Trae review

Trae is ByteDance’s AI coding IDE, built on a Visual Studio Code foundation. It combines a conventional editor with AI help for writing and debugging code, plus an agent mode for planning and carrying out tasks across multiple files. It is aimed at developers who want code generation, testing and deployment support in the same workspace, rather than a separate chat window alone.

There are two working modes. IDE Mode adds inline assistance as you work. SOLO is for project-level tasks: it can plan and execute work across files. Trae also supports built-in and custom agents, with settings for rules, skills, memories and hooks. It can index a codebase for context, use web searches and external documentation, and turn screenshots into interface code. Other listed tools include predictive completions, agent-driven code review, and a browser preview with inspection and console debugging.

Trae’s appeal is the breadth of its coding workflow. IDE Mode covers inline assistance, while SOLO is designed for multi-file planning and execution. That gives teams a path from a small edit to a larger project task without switching modes of work. Codebase indexing, contextual references, web search and external documentation can also give agents more information than a prompt alone.

The feature set includes:

• Built-in agents, plus custom agents configured with rules, skills, memories and hooks.
• Screenshot-to-interface-code input, predictive completions and agent-driven code review.
• A browser preview with inspection and console debugging.
• Access to named AI models, with the option to add custom API endpoints.

The pricing page lists a free plan, which makes it possible to try Trae without a monthly subscription, though usage is limited. Reddit reactions are mixed rather than uniformly negative: one commenter said they still enjoyed the experience, and others described Trae as useful or kept it as a no-cost alternative. Those are individual reports, not evidence that its models will outperform your current editor.

The free plan is constrained: it allows Auto mode only, 5,000 Autocomplete completions per month and two concurrent cloud tasks, with a standard queue. Pro costs $20 per month and includes $20 of basic usage, Auto plus all models, unlimited Autocomplete and 10 concurrent cloud tasks. Pro+ is $60 per month with $60 of basic usage and 15 tasks; Ultra is $200 per month with $200 of basic usage, 20 tasks and early model access. The pricing page says annual billing saves 17%, but does not publish annual prices. It also does not clarify here how far the included basic usage goes for a particular workload.

Reddit users report concerns about model selection, data use and changing value. In one pricing thread, commenters described exhausting credits quickly, including one report of a balance running out after two requests. These are user reports, not a verified general limit; still, check the usage terms and model costs before paying. One commenter also said they left over data concerns, but that does not establish Trae’s data policy.

The pricing page lists four monthly plans: Free at $0, Pro at $20, Pro+ at $60 and Ultra at $200. That makes the first paid step $20, but moving from Pro to Pro+ triples the monthly price; Ultra costs ten times as much as Pro.

Free includes Auto mode, 5,000 autocomplete completions a month and two concurrent cloud tasks. Pro adds all models, unlimited autocomplete and 10 concurrent tasks. Pro+ raises the task limit to 15, while Ultra raises it to 20 and adds early model access. Each paid tier lists a matching amount of basic usage: $20, $60 or $200 a month.

The pricing page says yearly billing saves 17%, but it does not publish annual prices. No add-ons are listed. Check the usage details before paying: the published plan summary does not specify a broader usage quota, and “unlimited” applies to autocomplete, not necessarily every AI feature.

Trae is built on a Visual Studio Code foundation, so VS Code-oriented developers may find the environment familiar. But the published information does not confirm which extensions, settings or project data transfer directly. Check those details before treating a move as a simple install.

Trae indexes codebases and supports custom agents with rules, skills, memories and hooks. Those workflows could take effort to recreate elsewhere; the available information does not say whether custom-agent configurations can be exported. It also does not describe a code or conversation export tool. Trae supports custom API endpoints, which offers some model flexibility, but that alone does not remove workflow lock-in.

The published material includes no Reddit posts, so we can’t reliably say what users report as their reason for leaving. Before switching in either direction, verify data handling, export options and whether your agent setup can move with you. Those details are not published in the published information.