ByteDance just open-sourced a 45K-star AI agent army
DeerFlow 2.0 from ByteDance hit 45K GitHub stars — an open-source SuperAgent that researches, codes, and creates. Free, works with GPT, Gemini, Claude, and more.
DeerFlow 2.0, ByteDance's open-source AI agent platform, just hit 45,600 GitHub stars — gaining nearly 3,800 in a single day. It's a "SuperAgent" that can research topics, write code, generate reports, create presentations, and even produce images and videos, all by orchestrating multiple AI sub-agents that work together automatically.
Think of it as giving AI a team of specialists instead of one generalist. You describe a complex task, and DeerFlow breaks it down, assigns pieces to different agents, runs them in sandboxed environments (isolated containers so nothing goes wrong on your computer), and delivers the final result.
What DeerFlow Can Actually Do
DeerFlow isn't just another chatbot wrapper. It's an orchestration platform — meaning it coordinates multiple AI models and tools to complete tasks that no single model could handle well alone.
Deep research — Give it a question and it searches the web, reads sources, cross-references facts, and writes a structured report with citations
Code generation — Spawns coding sub-agents that write, test, and debug code in isolated Docker containers (virtual boxes that keep experiments safe)
Presentation builder — Generates full slide decks from research findings, complete with formatting
Image and video creation — Integrates with generation tools to produce visual content as part of larger workflows
The key difference from tools like ChatGPT or Claude: DeerFlow handles multi-step tasks that take minutes to hours. You don't sit there typing follow-up prompts — you give it a goal and walk away.
Works With Almost Every AI Model
DeerFlow doesn't lock you into one AI provider. It supports:
OpenAI (GPT-4, GPT-5) — Google Gemini — DeepSeek v3.2 — Anthropic Claude (via Claude Code integration) — ByteDance's own Doubao models — and more through its configuration system.
You choose which model handles which task. For example: use GPT-5 for planning, DeepSeek for code generation, and Gemini for research — all in the same workflow. The platform uses LangChain as its backbone, so any model that LangChain supports works here too.
Not Just a Web App — It Lives in Your Chat
Beyond the web interface, DeerFlow connects directly to Telegram, Slack, and Feishu/Lark as a bot. This means you can message your DeerFlow agent from your phone and get research reports, code, or analysis without ever opening a browser.
There's also a Claude Code integration — install the claude-to-deerflow skill, and you can control DeerFlow directly from your terminal while coding. Ask it to research something, generate a component, or analyze a codebase, and the results stream back in real-time.
How to Get Started
DeerFlow requires some technical setup (Python, Node.js, and optionally Docker), but the team has simplified it significantly with make commands:
# Clone and setup
git clone https://github.com/bytedance/deer-flow.git
cd deer-flow
make config
# Edit config.yaml with your API keys
# Then start with Docker (recommended):
make docker-start
# Or start locally:
make install && make devOnce running, open http://localhost:2026 in your browser and you'll see the full agent interface.
Who Is This For?
Researchers and analysts: Automate literature reviews, competitive analysis, and report generation. Tell DeerFlow to "research the latest developments in quantum computing and write a 10-page report" and come back to a finished document.
Developers: Use the coding agents to prototype features, debug issues, or generate boilerplate — all in sandboxed environments that won't break your production code.
Content creators: Chain research → writing → presentation → image generation into a single workflow. One prompt, multiple deliverables.
Small teams: DeerFlow is like having an entire research department that works 24/7 for the cost of API calls.
From Research Tool to Full Agent Platform
DeerFlow v1 was a specialized deep research tool. Version 2.0 is a complete rewrite that transforms it into a general-purpose agent orchestration platform. It hit #1 on GitHub Trending when it launched in February 2026, and it's climbing again with today's surge of 3,787 new stars.
With 254 open issues and active development, the project shows strong community engagement. The GitHub repository includes comprehensive documentation, skill development templates, and contributing guidelines for anyone who wants to extend it.
The fact that ByteDance — the company behind TikTok — is open-sourcing its internal AI agent infrastructure is notable. It suggests the company believes the value isn't in the agent framework itself, but in the data and services built on top of it.
Related Content — Get Started with Easy Claude Code | Free Learning Guides | More AI News
Stay updated on AI news
Simple explanations of the latest AI developments