Trae

Trae is ByteDance's free AI-native IDE built on a VS Code fork. It ships with built-in AI chat, inline completions, and an autonomous Builder mode at zero cost. For developers priced out of Cursor or Windsurf, it's a serious alternative.

The pitch for Trae is simple enough that it barely needs explaining: it's a VS Code fork with AI baked in, and it costs nothing. ByteDance released it in late 2024 and has been iterating quickly since. If you've been watching the Cursor-vs-Windsurf debate from the sidelines because $20 a month feels like a lot for a tool you're not sure will stick, Trae is the obvious place to start.

That said, free is not the same as good. This review covers what Trae actually delivers, where ByteDance's backing is a legitimate concern, and who should consider it over the paid alternatives.

Quick verdict

Trae is a credible AI IDE for individual developers who aren't ready to pay for Cursor or Windsurf. The core loop, write code, ask the AI to extend or fix it, review the diff, is functional and fast. The Builder mode handles multi-step tasks well enough that it doesn't feel like a stripped-down version of a paid product. The main trade-offs are fewer polish details than Cursor and the unavoidable fact that ByteDance is the company holding your code during AI processing. If neither of those bothers you, Trae is worth trying immediately.

What Trae is

ByteDance launched Trae in November 2024 as an AI-native IDE for developers. It's built on a VS Code fork, which means your existing extensions, keybindings, and project files work without reconfiguration. The company frames it as their entry into the developer tools market, which in 2025 and 2026 means competing with Cursor, Windsurf, and to a lesser extent Claude Code.

The feature set covers the same ground as Cursor: inline tab completions, a chat panel with codebase indexing, and a multi-file agent mode called Builder. The underlying models have shifted over time. At launch Trae used proprietary ByteDance models. By early 2026 it ships with a combination of hosted models plus the option to bring your own API key for Claude Sonnet, GPT-4o, or others. That flexibility is meaningful: you're not locked into whatever ByteDance decides to run on the backend.

MCP support arrived in early 2026, which matters for anyone building workflows that connect the IDE to databases, external APIs, or custom tools. The ability to wire in an MCP server for Postgres schema inspection or Jira ticket lookup puts Trae in the same capability tier as Cursor on that front.

The VS Code fork angle is worth taking seriously. Cursor has been at it since 2022 and has had time to smooth over the rough edges of the fork. Trae is newer and it shows occasionally: extension compatibility is slightly narrower, and some VS Code features that rely on internal APIs behave differently. For most workflows this is invisible. For specific extension-heavy setups, it's worth testing before you commit.

The features that matter

Inline completions and chat

The tab completion in Trae is fast. ByteDance has put real engineering behind the latency, which is the thing you notice most in day-to-day use. Completions appear before you've stopped typing, and multi-line suggestions are common rather than rare. The quality is good, not quite at the level of Cursor's fine-tuned Tab model after months of learning your project, but close enough that most developers won't feel shortchanged.

The chat panel indexes your codebase on first open and stays current as you edit. You can ask it where authentication happens, what a function returns, or where a particular pattern is used, and get referenced answers rather than hallucinated ones. The indexing is fast on projects under 100k lines; larger monorepos take longer on first run but update incrementally after that.

Builder mode

Builder mode is Trae's version of Cursor's Agent mode or Claude Code's autonomous task execution. You give it a natural-language task, it plans a set of file changes, shows you the plan, and then executes with your approval. The plan-and-approve flow is the right safety model for anything touching production code: you catch conceptual mistakes before they propagate.

In practice, Builder mode handles well-scoped tasks reliably. "Add an email validation field to the registration form, update the backend validation endpoint, and add a test" produces a coherent set of diffs across three or four files that you can review and accept. Larger tasks with ambiguous scope need more hand-holding, same as any agentic tool. The failure mode when it goes wrong is usually to make changes in the right direction but with wrong assumptions about your data model rather than to produce completely broken code.

Multi-model support

This is a feature Trae handles better than you'd expect from a free tool. In settings, you can add API keys for Anthropic, OpenAI, or other providers. Once configured, you can route different interaction types to different models: fast completions to a cheap model, complex Builder tasks to Claude Sonnet or GPT-4o. This is smart product design. ByteDance absorbs the cost of their own hosted model for routine use while letting power users upgrade to frontier models on their own dime.

MCP connections

Model Context Protocol support means Trae can talk to external tools through MCP servers, just as Claude Code does. The setup is a config file pointing at your MCP servers. Once it's running, you can ask Trae to introspect your database schema while writing a migration or reference an issue tracker while implementing a feature. For developers already in an MCP-connected workflow with Claude Code, Trae fits in without much friction.

The ByteDance question

You can't review Trae without addressing this. ByteDance is a Chinese company subject to Chinese law, which requires companies to provide data to government authorities on request. The same concern has followed TikTok for years, and Trae operates in the same environment.

The practical question is what data Trae actually handles. Like any cloud-based AI IDE, it sends code to servers for processing. Unlike Cursor, which processes your code through Anysphere's servers in the United States, Trae routes it through ByteDance's infrastructure.

For personal projects, open-source work, and most early-stage startups, this is a judgment call that reasonable people make differently. For anything involving IP-sensitive proprietary code, financial systems, healthcare data, or US government contracts, the decision should involve your legal and security teams, not just a developer preference.

ByteDance has not published detailed enterprise compliance documentation for Trae comparable to Cursor's SOC 2 report. That gap matters if your organization has formal data handling requirements.

Pricing, what you actually pay

Trae is free. That's not "free tier with a hidden cap"; as of May 2026, there is no paid tier. ByteDance has not announced any intention to charge for the product, though that could change. The model is similar to how ByteDance has subsidized other developer tools to build market position.

This contrasts sharply with the rest of the market. Cursor costs $20 a month for Pro. Windsurf is similar. Codeium has a free tier but limits agentic features. If you're purely optimizing for cost, Trae is the obvious choice.

The risk is that free pricing at scale requires a business model at some point. ByteDance might introduce a paid tier, might start using your code for training, or might keep it free indefinitely as a strategic investment in developer mindshare. None of those outcomes are certain, which means building a deep workflow dependency on Trae carries some longevity uncertainty that building on Cursor doesn't.

Where Trae wins and where it doesn't

Trae wins on cost and on surprising functionality-for-free. The Builder mode genuinely works. The multi-model support is smarter than most free tools bother to be. If you're a student, an independent developer, or a founder in the early stages of a project, Trae removes the cost barrier entirely.

It falls behind Cursor on polish. Tab completion is fast but slightly less accurate on project-specific patterns, because Trae doesn't have the same training feedback loop that Cursor has accumulated from millions of paying users. The diff review UI is functional but not as refined. Extension compatibility has occasional gaps. None of these are deal-breakers individually; combined, they add up to a slightly rougher day-to-day experience.

The community and documentation gap is also real. Cursor has a large user base producing tutorials, .cursorrules templates, and workflow guides. Trae is newer and the surrounding ecosystem is thinner. If you hit a problem, you're more likely to find a Cursor answer on Stack Overflow or Reddit than a Trae answer.

Who should use Trae

The clearest use case is an individual developer who wants a modern agentic IDE but isn't ready to pay $20 a month before knowing whether the workflow actually fits. Trae gives you a genuine evaluation environment, not a stripped-down trial that hides the real features behind a paywall.

Students and open-source contributors are a natural fit. The zero-cost entry point and VS Code compatibility mean you can adopt it without changing your existing project setup or toolchain.

Teams where some developers are in markets with tight budgets will find Trae useful as a way to extend AI coding access across the team without a per-seat cost that scales poorly.

The developers who probably shouldn't use Trae as their primary tool: anyone with compliance requirements around code data handling, teams that need proven enterprise support contracts, and engineers who want the most polished and actively improved agentic IDE. For those profiles, Cursor or Claude Code is a more defensible choice.

Trae vs the alternatives

Trae vs Cursor

Cursor costs $20 a month and is more polished. Tab completion is better trained on project patterns, the diff review workflow is smoother, and the community resources are vastly larger. Cursor also has SOC 2 compliance documentation and a clear US-based data handling story.

Trae is free and good enough for most individual development work. The decision comes down to whether $20 a month matters to you and whether ByteDance's data handling is a concern for your use case.

Trae vs Codeium

Codeium also has a strong free tier for individual developers, but its agentic features are gated behind the paid Windsurf tier. Trae's Builder mode gives you more autonomous capability for free. If you specifically want the agent-mode workflow without paying, Trae is the stronger choice as of mid-2026.

Trae vs Claude Code

Claude Code is a terminal-native agent, not an IDE. They're solving different problems. If you live in a GUI editor, Trae and Claude Code aren't direct competitors. Some developers run both: Claude Code for large cross-cutting tasks orchestrated from the shell, Trae for the moment-to-moment editing experience. That combination covers both ends of the spectrum at a combined cost of $17 a month for Claude Pro plus zero for Trae.

Getting started

Download Trae from trae.ai. The installer detects your existing VS Code setup and imports settings, extensions, and themes. First launch asks you to sign in with a ByteDance or Google account.

Open an existing project. Trae will index the codebase in the background. While it does, open Settings and add your Anthropic or OpenAI API key if you want to route complex tasks to a frontier model. Set that model as your default for Builder mode, and leave the faster hosted model for completions.

Try the chat panel with a question about your codebase, then try Builder mode with a small scoped task: adding a field to a form and updating the backend validation to match. Watch the plan, review the diffs, accept them, run your tests. If that loop feels natural, you've found a free AI IDE worth keeping.

The bottom line

Trae is a better free AI IDE than most developers expect from ByteDance's first attempt at a developer tool. The core features work, the MCP support is genuinely useful, and the multi-model flexibility is smart product design. For individual developers not ready to pay for Cursor, it's worth a serious try.

The caveats are real: ByteDance's data handling deserves scrutiny, the long-term pricing stability is uncertain, and the polish gap with Cursor is noticeable if you've used both. But as a starting point for exploring agentic coding IDEs without a financial commitment, Trae earns its place on the shortlist.

Key features

• Built-in AI chat and inline completions at no cost
• Multi-file Composer-style agent for cross-file edits
• Supports multiple models including Claude and GPT-4o via settings
• VS Code extension compatibility out of the box
• Builder mode for autonomous task execution
• Codebase indexing for context-aware chat
• MCP server support for external tool connections

Pros and cons

Who is Trae for?

• Individual developers who want agentic AI coding without a monthly subscription
• Students and open-source contributors on tight budgets
• Teams evaluating AI IDEs before committing to a paid tool
• Developers in regions where Cursor pricing is a barrier

Alternatives to Trae

If Trae isn't quite the right fit, the closest alternatives are cursor , windsurf , claude-code , and codeium . See our full Trae alternatives page for side-by-side comparisons.