“

Preface
As software delivery accelerates, developers often juggle between the CLI, scripts, tests, and documentation. Trae Agent empowers you to execute complex workflows—code edits, testing, deployments—using simple natural‑language commands, freeing up both your hands and your focus.

Trae Agent: Your AI‑Powered Automation Companion for Software Engineering

Introduction to Trae Agent

Trae Agent is an LLM‑driven agent designed to streamline everyday software engineering tasks. Whether you need to generate a script, fix a bug, write tests, or update documentation, just issue a natural‑language instruction:

trae-cli run "Generate a project README"

Key benefits include:

• Natural‑Language Interface
  Execute end‑to‑end workflows without memorizing shell commands.

• Multi‑Model Support
  Compatible with both OpenAI and Anthropic APIs—switch between models for precision or cost‑efficiency.

• Rich Toolset
  Built‑in file editing, Bash execution, sequential reasoning, and more to cover a breadth of development scenarios.

• Interactive Mode
  Enter a live REPL‑style session (trae-cli interactive) to iterate on tasks, view history, and invoke help.

• Configurable & Traceable
  Customize via JSON or env vars, and record detailed execution logs (trajectories) for debugging and audit.

Core Features at a Glance

Installation & Quickstart

Follow these four steps to get up and running:

1. Clone the Repository

   git clone <repository-url>
   cd trae-agent
   

2. Sync Dependencies

   uv sync
   

3. Set API Keys

   export OPENAI_API_KEY="your_openai_key"
   export ANTHROPIC_API_KEY="your_anthropic_key"
   

4. Run Your First Task

   trae-cli run "Create a Hello World script"
   

• Override any setting on the fly with flags such as --model, --provider, or --max-steps.

• Launch interactive mode:

  trae-cli interactive --provider openai --model gpt-4o
  

Configuration Details

Trae Agent reads settings in this priority:

1. Command‑Line Arguments
2. Configuration File (trae_config.json)
3. Environment Variables
4. Built‑In Defaults

Example trae_config.json:

{
  "default_provider": "anthropic",
  "max_steps": 20,
  "model_providers": {
    "openai": {
      "api_key": "your_openai_key",
      "model": "gpt-4o",
      "max_tokens": 128000,
      "temperature": 0.5,
      "top_p": 1
    },
    "anthropic": {
      "api_key": "your_anthropic_key",
      "model": "claude-sonnet-4-20250514",
      "max_tokens": 4096,
      "temperature": 0.5,
      "top_p": 1,
      "top_k": 0
    }
  }
}

• default_provider sets your primary LLM.
• max_steps caps the agent’s reasoning iterations.
• model_providers details each provider’s parameters.

Deep Dive: Trajectory Recording

Why Trajectories Matter

• Debugging: Rewind every decision, tool call, and LLM response.
• Audit & Compliance: Retain immutable logs for regulatory needs.
• Optimization: Analyze token usage, model efficiency, and tool frequency.

Trajectory File Structure

Trajectories are saved as JSON with fields like:

Sample Snippet:

{
  "timestamp": "2025-06-12T22:05:47.000000",
  "provider": "anthropic",
  "input_messages": [...],
  "response": {
    "content": "Here’s your script...",
    "tool_calls": [...]
  }
}

Enabling & Customizing Trajectories

• Default Filename:

  trae-cli run "Refactor module"
  # Generates trajectory_20250707_160500.json
  

• Custom Filename:

  trae-cli run "Fix main.py bug" --trajectory-file debug_log.json
  

Tool Ecosystem & Extensions

1. str_replace_based_edit_tool – View, create, and patch files.
2. bash – Run shell commands and scripts.
3. sequential_thinking – Break tasks into stages and reflect iteratively.
4. task_done – Flag completion and produce a concise summary.

“

Example:

trae-cli run "Add unit tests for utils module" --working-dir ./project

The agent will think through each step, write tests, execute them, and report results.

Frequently Asked Questions (FAQ)

How‑To: Start Recording Your Workflow

# 1. Clone & install
git clone <repository-url>
cd trae-agent
uv sync

# 2. Set up env vars
export OPENAI_API_KEY="..."
export ANTHROPIC_API_KEY="..."

# 3. Run & record
trae-cli run "Refactor database module" --trajectory-file db_refactor.json

Conclusion

Trae Agent bridges the gap between natural‑language instructions and robust development workflows. With multi‑model support, a versatile toolset, and transparent trajectory logs, you can automate routine coding tasks, speed up CI/CD integration, and maintain full traceability. Try Trae Agent today and focus on crafting innovative solutions while your AI companion handles the repetitive work.