Trae review

Trae is a proprietary AI-native integrated development environment from ByteDance, built on a Visual Studio Code foundation. It is an editor application, not a coding library, standalone API or self-hosted server. The desktop IDE is available for macOS, Windows and Linux; the vendor also lists web, mobile and third-party-editor plugin clients.

Trae puts AI assistance inside the coding environment. IDE Mode provides inline help, while SOLO plans and executes multi-file project tasks. The IDE indexes codebases for contextual references, can use web searches and external documentation, and accepts screenshots for image-to-code workflows. Built-in agents can be supplemented with custom agents and sub-agents.

In a stack, Trae sits where an editor normally sits, with model access and agent workflows layered into the development loop. It supports custom model API endpoints and local MCP servers, but its backend infrastructure has no native self-hosting option. That makes it an editor choice with service and data-handling implications, not an independently deployable coding component.

Trae is primarily an IDE integration, not a dependency you call from application code. The vendor lists Visual Studio Code extension compatibility, Model Context Protocol (MCP) connections to custom servers, and configurable custom API endpoints for models. It also names direct tooling integrations with GitHub, Figma, Supabase, Lark/Feishu and Vercel. The homepage does not name a public Trae API, SDK, supported programming language list or CLI; confirm SDK and CLI support in the vendor’s docs before designing around them.

For day-to-day use, IDE Mode offers inline assistance and predictive CUE completions. SOLO handles multi-file plans and execution; agents can use rules, skills, memories and hooks, and call sub-agents. Developers can reference code with @ symbols, bring in web searches or external documentation, and use the browser preview for inspection and console debugging. Trae also lists agent-driven code review and screenshot-based interface generation.

The integration surface is therefore the editor, VS Code extensions, MCP and model endpoints. Check vendor documentation for the exact setup and compatibility requirements for each integration.

The Free plan costs $0 per month and includes Auto mode only, limited usage, 5,000 autocomplete completions per month, two concurrent cloud tasks and the standard queue. Pro costs $20 per month and includes $20 of basic usage, Auto plus all models, unlimited autocomplete, 10 concurrent cloud tasks and the fast queue. Pro+ is $60 per month with $60 of basic usage and 15 concurrent cloud tasks; Ultra is $200 per month with $200 of basic usage, 20 concurrent cloud tasks and early model access. Both include the features and fast queue listed for lower paid plans.

The listed monthly prices are in USD. The pricing page says yearly billing saves 17%, but does not publish annual prices. It lists no add-ons. The paid tiers’ basic-usage amounts are not a published overage schedule: Trae does not publish what happens when that usage runs out, how extra usage is billed or whether requests are throttled. Teams should confirm those terms and the model-specific usage costs before budgeting production work. Free is a useful evaluation tier, not an unlimited plan.

Trae is proprietary, closed-source software. The core IDE and backend infrastructure do not offer a native self-hosting option, so teams cannot deploy the full service under their own control. Enterprise deployments through BytePlus offer centralized governance, SSO and regional data-storage controls, according to the vendor’s stated feature set.

The vendor supports custom model endpoints and local MCP servers, which can give teams options for connecting existing infrastructure. Those options do not make the IDE or its backend open source. Trae’s published feature list does not describe a data-export path, so ask the vendor how to retrieve code, prompts, agent configuration and other account data before committing.

Migration away from the editor should be less involved than migration from a proprietary runtime: code remains in project files, and Trae supports standard VS Code extensions. Agent configuration and service-specific workflows may need to be recreated elsewhere. Verify which settings and data can be exported rather than assuming they transfer.

The plan limits that matter operationally are autocomplete allowances on Free and concurrent cloud-task caps across all tiers: two on Free, 10 on Pro, 15 on Pro+ and 20 on Ultra. Paid plans list basic-usage amounts, but Trae does not publish overage rates or what happens when an allowance is exhausted. The pricing page also does not publish service-level or rate-limit commitments. Confirm those points with the vendor before depending on agent tasks in a release workflow.

For security and compliance, review data handling, retention and model routing against your requirements. Reddit users raise concerns about code data, while other commenters question model quality and the value of changing plans; those are reports and opinions, not verified policy or performance findings. BytePlus enterprise options list SSO, centralized governance and regional storage controls. Ask what those controls cover and what support commitments apply.

Cloud-task concurrency is not a reliability guarantee. Treat agent output as code requiring review and testing, and keep a fallback editor and model workflow for outages or quota exhaustion.