Trae
Also known as: Trae, Trae AI, trae.ai, TRAE, TRAE SOLO, Trae IDE
AI native IDE from ByteDance, a VS Code fork with autonomous Builder and SOLO agents that scaffold and ship projects from natural language, plus free access to frontier models.
Trae is an AI native integrated development environment built by ByteDance, delivered as a fork of VS Code so existing extensions, themes, and keybindings carry over with no learning curve. Launched in early 2025, it has grown quickly, reporting well over a million monthly active users, helped by an aggressive free tier that hands developers frontier models at no cost. The pitch is simple: much of what paid editors like Cursor charge for, Trae offers free, funded by ByteDance's scale. It ships desktop apps for macOS and Windows and a browser based cloud version, and its core agent component has been open sourced.
Its headline is agentic. Builder Mode takes a natural language description and scaffolds a complete project, generating the file structure, writing frontend, backend, and configuration code, running terminal commands, and iterating when something breaks, rather than producing a single shot answer. SOLO mode, released as a standalone app in early 2026, goes further, running as a more autonomous agent that carries a task from requirements analysis through code, terminal commands, browser testing, and deployment, in a Code mode and a broader product mode. Background cloud tasks let several agent runs execute in parallel while a developer keeps coding locally.
Trae is strongly multi model, giving free and paid users switchable access to Claude, GPT, Gemini, DeepSeek, and ByteDance's own Doubao models without an API key, with regional versions adding local model providers. It supports the Model Context Protocol for connecting external tools, lets developers configure custom agents with their own tools, skills, and logic, and can work from design mockups, database schemas, and documents as well as code. Its own codebase index, CodeGraph, grounds suggestions, though users report it can lose track of structure on large repositories where more mature indexing still leads.
The important caveat is governance. Trae is cloud connected and routes code through ByteDance infrastructure, collects telemetry shared with ByteDance affiliates with multi year retention and only limited controls, and as of mid 2026 offers no team plan, single sign on, administrator controls, or compliance certifications. Reviewers consistently advise regulated, government, and enterprise teams handling sensitive code to choose a different tool, while noting Trae is a strong fit for individual developers, students, and rapid prototyping. Pricing is low and usage aware: a generous free tier, then five token metered plans from three to one hundred dollars a month, with the ten dollar plan the common default and higher tiers aimed at heavy agentic and parallel cloud workloads.
Vendor details
Canonical URL
https://trae.ai
Category
Coding agent
Subcategory
AI native IDE with autonomous Builder and SOLO agents (ByteDance)
Funding status
Trae is a first party product of ByteDance, the company behind TikTok, operated through a Singapore based subsidiary. It is not independently funded; it is backed directly by ByteDance's resources. Trae launched in early 2025 and reached more than one and a half million monthly active users and over six million registered users by late 2025. Its core agent component, Trae Agent, was open sourced in 2025.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Built as a VS Code fork, so the full VS Code extension, theme, and keybinding ecosystem transfers directly. Supports the Model Context Protocol for connecting external tools, interfaces, and data sources, and lets developers configure custom agents with specific tools, skills, and logic. Agents run terminal commands, edit files across a project, perform browser based testing, and can deploy, and Trae accepts multimodal input including design mockups, database schemas, and documents. Background cloud tasks run agent work in parallel. Available as desktop apps for macOS and Windows and as a browser based cloud IDE.
In practice
You want to build a prototype from one sentence. Trae's Builder and SOLO modes scaffold a full project, write the frontend and backend, run terminal commands, test in a browser, and can carry it to deployment.
You want frontier models without paying for API keys. Trae's free tier gives switchable access to Claude, GPT, Gemini, and DeepSeek inside a familiar VS Code style editor, with paid plans from three dollars a month.
You handle sensitive or regulated code. Trae routes code through ByteDance infrastructure with telemetry you cannot fully disable and offers no single sign on, admin controls, or certifications, so a private alternative is safer.
Sources & related URLs
Related / legacy domains
Capability coverage
7.0 / 14 capabilities · 50%
| Integrations & Tool CallingIs a VS Code fork so the full extension ecosystem transfers, supports the Model Context Protocol to connect external tools and interfaces, lets developers configure custom agents with their own tools and skills, and runs terminal commands, browser tests, and multimodal inputs like design mockups and schemas, broad named integrations with genuine action. | Full |
|---|---|
| Workflow OrchestrationShips autonomous agents in Builder Mode, which scaffolds a whole project and iterates on errors, and SOLO mode, which runs as a more autonomous agent carrying a task from requirements through code, terminal commands, browser testing, and deployment, with parallel background cloud tasks, genuine end to end orchestration. | Full |
| Knowledge Grounding & RAGGrounds suggestions with its own codebase index called CodeGraph and accepts project context and multimodal inputs, a real context capability, though reviewers report it can lose track of structure on large repositories and that its indexing is less deep than the most mature competitors, so grounding is present but not a standout. | Partial |
| Human Oversight & GuardrailsSplits a controlled editor mode from an agent mode, shows diffs and previews, and gates the agent so it seeks the developer's approval on direction before spending usage, real propose and approve oversight, though not a comprehensive runtime guardrail enforcement engine. | Partial |
| Security, Identity & GovernanceOffers no team plan, single sign on, administrator controls, or compliance certifications as of mid 2026, and reviewers explicitly state it is not ready for teams that need those, so a first class security, identity, and governance capability is not documented. | Unable to verify |
| Observability & AuditabilityProvides in editor diffs, live previews, tool panels, and visibility into concurrent cloud tasks and plan level usage, real process visibility, though per interaction cost is hard to read and there is no comprehensive agent execution tracing, audit log, or analytics suite. | Partial |
| Memory & State PersistenceMaintains context within a session and a codebase index, but long term context memory is a documented weakness, with the agent losing track on large repositories, and no distinct persistent agent memory or checkpoint capability is documented as first class. | Unable to verify |
| Deployment & Data ResidencyIs cloud connected and routes code through ByteDance infrastructure with no self hosted, on premises, or air gapped option, so there is no data residency capability under the customer's control, and reviewers point sensitive teams to self hosted alternatives instead. | Unable to verify |
| Prebuilt Agents, Templates & PacksLets developers configure custom agents with specific tools, skills, and logic to build repeatable workflows, real custom agent scaffolding, though not a broad browsable library or marketplace of prebuilt agents, templates, and packs to adopt off the shelf. | Partial |
| Triggers & Channel CoverageReaches developers through desktop apps for macOS and Windows, a browser based cloud IDE, and a standalone SOLO app, with background cloud tasks running agent work in parallel, real channel coverage that is developer invoked during coding rather than driven by broad external event triggers. | Partial |
| Model Flexibility & RoutingGives switchable access without an API key to Claude, GPT, Gemini, DeepSeek, and ByteDance's own Doubao models, with regional versions adding further providers and user supplied keys, genuine multi provider model flexibility and in editor routing across frontier and open models. | Full |
| APIs, SDKs & MCP ExtensibilitySupports the Model Context Protocol as a first class integration surface, lets developers build custom agents, and has open sourced its core agent component for deeper extension, a strong extensibility surface, though a dedicated public application programming interface and software development kit for the product are not clearly documented. | Partial |
| Testing, Debugging & OptimizationIterates on errors while building, offers intelligent bug troubleshooting, and in SOLO mode performs browser based testing of its own work and adjusts on intermediate results, a real test and debug loop inside the agent rather than a dedicated standalone testing or code quality product. | Partial |
| Browser & Computer UseRuns terminal commands, executes code, performs browser based testing, and can carry a project through to deployment within its agent modes, real code execution and computer use, though the browser use serves testing and building rather than being general autonomous browser automation as the core product. | Partial |
Pricing
Free tier with frontier model access and limited SOLO. Paid plans are Lite at 3 dollars, Pro at 10 dollars, Pro plus at 30 dollars, and Ultra at 100 dollars a month, all token metered by Basic and Bonus usage.
Flat monthly subscription tiers with token metered usage inside each. AI work draws down Basic and Bonus token credits rather than a fixed request count, so heavier context, stronger models, and more agentic execution consume the allowance faster and push a user toward a higher tier.
Included quota
Free: about five thousand autocompletions a month, limited Basic usage, two concurrent cloud tasks, limited SOLO. Lite at three dollars adds reliability but keeps two concurrent cloud tasks. Pro at ten dollars includes SOLO, unlimited autocomplete, fast queue priority, and ten concurrent cloud tasks. Pro plus at thirty dollars raises this to fifteen concurrent tasks, and Ultra at one hundred dollars to twenty, for heavy agentic and parallel workloads. All paid tiers meter AI work as Basic plus Bonus token usage.
What is public
Public: the five tier prices, the free tier contents, token based Basic and Bonus usage, and concurrent cloud task limits per tier. Not public: exact token consumption per task, and any enterprise pricing, which does not exist.
Billing mechanics
Flat monthly tiers from three to one hundred dollars, with a free tier, each metering AI work as Basic plus Bonus token usage and adding more concurrent cloud tasks at higher tiers. There is no enterprise or per seat plan.
Cost watchouts
Usage is token metered, so agentic runs, large context, and stronger models burn the allowance faster than simple autocomplete, and per interaction cost is hard to see while working. Heavy users may need Pro plus or Ultra for concurrent cloud tasks. Old guides citing fixed fast request counts are outdated.
Variable cost rationale
Billing is token metered on Basic and Bonus usage, so cost tracks how much context, model power, and agentic execution each workflow consumes, and heavier autonomous runs burn the allowance quickly. Although tiers are flat monthly prices, exhausting usage forces a tier upgrade, so effective spend scales with activity. Exposure is high and, by common report, hard to predict.
Additional watchouts
Token metered cost is hard to predict, and the biggest watchout is not price but governance: no team plan, single sign on, admin controls, or certifications, plus ByteDance telemetry with multi year retention, make it unsuitable for sensitive or regulated code regardless of cost.
Overage / add-ons
When a tier's Basic and Bonus token usage is exhausted, work is limited or slowed until it resets or the user upgrades. There is no separate per token overage purchase documented; the model is to move up a tier.
Sales call required
No — self-serve available
Free / trial
A permanently free tier includes about five thousand autocompletions a month, limited usage, two concurrent cloud tasks, limited SOLO mode, and access to premium models like Claude, GPT, and DeepSeek without an API key. The Pro plan adds a seven day free trial.
Lowest paid plan
Lite at three dollars a month is the lowest paid plan, above a free tier. The main developer plan is Pro at ten dollars a month.
Commercial notes
Trae undercuts paid rivals, offering frontier models and an autonomous agent free or cheaply, with Pro at ten dollars positioned as half the price of comparable competitors. The catch is that value is capped by token usage and, more importantly, by the absence of enterprise governance, which pushes serious teams elsewhere despite the low price.
Key ambiguities
Because billing is token based, the real monthly cost depends on model choice, context size, and how much agentic and cloud execution a workflow uses, which is hard to predict from the plan price alone. There is no team or enterprise tier, so per seat and admin pricing do not exist.
Cancellation / refund
Self serve monthly plans can be upgraded, downgraded, or cancelled by the user. There is no enterprise contract. A 7 day free trial applies to the Pro plan.
Missing data
Exact token consumption per task and per model, and any enterprise or team pricing, are not published, since no enterprise tier exists.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- Aider — Open-source, model-agnostic terminal coding agent that works…
- Anthropic Claude Code — Anthropic's terminal and IDE-native coding agent with sub-agent…
- Augment Code — AI coding assistant evolving into an AI-native engineering platform
- Blackbox AI — AI coding assistant for code generation and developer workflows
- Blink — Agentic AI app builder that turns a natural language prompt into a…