PocketMind — On-device AI Chat

📱 An AI chatbot that runs right on your phone

PocketMind is an on-device AI chatbot for Android. Every conversation is processed on the device itself and is never sent to an outside server, and it works the same way even with no internet connection.

━━━━━━━━━━━━━━━━━━━
✨ Why you'll like it
━━━━━━━━━━━━━━━━━━━
🔒 Privacy — inference never sends your prompts or replies to a model server.
✈️ Offline — works in airplane mode, on the subway, or abroad with no network.
📴 Local inference — once a model is ready, generation needs no network connection.
🌐 English & Korean — switches automatically with your device language.

━━━━━━━━━━━━━━━━━━━
🧠 Bring the model you want
━━━━━━━━━━━━━━━━━━━
• A curated model catalog, ranked for your device specs.
• Supports GGUF (llama.cpp) and LiteRT-LM formats.
• Open-license models can be fetched inside the app; for others, import a file you already have in one tap.
• Compare open-source models like Gemma, Qwen, Phi, and SmolLM.
• Search, filter, and sort by engine, Korean quality, speed, size, RAM fit, license flow, and install state.

━━━━━━━━━━━━━━━━━━━
⚙️ Fine-grained control
━━━━━━━━━━━━━━━━━━━
• One-tap Precise / Balanced / Creative presets
• Adjust Temperature / Top-P / Top-K / max tokens directly
• System prompt to set tone, role, and format
• Multi-turn context you can toggle on or off

━━━━━━━━━━━━━━━━━━━
💬 A comfortable chat experience
━━━━━━━━━━━━━━━━━━━
• Markdown rendering (headings, lists, quotes, bold, code)
• Code-block syntax highlighting (Kotlin / Python / JSON)
• Token count and generation time for each reply
• Copy, regenerate, or delete messages; long-press for a quick menu
• Save and search multiple chat sessions, grouped by time
• Material You dynamic color + Light / Dark theme
• Live device resources (memory, CPU, temperature)
• Assistant Lab benchmarks TTFT and tok/s on your own device
• Conversation personas, custom persona import/export, and system text-to-speech
• Choose a gallery image or take a photo for on-device OCR, review and edit the text, then add it to a question or chat knowledge

━━━━━━━━━━━━━━━━━━━
🧰 Local assistant workspace
━━━━━━━━━━━━━━━━━━━
• Add TXT, Markdown, JSON, PDF text layers, or DOCX/XLSX/PPTX Office text per conversation or globally and inspect answer sources
• Long-term Memory with global, persona, and conversation scopes, expiry, and provenance
• Restricted local calculator, date/time, and unit-conversion Tools with typed inputs and visible results
• Review and approve local reminders before creation, update, or cancellation
• Prompt Lab templates, parameter snapshots, and sequential A/B comparisons

━━━━━━━━━━━━━━━━━━━
📌 How to start
━━━━━━━━━━━━━━━━━━━
1. Pick a model that fits your device from the catalog.
2. Get it inside the app or import a file you downloaded.
3. Start chatting. Once the model is ready, it runs with no network.

━━━━━━━━━━━━━━━━━━━
💡 Good to know
━━━━━━━━━━━━━━━━━━━
• This app does not bundle model weights. You provide the model yourself.
• It runs small models (0.3B–4B), so quality differs from large cloud models.
• It shines at short summaries, casual chat, and code explanations.
• Supports arm64-v8a devices.

Contact: sia@siamakerlab.com

Recent changes:
• Stronger purchase signature verification and remove-ads entitlement handling
• Improved cancellation and concurrency for model operations
• Safer consent, privacy boundaries, and error presentation
• Better foldable layouts and accessibility across Chat and Settings
• Build stability updates for Gradle 9.7.1 and AGP 9.3.2