LLM Chat is an efficient chat application that interfaces with various Language Model providers. It features a lightweight GUI and leverages the zai library for LLM interactions.
- GUI built with WebUI, using vanilla JS/HTML/CSS
- Supports multiple LLM providers (OpenAI, TogetherAI, OctoAI, OpenRouter)
- Customizable model parameters (temperature, top_p, frequency_penalty, presence_penalty, max_tokens)
- Efficient state management in Zig
- Streaming responses from LLMs
- Copy-to-clipboard functionality for messages
- Scrollable chat history with arrow key navigation
- Real-time model switching without restarting the application
- Error handling and user-friendly error messages
zig build run
zig build install -Doptimize=ReleaseFast --prefix ~/.local
zig build
zig-out/bin/llm-chat
To use different LLM providers, set the following environment variables:
- OpenAI:
OPENAI_API_KEY - TogetherAI:
TOGETHER_API_KEY - OctoAI:
OCTOAI_API_KEY - OpenRouter:
OPENROUTER_API_KEY
- All state and application logic reside in the Zig codebase, with HTML/JS/CSS as a frontend wrapper.
- The binary is small, using the system's installed web browser or webview.
- Implements a chat completion API with streaming support.
- Uses a general-purpose allocator for memory management.
- Implements custom error handling and propagation.
- Expand to include a CLI version with optional WebUI
- Incorporate retrieval elements
- Add local file editing features
Issues and pull requests are welcome! Feel free to contribute to the project.
This project is licensed under the MIT License.
The tech stack (Zig + WebUI + zai) provides a robust and efficient foundation for building LLM-powered applications. It combines Zig's performance and safety with web technologies' flexibility, making it an excellent choice for cross-platform GUI applications with LLM capabilities.