What's Changed
Reasoning/Thinking Support
- Per-provider reasoning toggles in Settings (Anthropic, OpenAI, Gemini)
- Anthropic: Extended thinking with configurable budget (1024-32000 tokens)
- OpenAI: Reasoning effort (low/medium/high) for o3, o4-mini, GPT-5.x models
- Gemini: Thinking level (low/medium/high) for gemini-3-*-preview models
- DeepSeek: Automatic reasoning for deepseek-reasoner model
- Chat indicator shows when reasoning/thinking was used
Per-Provider Action Reasoning
- Actions can now configure reasoning per-provider independently
- Replaced single "Force ON" with per-provider settings dialog
- Supports Anthropic budget, OpenAI effort, Gemini level per action
Debug Improvements
- Debug info now stored per-message so you can come back to it for debugging
- Debug info excluded from chat exports
Model Constraints System
- Centralized model parameter constraints in model_constraints.lua
- Auto-adjusts temperature for reasoning models (forced to 1.0)
- Prefix matching for versioned model names
Bug Fixes
- Fixed Gemini thinking API format (camelCase, inside generationConfig)
- Fixed reasoning indicator for streaming mode - Fixed hardcoded provider lists in action dialogs
- Fixed OpenAI reasoning_effort parameter placement
Testing
- Added model validation tests
- New test infrastructure for constraint verification
Full Changelog: v0.7.1...v0.8.0