When you're too exhausted and lazy to discipline yourself, RoastGuard monitors your screen time in the background and unleashes savage, context-aware Google Gemma 4 roasts the second you overstay your welcome.
Experience what happens when you get caught doomscrolling. Choose your offending app, elapsed duration, and roast intensity.
Generating response...
Traditional screen-time apps demand active effort and sell your data. RoastGuard takes over the cognitive load entirely.
Runs as a background foreground service using Android's native SYSTEM_ALERT_WINDOW. When your time limit is crossed, it renders directly over the offending app with a compulsory reality check.
Trained on open weights for creative reasoning. No corporate sanitization or polite platitudes—Gemma 4 delivers authentic, biting comedic roasts calibrated to your exact habits.
gemma-4-26b-a4b-it
• Temperature: 0.85
• Brevity Enforced: < 25w
• Zero Cloud Tolls
Open reasoning models frequently leak internal <think>...</think> loops. Our 4-tier pipeline scrubs raw thoughts at stream, regex, and native Kotlin levels.
input.replaceAll(
RegExp(r'<think>[\s\S]*?</think>'),
''
).trim()
Waiting 3 seconds for an LLM when an overlay pops up ruins the psychological effect. Riverpod background workers pre-generate roasts into an LRU cache ahead of time.
unawaited(RoastAiService
.prefetchRoasts(intensity));
// Instant local delivery
Synced slider boundaries from 1 to 120 minutes with zero assertion clipping. Configure strict 5-minute caps for TikTok while allowing 45 minutes for Reddit.
slider.clamp(1.0, 120.0)
// Fully aligned bounds
// Settings + Dashboard sync
From native usage polling to model inference, token scrubbing, and immediate UI overlay presentation.
Monitors package foreground events via native Kotlin service every 15 seconds without draining battery.
Routes requests to Google Gemma 4 open weights as primary, falling back to ultra-low-latency Groq Llama 3 when offline.
Strips <think> blocks in memory before strings are written to SharedPreferences or dispatched to the UI.
Floats over third-party apps, locking touch focus until the user either closes the app or acknowledges the shame timer.
Why exert mental energy trying to fight dopamine loops when open-weight AI can automate it for you?