EveryTalk - Next-Generation Intelligent Conversation Platform

🚀 Ultra-smooth · 🎨 Highly Customizable · 🔒 Privacy First · 🌐 All-in-one Multimodal

A revolutionary Android AI client integrating cutting-edge technology stacks, supporting any LLM.
From text chat to image generation, voice interaction, and real-time web search, build your custom AI companion.

🧭 Project Status

• Project Positioning: A native Android AI client focusing on multi-model integration, real-time streaming, multimodal features, and local privacy.
• Target Platform: Android 8.1 and above (Android 10+ recommended); Development environment requires Android Studio + JDK 17.
• Capabilities: Covers text conversation, vision, image generation, live voice, LaTeX offline rendering, native web search, and OpenClaw remote control integration.
• Connection Modes: Supports direct upstream API calls (serverless), as well as routing through an optional proxy backend for image generation and extra endpoints.
• Minimum Requirement: First-time use requires at least one set of API endpoint, model name, and API key. Advanced capabilities depend on upstream models or backend support.

---

📌 Target Audience & Suitability

Suitable For

• Personal users who want to unify multiple LLMs (such as OpenAI o1/o3-mini, Gemini 2.5/2.0, Claude 3.7 Sonnet, DeepSeek-R1) in a single Android app.
• Heavy AI users who require high-performance streaming, multimodal inputs, advanced image generation (Flux.1, Imagen 3, DALL-E 3), and voice conversations.
• Developers wanting to learn modern native Android AI applications using Kotlin and Jetpack Compose.
• Advanced users requiring remote server control via the OpenClaw / Bridge setup.

Not Suitable For

• Users looking for a "ready-to-go" application with no API configurations or keys required.
• Users who only want the official client experience of a single model and do not need customization.
• Scenarios requiring native iOS, Web, or Desktop clients.
• Environments where all search, voice, and image capabilities must work completely offline.

---

⚠️ Prerequisites & Warnings

• The primary deliverable of this project is the native Android client located in app1/.
• The application supports a serverless direct connection mode, but not all models provide identical features (e.g., web search only works on models that support native search parameters).
• Features like image generation, voice interaction, backend proxies, and OpenClaw depend on whether your upstream models or server endpoints support them.
• "Supported" in this README means the client contains the integration pipeline, not that every model works with every config by default.

---

📚 Document Index

• Project Overview: Current README.md (English) / README_zh.md (中文).
• OpenClaw Integration: docs/openclaw-integration.md, detailing the setup of EveryTalk -> Bridge -> Gateway.
• Design Logs: docs/plans/, containing design documentation and historical architectural decisions.
• Developer / Agent Docs: CLAUDE.md and AGENTS.md, helpful for maintainers wanting to build and run tests.

---

🌟 Core Features

🎯 Intelligent Chat System

• 🤖 Full Model Compatibility: Seamlessly connect to OpenAI (GPT-4o, o1, o3-mini), Google Gemini (Gemini 2.5, 2.0 Pro/Flash), Anthropic Claude (Claude 3.7 Sonnet, 3.5 Opus), Zhipu GLM, and local models via Ollama (Llama 3.3, Qwen 2.5, DeepSeek-R1).
• ⚡ Ultra-fast Streaming: Millisecond-level stream parser for fluid typing animation. Features a pass-through stream controller that accumulates tokens and updates UI immediately.
• 🧠 Reasoning Chain Visualizer: Real-time visualization of thinking processes for reasoning models like DeepSeek-R1, OpenAI o1, and o3-mini.
• 🌐 Deep Web Search:
  • Gemini: Supports native Google Search tool call integration.
  • Qwen: Supports native search parameters.
  • DeepSeek/OpenAI: Displays search citations and results if supported by the endpoint.
• 🐍 Code Interpreter:
  • Gemini: Detects computation/data execution tasks, runs Python code in a secure sandboxed environment, and renders output.
• 🔄 Auto Fallback: Seamlessly switches to direct connection mode when Cloudflare challenges are detected.
• 📐 LaTeX Math Rendering: Supports inline ($...$) and block ($$...$$) LaTeX formulas. Fully offline rendering with light/dark adaptive themes.

🎨 Multimodal Interactions

• 🖼️ Vision & Documents:
  • Image Recognition: Support for GPT-4o, Gemini 2.0/2.5 Flash/Pro, Claude 3.7 Sonnet.
  • Document Analysis: Directly upload and chat with PDF, Word, and Excel files (local PDF parser included).
• 🎭 AI Image Generation: Connects to Flux.1, Imagen 3, DALL-E 3, and other custom text-to-image endpoints.
• 🎙️ Live Voice Dialogue:
  • OpenAI: High-quality TTS/STT combos or OpenAI Realtime audio protocols.
  • Gemini: Low-latency native multimodal audio flow input/output (Live Voice).
  • Azure/Aliyun: High-fidelity cloud text-to-speech engine integrations.
• 🎨 Image Compression & Editing: Smart on-device image optimization before upload.

⚙️ Customizations & Controls

• 🎛️ Parameters Fine-tuning: Fine-grained controls over Temperature, Top-P, Max Tokens, etc.
• 📝 System Prompts: Create and switch custom AI personas for various roles.
• 🔧 Multi-Config Switcher: Save multiple API profiles and switch models in one tap.
• 🎨 Preset Aspect Ratios: Choose from 1:1, 16:9, 9:16 for image generation.
• 💾 History Manager: Full conversation persistence and history search.
• ☁️ Telegram Releases: Get automatically notified of new APK releases in the official channel.

🛡️ Security & Performance

• 🔒 On-Device Encryption: Configuration and chat logs are stored locally with AES-256 encryption.
• 🔐 Message Signatures: Optional HMAC-SHA256 signature protocol verification for backend API calls.
• ⚡ Performance Tuning: Smart scroll controls (pauses autoscroll when user scrolls manually) and list view caching.

🎨 Modern UI/UX

• 🌓 Perfect Dark Mode: Curated dark and light themes using Material Design 3.
• ✨ Smooth Motion: Crossfades, spring animations, and tactile feedback.
• 🎯 Dynamic Sidebar: Easily switch chats, access settings, and manage multiple profiles.

---

🚀 Quick Start

1. Download & Installation

1. Go to the project's Releases page.
2. Download the latest app-release.apk.
3. Allow "Install from Unknown Sources" on your Android device and install the package.

1.1 CI/CD & Telegram Notifications

The project is configured with a automated pipeline (see .github/workflows/build-artifacts.yml):

1. Auto Release: Merging a Release PR builds and signs the APK and AAB, uploading them directly to GitHub Releases.
2. Telegram Broadcast: A bot notifies the Telegram channel with release notes and download links instantly upon build completion.

2. Developer Configuration

Before building or running the project locally, set up your keys:

1. Copy app1/local.properties.example and rename it to app1/local.properties.
2. Populate the keys with your credentials:

# local.properties example configuration
GOOGLE_API_KEY="AIzaSy..."
GOOGLE_API_BASE_URL="https://generativelanguage.googleapis.com"
DEFAULT_OPENAI_API_BASE_URL="https://api.openai.com"
SILICONFLOW_API_KEY="sk-..."
# For more parameters, refer to local.properties.example

---

🛠️ Technical Architecture

Frontend Technology Stack

┌─────────────────────────────────────────┐
│         Jetpack Compose UI              │
│     Material 3 · Responsive Layout      │
└─────────────────────────────────────────┘
              ↓
┌─────────────────────────────────────────┐
│       Kotlin Coroutines & Flow          │
│   Async Process · SSE Stream · State    │
└─────────────────────────────────────────┘
              ↓
┌─────────────────────────────────────────┐
│         Ktor Client Network             │
│   SSE Parser · Fallback · Direct Call   │
└─────────────────────────────────────────┘
              ↓
┌─────────────────────────────────────────┐
│      Local Data (Room Database)         │
│  Encrypted SharedPreferences · Files    │
└─────────────────────────────────────────┘

Core Components

• ApiClient.kt: Manages network calls, SSE deserialization, error recovery, and Cloudflare bypass.
• StreamingOutputController.kt: Pass-through streaming controller updating Compose states immediately.
• GeminiDirectClient.kt / OpenAIDirectClient.kt: Handles serverless connections directly to Google/OpenAI.
• VoiceChatSession.kt: Manages continuous voice sessions (STT → Chat → TTS).

System Requirements

Component	Requirement
Minimum SDK	Android 8.1 (API 27)
Recommended SDK	Android 10+ (API 29+)
IDE	Android Studio Koala+ / Ladybug+
JDK	JDK 17+
Backend	Optional (Supports direct serverless mode)

Backend Service (Optional Proxy)

• Open-source Proxy: backdAiTalk
• Functions: API proxying, remote image generation, rate limiting.
• Deployment: Supports Docker containerized setup with multi-instance load balancing.

---

📁 Project Structure

EveryTalk/
├── app1/                                    # 📱 Android Client Module
│   ├── app/
│   │   ├── src/main/java/com/android/everytalk/
│   │   │   ├── config/                      # ⚙️ Configurations
│   │   │   ├── data/                        # 📊 Data Layer
│   │   │   │   ├── network/                 # 🌐 Network APIs
│   │   │   │   │   ├── ApiClient.kt         # Master Client
│   │   │   │   │   ├── AppStreamEvent.kt    # SSE Event Models
│   │   │   │   │   └── StreamEventParser.kt # Parser
│   │   │   │   └── DataClass/               # Room Entities
│   │   │   ├── statecontroller/             # 🎮 State Management
│   │   │   │   ├── AppViewModel.kt          # Main ViewModel
│   │   │   │   ├── facade/                  # UI State Facade
│   │   │   │   └── ViewModelStateHolder.kt  # State Holder
│   │   │   ├── ui/                          # 🎨 UI Screen Components
│   │   │   │   ├── screens/                 # Screens
│   │   │   │   └── components/              # UI Custom Components
│   │   │   └── util/                        # 🔧 Utilities
│   │   └── build.gradle.kts                 # Build Script
│   └── local.properties.example             # Template Properties
├── ET-Backend-code/                         # 🖥️ Optional Backend Proxy
└── README.md                                # 📖 Master English Document

---

🔌 API Protocols

Chat Completions (Stream)

Endpoint: POST /chat

Request Format: multipart/form-data

├── chat_request_json: ChatRequest (JSON)
│   ├── model: string
│   ├── messages: Array<Message>
│   ├── useWebSearch: boolean
│   ├── showReasoning: boolean
│   └── generationConfig: {...}
└── uploaded_documents: File[] (Optional)

Response Format: Server-Sent Events (SSE)

data: {"type":"content","text":"Hello","output_type":null,"block_type":null}

data: {"type":"reasoning","text":"Let me think..."}

data: {"type":"web_search_results","results":[...]}

data: {"type":"finish","reason":"stop"}

data: [DONE]

Event Types:

• content: Incremental token.
• content_final: Completed message token.
• reasoning: Mind chain tokens (OpenAI o1/o3-mini, DeepSeek-R1).
• reasoning_finish: End of thinking step.
• web_search_status: Real-time web crawling status.
• web_search_results: Crawled search citations.
• tool_call: Tool invocations.
• error: Error logs.
• finish: Stream close trigger.

Image Generations

Endpoint: POST /v1/images/generations

Request Format: application/json

{
  "model": "flux-1-schnell",
  "prompt": "A beautiful sunset over the mountains",
  "image_size": "1024x1024",
  "batch_size": 1,
  "apiAddress": "https://api.siliconflow.cn",
  "apiKey": "sk-...",
  "forceDataUri": true
}

Response Format:

{
  "images": [
    {
      "url": "data:image/png;base64,..."
    }
  ]
}

---

🦞 OpenClaw Guide

If you wish to control your remote VPS instances running Claw via EveryTalk:

curl -fsSL https://claw.everytalk.cc | bash

See OpenClaw Integration for full configuration steps.

---

🤝 Contributions

1. Fork this repository.
2. Create your feature branch (git checkout -b feature/AmazingFeature).
3. Commit changes (git commit -m 'Add some AmazingFeature').
4. Push your branch (git push origin feature/AmazingFeature).
5. Create a Pull Request.

Make sure to follow Kotlin style guides and cover your changes with tests.

---

❓ FAQ

Q1: Why can't I connect to my local Ollama server?

A: Check the following:

1. Ensure your Android device and computer are on the same Wi-Fi network.
2. Use your computer's local IP address (e.g., 192.168.1.100), not localhost or 127.0.0.1.
3. Check your PC's firewall settings for the Ollama port (usually 11434).
4. Ensure Ollama is configured to listen on all interfaces by setting environment variable OLLAMA_HOST=0.0.0.0.

Q2: How does Web Search work?

A: The app routes web queries natively:

1. Gemini: Activates the native google_search tool call.
2. Qwen: Sends the query with native search parameters (enable_search=true).
3. DeepSeek: Triggers the search parameters. Search results will render in citation cards. Tap cards to open the link directly in your browser.

Q3: What are the latest supported models?

A: EveryTalk supports all OpenAI-compatible APIs. Recommended latest models include:

• Reasoning Models: DeepSeek-R1, OpenAI o1, o3-mini.
• Commercial Flagships: Claude 3.7 Sonnet, Claude 3.5 Opus, Gemini 2.5 Pro/Flash, GPT-4o.
• Open-source / Local: Llama 3.3, Qwen 2.5 (deployed via Ollama, LM Studio, vLLM).
• Image Generation: Flux.1, Gemini Imagen 3, DALL-E 3.

Q4: Does the app collect user telemetry?

A: No. All settings, configurations, keys, and chat histories are saved locally in the SQLite Room Database with AES encryption. Telemetry SDKs are not included.

Q5: The message list jumps when the streaming finishes. How do I fix it?

A: This is caused by spacing variations between draft markdown and finished structured components (like code blocks and tables). We resolve this with two features: streaming height placeholders and single-swap rendering. You can tweak these options in PerformanceConfig.kt:

ENABLE_STREAMING_HEIGHT_PLACEHOLDER = true
ENABLE_SINGLE_SWAP_RENDERING = true

---

📄 License

This project is licensed under the MIT License.

---

If this project helped you, please give us a ⭐ Star!

Made with ❤️ by the EveryTalk Team