Trae Agent

Trae Agent is an open-source, LLM-based software engineering agent from ByteDance with more than 12,100 stars on GitHub, built to handle general-purpose development tasks autonomously from the command line.

Overview

Trae Agent is the open research-and-tooling counterpart of ByteDance’s Trae AI IDE. Given a task in natural language — fix a bug, add a feature, explain a module — it explores the repository, reasons about the relevant code, edits files, and runs tests to check its own work, iterating until the task is done. It is model-flexible by design: instead of being married to one vendor, it can be configured with different LLM providers, which makes it useful both as a practical coding assistant and as a testbed for studying how agent behavior changes with the underlying model. It also records execution trajectories, so researchers can analyze and learn from real agent runs. The project is MIT licensed and comes from the team behind one of the most widely used AI IDEs, which gives it an unusually production-informed design for an open-source agent. For developers who like the Trae experience in an IDE, the agent offers the same philosophy in a scriptable, terminal-first form.

Key Features

• Autonomous task execution – describes a goal once; the agent explores, edits, and verifies on its own.
• Multi-model support – configure different LLM providers and models instead of being locked to one.
• Full tool use – file editing, shell commands, and code search as part of the agent loop.
• Trajectory recording – capture complete run histories for analysis, debugging, and research.
• Terminal-first workflow – a CLI tool that fits scripts, servers, and headless environments.
• Backed by the Trae team – design informed by ByteDance’s production AI IDE.
• MIT licensed – permissive open source, free to use and modify.

💳 Pricing

Trae Agent itself is free and open source under the MIT license — you only supply your own LLM API keys. The related Trae IDE offers a free tier plus Pro plans for heavier usage; check trae.ai for its current pricing.

Who Is It For

Developers who want an open, model-agnostic coding agent in the terminal, and AI researchers who need recorded agent trajectories and a configurable harness for software-engineering benchmarks.