v3.3.1

Full Changelog: https://github.com/daemon-001/Mark-VII/blob/main/CHANGELOG.md

v3.3.x Groq Integration, Retry UX & Performance Improvements

Groq AI Provider (v3.3.0)

• Added Groq as a third AI provider alongside Gemini and OpenRouter
• Support for ultra-fast models (Llama 3, Mixtral, Gemma, etc.)
• Dynamic model fetching from /models endpoint (no hardcoding)
• SSE streaming with conversation context (last 6 messages)
• Firebase-managed API key with optional user override
• Integrated into model selector and settings UI

Retry UX & Traceability (v3.3.1)

• Retry actions linked to original prompts (retryOfPrompt)
• "Retried From" tag with quick navigation to source prompt
• Source prompt highlight animation for visibility
• Cleaner retry context without stale responses

Performance & Rendering

• Throttled streaming updates (~60 FPS) for smoother UI
• Introduced StreamingMarkdown for low-flicker live rendering
• Full markdown rendering with code copy after stream completion
• Optimized haptic feedback handling
• Improved mic waveform rendering using single Canvas
• Refined auto-scroll behavior

Sessions & Stability

• Per-user local session caching (chat_sessions_<userId>.json)
• Cache-first session loading with Firestore sync
• Improved IO threading to reduce UI blocking
• Stabilized drawer recomposition and callbacks

UI & Model Selector

• Added Groq tab in model selector
• Increased dropdown width for better layout
• Improved tab text sizing and spacing
• Removed unnecessary scaffold hacks

Fixes & Improvements

• Enabled Groq models in retry selector
• Added warning for unsupported image retries with Groq
• Fixed duplicate prompt issue in OpenRouter requests
• General cleanup and stability improvements