trae-agent

Trae Agent is an open-source software engineering agent from ByteDance. It takes natural language instructions from a developer and carries out coding tasks autonomously: writing code, editing files, running commands, and working through multi-step workflows. You describe what you want done in plain English and the agent figures out the steps.

The agent runs through a command-line interface and can be connected to several AI model providers, including OpenAI, Anthropic, Google Gemini, and others. You configure which model to use through a YAML file or environment variables, then point the agent at a project directory and give it a task. An interactive mode is also available for back-and-forth conversation with the agent as it works.

One feature called Lakeview generates short summaries of what the agent is doing at each step, making it easier to follow along without reading every action in detail. The agent also records a full log of all actions it takes, which can be replayed or analyzed afterward. Tasks can optionally run inside a Docker container for isolation, which is useful when you want the agent to have access to a specific environment without affecting your main machine.

The project describes itself as research-friendly. The codebase is designed to be modular and easy to modify, making it suitable for people studying how AI agents work or experimenting with new agent capabilities. It is actively developed and accepts contributions. The full README is longer than what was shown.