Trae
Also known as: TRAE SOLO, Trae IDE, TraeCode, TraeWork
ByteDance's AI coding platform. The all new TRAE joins an AI native IDE window with an Agent window for background, cloud and scheduled tasks, with custom agents, subagents, memories, marketplace plugins, browser use and built in or custom models. The enterprise offering comes through BytePlus.
Trae is ByteDance's AI coding platform. In October 2026 TraeCode and TraeWork merged into the all-new TRAE, which pairs an AI native IDE window with an Agent window. It supports custom agents, subagents, skills, rules, global and project memories, automatic codebase indexing, MCP servers and plugins from a Marketplace, browser use in a built in browser or linked Chrome, approval modes, and built in and custom models. The Agent window runs tasks in the background and in the cloud, and automations run them on a schedule.
Enterprise customers are served through BytePlus. The enterprise offering includes role based permission management, SOC 2 Type 2 certification, zero retention of code with no model training on it, org wide spending caps with live consumption, and visibility into member activity and tool usage, with the option to plug in custom enterprise models. Delivery is SaaS, with a VPC option announced as coming soon.
Individual plans are monthly or annual tiers with a dollar usage allowance above a free Auto mode tier.
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 subsidiary based in Singapore. 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
first party product
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
TraeCode supports MCP servers inside agents (three transport types), editor extensions, custom agents with skills and rules, and codebase indexing. TraeWork runs on desktop, web and mobile.
In practice
You want an agent to carry work in the background. TraeWork runs several tasks at once and returns the output for you to review, on desktop, web or mobile.
You want to choose the model. TRAE offers built in frontier models and lets enterprises plug in their own custom models.
You need enterprise controls. TRAE's enterprise offering through BytePlus includes role based permissions, SOC 2 Type 2, zero code retention and org wide spending caps, and a VPC option is announced as coming soon.
Sources & related URLs
Research sources
Agentic Index coverage score
11.0 / 14 capabilities · 79%
| Integrations & Tool Calling | Full |
|---|---|
|
MCP servers plug into TraeCode agents over three transport types, and customers create and manage custom agents with their own skills, so agents take tools the customer adds. TraeWork calls on the right tools for each task and keeps all of a project's files in a single Workspace. TRAE's documentation lives at docs.trae.ai. SourceTRAE, trae.ai/solo and docs.trae.ai/ide/use-mcp-servers-in-agentsread 2026-10-06 |
|
| Workflow Orchestration | Full |
|
TraeWork takes a defined task, breaks it down on its own and calls the right tools to get it done, running multiple tasks simultaneously in the background. The tasks run in the cloud, with no local resource limits, so work carries on while the user is offline. TraeCode supports subagents beside custom agents, so work is split across agents. SourceTRAE, trae.ai/solo and docs.trae.ai/ide/subagentsread 2026-10-06 |
|
| Knowledge Grounding & RAG | Full |
|
Automatic codebase indexing builds an index of the workspace and keeps it current as the code changes, with a manual rebuild, so the agent draws on a maintained index rather than gathering context per run. Whether the index persists between sessions is not stated outright. Sourcedocs.trae.ai/ide/codebase-indexingread 2026-10-25 |
|
| Human Oversight & Guardrails | Full |
|
In Manual Approval mode TRAE shows an approval card before the agent touches files outside scope or runs dangerous commands, and a custom mode can require approval of MCP calls. Enterprises can also set org wide spending caps. Automatic Approval is not offered in the US. Sourcedocs.trae.ai/ide/permission-and-approvalread 2026-10-25 |
|
| Security, Identity & Governance | Full |
|
Access is governed by 'Role-based permission management to prevent unauthorized access', with member activity and tool usage visible across the organization. The product is also 'SOC 2 Type 2 certified', with zero retention of code and no model training on it. Sourcetrae.ai/enterpriseread 2026-09-29 |
|
| Observability & Auditability | Partial |
|
Automated tasks keep a run history, and the enterprise console shows member activity, tool usage and spend across the organization in aggregate. No retained step by step record of what an agent did in each run is described. Sourcedocs.trae.ai/ide/automated-tasks and trae.ai/enterpriseread 2026-10-25 |
|
| Memory & State Persistence | Full |
|
Memories come in global and project scopes, stored at stated file paths on the user's machine. The AI creates and updates them as it works, and users can edit or delete any memory. Sourcedocs.trae.ai/ide/memoriesread 2026-10-25 |
|
| Deployment & Data Residency | Not documented |
|
The enterprise offering is 'SaaS or VPC (coming soon)', so TRAE is sold as a hosted service and the VPC option is announced but not shipped. No hosting region is published. TraeWork runs its tasks in TRAE's cloud rather than on the user's machine. Sourcetrae.ai/enterpriseread 2026-09-29 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
Customers install plugins that bundle MCP servers and skills from TRAE's Marketplace, and automations start from preset templates, beside the built in Builder agent and the customer's own agents. Sourcedocs.trae.ai/ide/marketplaceread 2026-10-25 |
|
| Triggers & Channel Coverage | Full |
|
Automation runs agent tasks at a fixed time or on a recurring interval, in the cloud or locally, with a run history, so work starts without a person invoking it. Scheduled tasks arrived with v3.6.0 on 8 October 2026. Sourcedocs.trae.ai/ide/automated-tasks and changelogread 2026-10-25 |
|
| Model Flexibility & Routing | Full |
|
The enterprise offer is to 'Access world-class frontier models or plug in your custom enterprise models', and TraeCode supports built in and custom models, so the customer picks the model. BytePlus explains how to bring its ModelArk models into TRAE. Sourcetrae.ai/enterpriseread 2026-09-29 |
|
| APIs, SDKs & MCP Extensibility | Partial |
|
Editor extensions and the MCP servers its agents call let TraeCode reach out to other tools. TRAE names no API or SDK for driving it from outside, and does not say how other systems could start or steer its agents. Sourcedocs.trae.ai/ide/manage-extensionsread 2026-09-29 |
|
| Testing, Debugging & Optimization | Not documented |
|
TRAE describes no way for a customer to test or evaluate an agent's work before relying on it. It has published a benchmark of its own agent, which customers cannot run themselves. In TraeWork, users review each result in a live preview and request revisions. SourceTRAE, trae.ai/solo; docs.trae.airead 2026-10-06 |
|
| Browser & Computer Use | Full |
|
Browser use lets the agent click and fill forms in TRAE's built in browser or a linked Chrome. The plain chat agent cannot use it. Sourcedocs.trae.ai/ide/browser-useread 2026-10-25 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
The Agent window, launched in TRAE v3.6.0, is a workspace for orchestrating multiple agents across projects from one window, with cloud tasks, scheduled tasks and one place to manage outputs such as code, documents, presentations, images and video. The Agent window can also connect to a directory on a remote host and run tasks there, and users can switch between it and the IDE window with data kept in sync.
Bears on: Deployment / data residency
View sourceTrae has integrated with the BytePlus ModelArk Coding Plan. Users can configure Trae to use the BytePlus Plan service provider, granting direct access to a suite of foundation models including DeepSeek, Kimi, GLM, and ByteDance's Seed models.
Bears on: Integrations
View sourcePricing
Free tier in Auto mode. Pro is $20 a month ($200 a year), Pro+ $60 and Ultra $200, each with a matching Dollar Usage allowance. Lite at $8 is sold only in selected countries, and existing subscribers keep their earlier prices.
Flat monthly or annual tiers, each carrying a Dollar Usage allowance equal to its price that model calls draw down at model specific rates, with pay as you go On-Demand Usage in $3 steps once the allowance is spent.
Included quota
Free: Auto mode, limited autocompletion and Dollar Usage, 2 cloud tasks. Lite ($8, selected countries): $8 of Dollar Usage. Pro ($20): $20. Pro+ ($60): $60. Ultra ($200): $200. Annual plans cost ten times the monthly price.
What is public
The docs plans page publishes the new tier prices, their Dollar Usage allowances, the countries where Lite is sold and the On-Demand Usage step. Enterprise pricing through BytePlus is not public.
Billing mechanics
Monthly or annual tiers from $20 (Lite at $8 regionally), each with a Dollar Usage allowance and On-Demand Usage beyond it, plus a free Auto mode tier. Enterprise terms are arranged through BytePlus.
Cost watchouts
Agentic runs, large context and stronger models spend the Dollar Usage allowance faster, and On-Demand Usage then bills in $3 steps. Some models (Seed, MiniMax, GLM) and Automatic Approval are not available in the US. Guides citing $3 to $100 tiers or token credits are outdated.
Variable cost rationale
Model calls draw down a dollar allowance at model specific rates, and On-Demand Usage bills beyond it in $3 steps, so cost tracks context, model power and agentic and cloud execution, and heavy autonomous runs spend the allowance quickly.
Additional watchouts
Enterprise controls (role based permissions, SOC 2 Type 2, zero code retention, spending caps) sit on the enterprise offering through BytePlus, and its VPC deployment is listed as coming soon. The individual plans do not list those controls.
Overage / add-ons
Once the Dollar Usage allowance is spent, On-Demand Usage bills pay as you go in $3 steps, or work waits for the next cycle.
Sales call required
No, self serve available
Free / trial
A free tier runs in Auto mode with limited autocompletion, limited Dollar Usage and 2 cloud tasks, without SOLO mode. Claude is available only with a custom key. No trial is offered on the new plans.
Lowest paid plan
Pro at $20 a month ($200 a year) with $20 of Dollar Usage. Lite at $8 is sold only in selected countries, not the US.
Commercial notes
Trae undercuts paid rivals, offering frontier models and an autonomous agent free in Auto mode or from $20 a month on Pro. An enterprise offering sold through BytePlus adds role based permissions, SOC 2 Type 2, zero code retention and org wide spending caps, priced on request.
Key ambiguities
Real monthly cost depends on model choice, context size and agentic and cloud execution. The pricing page's server fallback and older billing data still show the legacy $3, $10, $30 and $100 prices, which apply only to existing subscribers, and the pricing page and docs disagree on whether the free allowance is one time or monthly.
Cancellation / refund
Self serve monthly plans can be upgraded, downgraded, or canceled by the user. Enterprise terms are arranged through BytePlus. A 7 day free trial applies to the Pro plan.
Missing data
Per model usage rates in practice, and enterprise pricing.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- 10Web — Agentic website platform whose AI agents build, host and manage…
- AgentUI — Managed AI app builder for operations teams that generates and hosts…
- Aider — Open source, model agnostic terminal coding agent that edits your…
- Amp — Amp is a coding agent from Amp Frontier Corporation that runs each…
- Anthropic Claude Code — Anthropic's agentic coding system across terminal, desktop, IDE, web…
Alternatives to Trae
The closest documented capability profiles to Trae among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Kilo Code13.0 / 14Adds documented Deployment & Data Residency and Testing, Debugging & Optimization
- Verdent11.0 / 14Adds documented Testing, Debugging & Optimization
- Cognition13.5 / 14Adds documented Deployment & Data Residency and Testing, Debugging & Optimization
- GitHub Copilot13.5 / 14Adds documented Deployment & Data Residency and Testing, Debugging & Optimization
- Kiro13.5 / 14Adds documented Deployment & Data Residency and Testing, Debugging & Optimization
- Refact.ai12.5 / 14Adds documented Deployment & Data Residency and Testing, Debugging & Optimization
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded