Overview
TRAE Work (formerly TRAE SOLO) is an AI agent platform under ByteDance, first released on 2026/03/31 under the name "TRAE SOLO". It represents a paradigm shift in AI programming tools from "assistance" to "dominance"—the core concept is "context engineering": integrating development tools into AI, rather than integrating AI into tools.
**On 2026/06/09, it was officially rebranded as "TRAE Work"**, with its positioning expanded from "developer-exclusive" to "more than code, great potential"—covering broader knowledge work scenarios beyond software development, including research (reading papers to produce summaries), data analysis (mining market data/trend analysis), and content creation (generating PPTs, developing multilingual websites). Users only need to propose tasks, and the AI automatically decomposes them and calls Skills and tools to execute, with users only needing final acceptance.
**On 2026/06/24, the Design mode was added**: a one-stop design capability covering the complete workflow from design draft generation, natural language batch modification, design system management, to design-to-code conversion. At the same time, **global memory** (automatically accumulating historical conversations into exclusive memory) and voice discussion with web search support were launched. TRAE Work adopts a dual-mode agent design with one-click switching between Work mode and Code mode, with desktop (Win/Mac) + web client synchronization, leveraging cloud computing power to support multi-task parallel background execution.
Key Features
- Brand Upgrade: SOLO → Work: Officially renamed on 2026/06/09, positioning expanded from "developer-exclusive" to a comprehensive AI work agent platform covering research/data/creation/development
- Design Mode (New): Launched on 2026/06/24, one-stop design workflow: design draft generation, natural language batch modification, design system management, design-to-code conversion
- Agent-Led Work Paradigm: AI is no longer a sidebar assistant but an intelligent hub that uniformly schedules the entire task flow, autonomously completing understanding requirements, decomposing tasks, and calling tools for execution
- Context Engineering: Organically integrates all contexts such as terminal, editor, documents, browser, Figma, etc., with the agent selecting the most suitable tool for each task
- Global Memory (New): When enabled, automatically remembers historical conversation context, accumulating into exclusive memory for reuse across tasks
- Multi-Task Parallel + Dual-Client Collaboration: Leveraging cloud computing power for multi-task background parallel execution, with desktop (Win/Mac) and web client synchronized for progress viewing
- Transparent and Traceable Execution Process: Aligns goals through structured information like Plans and to-do lists, with each step transparent and traceable, allowing intervention and adjustment at any time
Use Cases
- Non-developers quickly build MVPs or personal projects using natural language
- Study papers and automatically produce research summaries
- Research market/short video platform data and generate analysis reports or PPTs
- Developers delegate complex tasks to the Agent for independent completion
- Need Figma design-to-code conversion, or use Design mode for one-stop design to implementation
- Domestic teams use AI agents without proxy
Pros
- All features are available in the free version; payment only increases cloud task concurrency and fast-track priority
- Positioning expanded from pure programming to research/data/creation full scenarios, with broader application scope
- Strong Agent autonomy, capable of handling the complete process from requirements to deployment
- Fast direct connection domestically, excellent Chinese experience
- Design mode bridges design to code, reducing tool switching
- Backed by ByteDance, with rapid iteration (multiple versions released within June)
Pricing
TRAE Work's free version has all features available, supporting 2 simultaneous cloud tasks; paid plans are unified with the TRAE main site (Fast Track Pro from ¥59/month, Pro+ ¥239/month, Ultra ¥699/month, Express ¥1399/month). The core differences lie in "number of simultaneous cloud tasks" (up to 20) and "number of queue-skipping fast tracks". Higher payment yields higher task execution efficiency and priority, and priority access to SOTA models. Heavy users can also purchase the Ark Coding Plan for more computing power.
Summary
TRAE Work (formerly TRAE SOLO) is one of ByteDance's most groundbreaking AI agent products in 2026, redefining the collaboration between AI and users through "context engineering". The upgrade from "SOLO" to "Work" on 6/9 was not just a simple name change but an expansion of service boundaries from "AI programming" to "AI work"—emphasizing research, data analysis, content creation, and development equally; the new Design mode added on 6/24 further completed the design-to-code loop. Compared to Cursor's focus on IDE assistance and Claude Code's focus on terminal interaction, TRAE Work takes the route of AI fully taking over task execution, and its advantages of free basic features + direct domestic connection make it a strong choice for handling comprehensive knowledge work tasks in Chinese scenarios.
Version History
- Design Mode + Global Memory (2026/06/24): Launched one-stop Design mode (design draft generation/batch modification/design system management/code conversion); added global memory and voice discussion with web search
- TRAE SOLO Upgraded to TRAE Work (2026/06/09): Official brand rename, positioning expanded from "developer-exclusive" to a comprehensive AI work platform covering research/data/creation/development, supporting enterprise account login
- Builder/SOLO Builder Unified into Agent (2026/05/08): TRAE IDE merged Builder / Builder With MCP into Agent, and SOLO Builder / SOLO Coder into SOLO Agent, simplifying the naming system
- TRAE SOLO Initial Release (2026/03/31): Officially launched with Agent as the core, supporting dual-client collaboration, three-panel workspace, MTC/Code dual modes, proposing the "context engineering" concept