little-agent is a lightweight embedded agent framework (similar to Claude Code and OpenAI Codex). It supports multiple model providers and can be integrated with your apps easily.
There are already many agent projects around the world, and they are valuable both for practical use and for study. However, I want to write my own agent — one that is simple, maintainable and embeddable. Instead of designing an overly complex mechanism, I prefer to keep this project KISS. The agent is essentially a simple loop: the model receives user input and may output tool-call requests. The agent executes tools and sends the results back to the model. The loop continues until there are no more tool-call requests. With this simple design, I want to continuously explore the model’s capabilities. And since the project is written in Rust and exports C APIs, I can integrate it into my other apps and add agentic features to them.
It's recommended to use the little-agent crate, which provides some built-in tools and is ready to use. For more advanced use cases, you can also use the core crate, which enables finer-grained control over the agent.
little-agent also exports some C APIs. You can build it as a dynamic library, and link your apps against it. See include/little-agent.h for the interface.
For Rust APIs, run cargo doc to build the docs.
The project also provides a simple CLI. It's basic and mainly intended to demonstrate how to use the library.
To use the CLI, you need to set the following environment variables:
export OPENAI_API_KEY="<Your API Key>"
export OPENAI_BASE_URL="https://api.openai.com/v1"
export OPENAI_MODEL="gpt-5.3"Simply cargo run to run the CLI.
Licensed under MIT License, see LICENSE for more information.