The era of the ‘Copilot’ is officially getting an upgrade. ByteDance is changing the goals. While tech companies have spent the past two years becoming comfortable with AI which suggests codes or emails, the ByteDance Team has been moving them. They released DeerFlow 2.0, a newly open-sourced ‘SuperAgent’ framework that doesn’t just suggest work; it executes it. DeerFlow was designed to automate the creation of video, slide decks, websites and code.
Sandbox, An AI With Its Own Computer
DeerFlow’s approach to execution is what sets it apart. The majority of AI agents are restricted to a text box interface. They send queries to APIs and return a string. If you want that code to run, you—the human—have to copy, paste, and debug it.
DeerFlow changes the script. DeerFlow operates in a Real Docker Container.
The implications for software developers are huge. This isn’t an AI ‘hallucinating’ that it ran a script; it is an agent with a full filesystem, a bash terminal, and the ability to read and write actual files. When you give DeerFlow a task, it doesn’t just suggest a Python script to analyze a CSV—it spins up the environment, installs the dependencies, executes the code, and hands you the resulting chart.
By providing the AI with its own ‘computer,’ ByteDance team has solved one of the biggest friction points in agentic workflows: the hand-off. DeerFlow’s stateful memory, persistent filesystem and ability to remember preferences and writing styles across sessions is possible because it uses a persistent and stateful system.
Multi-Agent Operation: Conquer first, then Divide.
The ‘magic’ of DeerFlow lies in its orchestration layer. It uses a SuperAgent harness—a lead agent that acts as a project manager.
When a complex prompt is received—for example, ‘Research the top 10 AI startups in 2026 and build me a comprehensive presentation‘—DeerFlow doesn’t try to do it all in one linear thought process. DeerFlow uses task division instead.
- Lead agent Break the task down into sub-tasks.
- Sub-agents They are spawned simultaneously. Another might be responsible for competitor research, while a third could generate images.
- Convergence: The results of the tasks completed by the subagents in the respective sandboxes are sent back to the main agent.
- Final Delivery Final agents compile the data and deliver a final product, like a web app or slide show.
This parallel processing significantly reduces the time-to-delivery for ‘heavy’ tasks that would traditionally take a human researcher or developer hours to synthesize.
Full Stack Automation from Research Tool
DeerFlow was not originally designed to have this much functionality. DeerFlow began at ByteDance, as a tool for deep research. As the community within ByteDance began to use it, the capabilities of the tool were pushed.
The Docker execution enabled users to create automated data pipelines and real-time dashboards. They could even build full-scale websites from scratch. ByteDance, realizing that users wanted an engine to execute code rather than a simple search tool for the web, rewrote its framework completely.
DeerFlow is a flexible framework capable of handling:
- Deep Web Research Retrieving cited sources from the web.
- Create Content: Generating reports integrating charts, videos, and images.
- Code Execution Run bash and Python commands within a safe environment.
- Asset Generation Create slide decks that include all UI and complete components.
The Key Takeaways
- Execution-First Sandbox: DeerFlow, unlike traditional AI agents operates in an isolated environment. Sandbox based on Docker. It gives it a full filesystem with a bash console, the ability to actually run and execute commands, rather than simply suggesting them.
- Hierarchical multi-Agent orchestration The framework uses a ‘SuperAgent’ lead to break down complex tasks into sub-tasks. It is a spawner parallel sub-agents to handle different components—such as scraping data, generating images, or writing code—before converging the results into a final deliverable.
- The ‘SuperAgent’ Pivot: Originally a tool to do deep research, DeerFlow 2.0 has evolved into a powerful marketing and sales tool. Rewritten from scratch It is now a harness that can be used for any task. Now, it can build complex data pipelines and full stack web applications.
- Complete Model Agnosticism: DeerFlow was designed with the user in mind. LLM-neutral. The OpenAI API allows engineers to switch between GPT-4 models, Claude 3.5 models, Gemini 1.5 models, and even local models using DeepSeek or Ollama, without having to change the agent logic.
- Stateful Memory & Persistence: Agent features The memory system is a persistent memory It can track user preferences, writing style, and context of the project across sessions. This allows it to function as a long-term ‘AI employee’ rather than a one-off session tool.
Check it out GitHub Repo. Also, feel free to follow us on Twitter Join our Facebook group! 120k+ ML SubReddit Subscribe now our Newsletter. Wait! Are you using Telegram? now you can join us on telegram as well.

