Examples overview

The repository ships runnable examples under example/, one per integration style.

Example catalog#

• Basic app: Dart console apps for chat, embeddings, SQLite vector retrieval and decision models.
• Chat app: Flutter chat app with a model library, runtime controls, multimodal input and speech.
• Laya Tetris: Flutter game played in real time by a decision model.
• Laya Command Bar: Flutter text field that a decision model, an embedding model or a small LLM reshapes as you type.
• llama.cpp-Style CLI: terminal chat with llama.cpp-style arguments and parity tooling.
• OpenAI-compatible server: OpenAI-style HTTP API over an on-device model.
• TUI coding agent: nocterm terminal coding agent with four tools.
• LoRA Training Notebook: a Jupyter notebook that fine-tunes a LoRA adapter for Qwen2.5 0.5B Instruct with Hugging Face peft and converts it to GGUF for LoRA adapters.

Which one should I start with?#

• Learn the API surface: Basic App.
• Try embeddings: Basic App embedding CLI.
• Build retrieval with a vector database: Basic App SQLite vector CLI.
• Learn decision models: Basic App decision CLI.
• Build a product UI: Chat App.
• Run decision models in a real-time app: Laya Tetris.
• Drive interactive UI from a classifier: Laya Command Bar.
• Keep a llama.cpp terminal workflow: llama.cpp-Style CLI.
• Serve tools and agents over HTTP: OpenAI-Compatible Server.
• Build an interactive coding agent in the terminal: TUI Coding Agent.
• Train your own adapter: LoRA Training Notebook.

Global example requirements#

• Dart SDK >= 3.10.7
• Flutter SDK >= 3.38.0 for Flutter examples
• Flutter Apple example builds that use SwiftPM companion packages require deployment targets of iOS 16.4 or macOS 14.0 or newer
• Internet on first run (runtime bundle resolution)
• The Chat App enables both native runtime families, llama.cpp and LiteRT-LM, because it demonstrates GGUF and .litertlm models.