lovebrain

LoveBrain (Dating Advisor) is an open-source AI conversation assistance tool for Android, designed to help users express themselves more sincerely in romantic relationships. Its core interaction is: long-press any message in a chat app to automatically capture the context, and a floating panel streams reply suggestion cards in four styles—Recommended, Clear-headed, Playful, and Gentle—which the user can freely pick, edit, and manually send. The key differentiator is its long-term relationship memory system. It continuously models relationship status through a five-dimensional vector (Intimacy, Trust, Commitment, Passion, Security), reassessing every 3 topics with supporting reasons. It also tracks 8 relationship stages (Initial contact → Breaking the ice → Ambiguity → Passionate love → Adjustment → Stability → Crisis → Repair), assessing the current situation to suggest whether to advance, stabilize, or retreat. Additionally, every 5 topics, it reviews the last 25 rounds of conversation, distilling personalized insights into a local knowledge base to avoid repeating the same mistakes. A 3D profile feature continuously updates both parties' traits and interaction warmth, making reply suggestions increasingly aligned with real habits. On privacy, LoveBrain insists on zero telemetry and a backend-free design. The knowledge base, profiles, insights, and chat history are all stored in local files within the app's private directory, which users can view, edit, export, or delete at any time. The only network request occurs when the user taps "Generate," sending the current context to a user-configured AI Provider (default tested with DeepSeek; custom baseUrl, model, and API key are supported). The floating bubble runs as a foreground service, with no auto-start on boot or background residency. Beyond core reply generation, LoveBrain offers a Heart-to-Heart mode (deconstructing conflict situations via a six-step method), Daily Tips (providing relationship strategies based on the current stage), and Natural Openers (polishing smooth, unawkward conversation starters) as auxiliary features. In terms of performance, the developers tested 50 questions across 19 real chat scenarios, achieving a first-token latency of about 527ms, an 83% cache hit rate, and a total cost of approximately ¥0.17. The project uses dual licensing: AGPL-3.0 for open-source free use (personal, educational, or open-source projects), while closed-source commercial use requires a commercial license. It supports Android 8.0 and above; there is no iOS version currently.