Features — Private Chat Hub

🤖 Multi-Backend AI Support

Private Chat Hub supports four distinct AI backend types. Switch between them from the settings screen at any time:

• Ollama — Connect to any self-hosted Ollama server over your local network. Supports the full Ollama model library: Llama 3, Mistral, Gemma, Qwen, DeepSeek, Phi, and more. Full tool-calling and vision support for compatible models.
• LM Studio — Connect to LM Studio's local server running any GGUF model. Vision and multimodal input supported where the loaded model allows.
• OpenCode — Route requests through a self-hosted opencode server to cloud providers (OpenAI GPT-4o, Anthropic Claude, Google Gemini, and more). You supply your own API keys; traffic goes through your own server.
• On-Device (LiteRT) — Run Gemma 3 or Gemini Nano locally on the Android device with Google's LiteRT runtime. Completely offline—no network requests made.

⚡ Real-Time Streaming Responses

Responses appear token-by-token as the model generates them, just like a human typing. The app handles streaming for all supported backends:

• Ollama — newline-delimited JSON stream
• LM Studio — Server-Sent Events (SSE)
• OpenCode — SSE event stream
• On-Device — live inference callbacks

🛠️ Tool Calling / Function Calling

AI models can invoke built-in tools during a conversation to retrieve real-world information or perform actions. Tool calling is supported on Ollama (and cloud models via OpenCode).

Built-in tools:

• Current Date & Time — Returns the current timestamp
• Get Current Location — Returns the device's approximate location via IP geolocation
• Web Search — Searches the web using Jina AI and returns results with clickable source links
• Fetch URL — Retrieves and returns the contents of any web page or URL
• Show Notification — Sends a local Android notification from the AI response

Tools can be individually enabled or disabled per conversation in Chat Settings → Tools. The app uses an agentic loop: if the model requests a tool call, the app executes it and sends the result back, continuing until a final text response is produced.

🌐 Web Search Integration

The built-in web search tool uses Jina AI to fetch live search results. When the AI uses web search:

• Results include titles, snippets, and URLs
• Source links are rendered as tappable links in the response
• Works on any model that supports tool calling
• No Jina API key required for basic usage

👁️ Vision & Image Input

Attach images and documents directly to your chat messages:

• Pick photos from the gallery or take a new photo with the camera
• Attach files (PDFs, text, documents) for document-based Q&A
• Vision models (e.g. llava, gemma3, llama3.2-vision) can describe, analyze, and reason about attached images
• Images are encoded locally and sent to the model; they never leave your network
• Works with Ollama vision models and LM Studio vision models
• Receive shared images from other Android apps via Share intent

🔊 Text-to-Speech (TTS)

Have AI responses read aloud using your device's built-in TTS engine:

• Tap the speaker icon on any message to start playback
• Streaming TTS: speech begins playing before the full response is complete
• Tap again to stop playback at any time
• Uses Android's native TTS engine (no additional apps required)
• Works with all backends and all models

📊 Model Comparison

Run the same prompt against two or more models simultaneously and compare responses side-by-side:

• Open the comparison screen from the chat menu
• Select any two models (can mix backends, e.g. Ollama vs LM Studio)
• Responses appear in parallel columns, streaming in real time
• Quickly evaluate quality, speed, and style differences between models

📝 System Prompt per Conversation

Each conversation can have its own system prompt (AI persona / instructions):

• Set or edit the system prompt in Chat Settings
• The prompt is sent to the model at the start of every request
• Persisted with the conversation so it survives app restarts
• Examples: "You are a Python expert", "Respond only in French", "Be concise"

📁 Project Workspaces

Organise conversations into named projects or topics:

• Create projects from the conversation list screen
• Assign conversations to projects for easy navigation
• Filter the conversation list by project
• Ideal for separating work, personal, and experimental chats

🧠 Thinking / Extended Reasoning

Support for models that expose their internal reasoning steps:

• Thinking traces are displayed in a collapsible section above the final answer
• Works with Ollama models that support the think parameter (e.g. deepseek-r1, qwq)
• Helps you verify the model's reasoning process

📶 Offline & Message Queuing

• On-device models work completely offline—no network required
• Network connectivity is monitored in real time
• Messages sent while offline are queued locally and delivered automatically when reconnected
• Connection status is displayed in the app's status bar

💬 Conversation Management

• Full conversation history with search
• Edit the conversation title at any time
• Delete individual messages or entire conversations
• Export conversations as JSON or Markdown
• Import previously exported conversations
• Conversations are stored entirely on-device (SQLite / shared preferences)

✍️ Rich Rendering: Markdown & LaTeX

• Responses are rendered with full Markdown support: headings, bold, italic, lists, tables, code blocks with syntax highlighting
• LaTeX / MathJax expressions (inline and block) are rendered as proper math notation
• Code blocks include a copy button
• Links in responses are tappable

🔒 Privacy & Data Ownership

• All conversation data stored locally on your device
• No telemetry, analytics, or crash reporting sent anywhere
• Cloud API keys (for OpenCode) stored encrypted on-device
• Network traffic goes only to your own servers (or directly on-device)
• Open-source (MIT License) — audit the code yourself
• Export and own your full conversation history at any time

📱 Android Integration

• Receive text and images shared from other apps via Android's Share intent
• Share conversations or responses to other Android apps
• Local push notifications for long-running AI tasks
• Material Design 3 adaptive theming (dark mode always on)
• Background task support with notification actions
• Clipboard integration for quick copy/paste

📚 Learn More

Installation Guide • API Comparison (Ollama vs LM Studio vs OpenCode) • GitHub Repository