Trae Agent: open source software engineering task automation tools

Trae Agent is a software engineering task automation tool based on the Large Language Model (LLM) open-sourced by ByteDance. It receives natural language commands through the command line interface (CLI) to automate complex programming tasks such as writing code, fixing bugs or optimizing programs. The project is currently in alpha stage, still under active development, and community contributions are welcome.Trae Agent supports multiple large language models, such as OpenAI and Anthropic, and provides a rich ecosystem of tools, including file editing, script execution, and other features. Its modular design is easy for developers to customize and suitable for research and development of new intelligences. Users can quickly get started with a simple installation process and enjoy a transparent development experience.

Function List

• Support for natural language commands, such as "Create a Python script" to automatically generate code.
• Supports multiple LLM providers, including OpenAI, Anthropic, Doubao, etc. for flexible model switching.
• Provides file editing, bash script execution, sequential reasoning and other tools to meet the diverse programming needs.
• The interaction model supports multiple rounds of dialog and is suitable for iterative development.
• Record detailed operation logs and save them as JSON files for easy debugging and analysis.
• Flexible configuration is supported through JSON profiles and environment variables.
• Provides an easy installation process for quick deployment using pip or UV tools.

Using Help

Installation process

Trae Agent is easy to install, and a Python 3.12 environment is recommended. The following are the detailed steps:

1. Preparing the environment
   Ensure that Python 3.12 and pip are installed. it is recommended that you use the UV tool to manage your virtual environment and install UV:

   pip install uv
   

Create and activate a virtual environment:

uv venv
source .venv/bin/activate  # Linux/Mac
.venv\Scripts\activate      # Windows

2. cloning project
   Clone the Trae Agent repository from GitHub:

   git clone https://github.com/bytedance/trae-agent
   cd trae-agent
   

3. Installation of dependencies
   Install the dependencies using UV or pip:

   uv pip install -r requirements.txt
   

   Or:

   pip install -r requirements.txt
   

4. Configuring API Keys
   The Trae Agent supports a variety of LLM providers, and you need to configure the corresponding API keys. For example, for OpenAI and Anthropic Configure the key:

   export OPENAI_API_KEY='your_openai_api_key'
   export ANTHROPIC_API_KEY='your_anthropic_api_key'
   

   Verify that the key was set successfully:

   echo $OPENAI_API_KEY
   echo $ANTHROPIC_API_KEY
   

5. configuration file
   The Trae Agent uses a JSON configuration file to manage settings, located in the project root directory under the config.json. Example Configuration:

   {
   "default_provider": "anthropic",
   "max_steps": 20,
   "model_providers": {
   "openai": {
   "api_key": "your_openai_api_key",
   "model": "gpt-4o",
   "max_tokens": 128000,
   "temperature": 0.5
   },
   "anthropic": {
   "api_key": "your_anthropic_api_key",
   "model": "claude-sonnet-4-20250514",
   "max_tokens": 4096,
   "temperature": 0.5
   }
   }
   }
   

   After saving, run the following command to check the configuration:

   trae show-config
   

Functional operation flow

The core functionality of the Trae Agent is provided through the trae command invocation, the following is how to use the main functions:

1. Single-task implementation
   utilization trae-cli run command, you can enter natural language instructions to trigger the task. For example, create a Fibonacci array script:

   trae-cli run "Create a Python script that calculates fibonacci numbers"
   

   Specify the model and provider:

   trae-cli run "Fix the bug in main.py" --provider anthropic --model claude-sonnet-4-20250514
   

2. interactive mode
   Go into interaction mode for multiple rounds of dialog and iterative development:

   trae-cli interactive
   

   The model and maximum number of steps can be specified:

   trae-cli interactive --provider openai --model gpt-4o --max-steps 30
   

   In interactive mode, the user can enter commands continuously and the Trae Agent will complete the task step-by-step according to the context.

3. Save operation log
   An operation log can be generated for each task execution for debugging purposes. By default, the log is saved as trajectory_YYYYMMDD_HHMMSS.json, or specify a file:

   trae-cli run "Optimize the database queries" --trajectory-file optimization_debug.json
   

4. Force Generated Patch
   For tasks that require file modifications, patches can be forced to be generated:

   trae-cli run "Update the API endpoints" --must-patch
   

5. Customizing the working directory
   Specify the project directory to perform the task:

   trae-cli run "Add unit tests for the utils module" --working-dir /path/to/project
   

Featured Function Operation

• Lakeview Summary
  The Lakeview feature of the Trae Agent provides a concise summary of the task steps. After running the task, view the log file (e.g. trajectory_20250612_220546.json) in the Lakeview field for a quick overview of the implementation steps.
• Multi-LLM Support
  Users can access the --provider cap (a poem) --model parameter to switch between different models. For example, using the OpenRouter of GPT-4o:

  trae-cli run "Optimize this code" --provider openrouter --model "openai/gpt-4o"
  

• tool ecology
  Trae Agent has built-in tools for file editing, bash execution, and more. For example, automatic file editing:

  trae-cli run "Add documentation to main.py"
  

  The tool generates the document and saves it to the specified file.

caveat

• Ensure that the API key is valid, otherwise the task cannot be executed.
• The project is in alpha and may be unstable, so we recommend following the GitHub repository for updates.
• The log file records detailed operations and is recommended to be cleaned regularly to save space.

application scenario

1. Automated Code Generation
   Developers simply type in "Create a REST API framework" and Trae Agent generates the code framework in Python or other languages, saving time on manual writing.
2. Debugging and fixing code
   Type "Fix error in main.py" and Trae Agent will analyze the code, locate the problem and generate a patch, which is suitable for fixing bugs quickly.
3. Research and development of intelligent bodies
   Due to the modular design, researchers can modify the architecture of Trae Agent to test new tools or workflows, suitable for academic research.
4. Batch Task Processing
   Call Trae Agent via scripts to perform tasks in batch, such as adding unit tests for multiple modules, to improve efficiency.

QA

1. What languages does Trae Agent support?
   Python 3.12 is primarily supported, but code generation and editing for other languages can be supported through the tool ecology.
2. How do I switch between different LLM models?
   utilization --provider cap (a poem) --model parameters, such as trae-cli run "任务" --provider openai --model gpt-4oThe
3. How are log files used?
   Log files record every step of a task. Viewing JSON files allows you to analyze the execution process, which is suitable for debugging or optimization.
4. Is the project suitable for a production environment?
   It is currently in alpha and is recommended for development and research, but should be tested with caution in production environments.