 30B with 256K context, and DeepSeek R1 Distill for reasoning. A context usage indicator helps you track conversations. Tool selection is now smarter, automatically prioritizing based on your file types.

Chat Improvements

Assistant messages now render with proper markdown formatting. Better multi-turn conversations with cloud AI. Optimized handling for CSV files. The chat experience feels more polished and capable.

Smarter Markdown Editing

Press Enter in a list and it automatically continues. Smart list continuation makes writing markdown faster and more natural. Better activity tracking. The little details that make a big difference.

The Numbers

50+ commits. 4 major feature areas. 3 platforms. 12 weeks of continuous shipping.

50+
Commits
4
Feature Areas
3
Platforms
12
Weeks Shipping

Twelve Weeks of Shipping

V1 on Oct 31. V12 today. Twelve consecutive releases. Building in public, shipping every week.

View Full Changelog →

Twelve releases. Twelve weeks. Offline AI. New models. Smart editing. Download V12 now.

Download Diwadi V12
Full Changelog

Major Features

  • Local AI fully offline with bundled model catalog
  • New models: Gemma 3 (multimodal), SmolLM3 3B, Qwen3 Coder 30B-A3B, DeepSeek R1 Distill
  • Context usage indicator for conversations
  • Smart tool selection based on file types
  • Markdown rendering for assistant messages
  • Smart list continuation in markdown editor

Technical Improvements

  • Better multi-turn AI conversations
  • Improved CSV file handling in chat
  • Local LLM reliability fixes
  • Parse plain JSON tool calls
  • Detect when model echoes JSON instead of text
  • PowerPoint (DWPPT) export improvements

UI/UX Enhancements

  • Magic Triggers renamed from Automation with folder indicators
  • Local LLM support for automation agents
  • File navigator keyboard shortcuts (Cmd+C/X/V/A)
  • Better markdown editing activity tracking

Fixed

  • Fixed LLM runtime panic with spawn_blocking for tool execution
  • Fixed @ prefix stripping from file paths in chat
  • Fixed cursor position after @ mention selection
  • Fixed preview file size refresh after AI image generation
  • Fixed PDF Typst syntax corruption in system prompt
  • Fixed i18n missing translations
  • Fixed activity completion using correct conversationId