Agora: Android BYOK Multi-Agent LLM Client with Local Model Support and Remote Device Control

This article introduces the open-source project Agora, an Android BYOK (Bring Your Own Keys) multi-agent LLM client for advanced AI users. Its core features include: direct connection to over 8 large model providers, non-linear conversation trees, local model execution, and encrypted remote device control. The project is maintained by newo-ether, open-sourced on GitHub (link: https://github.com/newo-ether/Agora), developed natively in Kotlin under the MIT license, and released in June 2026.

Mainstream AI applications face pain points such as difficulty in ensuring data privacy, inflexible conversation history management, cumbersome multi-model switching, and lack of local/remote control capabilities. Agora is designed to address these issues: it uses the BYOK model to let users control API keys and data flow, provides agent workflow and remote device management capabilities, and is positioned as an open-source LLM client for Android.

Core Design: 1. Direct connection without intermediaries: Data does not go through third-party relays, only stored in the local Room database, with zero telemetry tracking; 2. Non-linear conversation architecture: Tree-structured message format, supporting message editing and re-generation, branch exploration, and context preservation.

Multi-Provider Support: Built-in support for 8 mainstream providers: OpenAI, Anthropic, Google Gemini, DeepSeek, Qwen (DashScope), OpenRouter, Ollama, and Local (llama.cpp); supports multi-key configuration, alias rotation, and custom base URLs.

Agora has built-in rich tool calling capabilities: 1. Web search: Integrates search engines like Brave, Serper, Tavily, and SearXNG; 2. Code execution: Executes code snippets via Gemini; 3. Remote Shell Control (featured function): Based on the Conch protocol, it can execute remote commands, edit files, and search code repositories, with end-to-end encryption using ECDH+AES-256-GCM throughout; 4. File operations and memory management: Remote file reading/writing, semantic conversation history search.

For users with offline or privacy needs, Agora provides local model support: Runs GGUF format models based on llama.cpp, supports quantized models, can run smoothly on mid-range Android devices, and is fully offline usable without network connection.

Agora's application scenarios include: 1. Mobile office assistant: Manage remote servers and handle tasks during commutes; 2. Privacy-first interaction: Suitable for sensitive professions like lawyers and doctors; 3. Multi-model comparison research: Quickly switch between models from different providers; 4. Creative writing: Non-linear conversation trees help explore plotlines.

Agora represents a new direction for mobile LLM clients: it returns control to users, meeting the privacy and flexibility needs of advanced users through the BYOK model, local-first architecture, and agent capabilities. As local model capabilities improve and device computing power increases, Agora will play an important role in the AI democratization process, serving as an open-source choice for desktop-level AI experiences on mobile devices.