Skip to content

CHANGELOG.md

Latest commit

 

History

History
84 lines (67 loc) · 2.59 KB

CHANGELOG.md

File metadata and controls

84 lines (67 loc) · 2.59 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[5.0.0] - 2026-02-19

Removed

  • Image generation (DALL-E, Stable Diffusion, FLUX, local SD)
  • Video generation
  • Voice cloning and TTS (MLX-Audio, Native TTS)
  • GitHub panel and GitHub tool
  • MCP server tool
  • Web fetch tool
  • News tool
  • Autonomous agent
  • Multi-model comparison and multi-model provider
  • RAG service
  • Cost tracker
  • Intent router
  • Smart code actions
  • Prompt library and prompt template manager
  • Performance dashboard, gauges, and token metrics views
  • AI Capabilities unified layer (5 files)
  • 41 files deleted total (~16,000 lines removed)

Changed

  • Default model changed to Qwen 2.5 7B (from Llama 3.2 3B)
  • System prompt rewritten — compact, honest about capabilities (~500 tokens)
  • Tool count reduced from 40+ to 11 focused tools
  • Few-shot examples rewritten to cover all core tools
  • Performance metrics display simplified from gauges to text
  • SettingsView simplified (removed image, GitHub tabs)
  • Version bumped to 5.0.0 reflecting breaking scope change

Fixed

  • Tool prompt no longer consumes most of the context window
  • System prompt no longer claims capabilities that don't work

[4.0.0] - 2026-02-19

Added

  • Chat template support via Python daemon (ChatML, Llama, Mistral formats)
  • Structured message passing (JSON arrays instead of flattened strings)
  • Tool tier system (core + development)
  • Tool approval flow (auto-approve read-only, ask for write/execute)
  • Tool approval UI (inline approve/deny in chat)
  • Context budget system with per-model token allocation
  • Word-based token estimation (replaces character/4 heuristic)
  • Smart context assembly with rule-based compaction
  • Project context auto-include (file tree, recent files)
  • Context window size detection from model config

Changed

  • maxTokens restored from 512 to 2048
  • RepetitionDetector limits increased (16K chars, 4K tokens)

[1.2.0] - 2026-02-04

Added

  • macOS WidgetKit widget extension (small, medium, large)
  • Real-time model status monitoring in widget
  • App Group data sharing for widget

[1.1.0] - 2026-01-27

Added

  • MLX backend via mlx_lm CLI subprocess
  • Model auto-detection
  • Streaming token generation

[1.0.0] - 2025-11-18

Added

  • Initial release
  • Chat interface with SwiftUI
  • Basic code generation
  • MIT License

For detailed release notes, see GitHub Releases.