AI Frameworks · bytedance

deer-flow

DeerFlow is an open-source agentic framework from ByteDance that orchestrates AI agents with sandboxes, memory, and extensible skills to handle long-running research and coding tasks. It supports multiple LLM providers and integrates web search, file operations, and sub-agent delegation for complex workflows.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/bytedance/deer-flow

76.4k

GitHub stars

10.4k

Forks

Python

Primary language

MIT

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	bytedance/deer-flow
Owner	bytedance
Primary language	Python
License	MIT — OSI-approved
Stars	76.4k
Forks	10.4k
Open issues	930
Latest release	v2.0.0 (2026-06-25)
Last updated	2026-07-07
Source	https://github.com/bytedance/deer-flow

What deer-flow is

A Python/Node.js agent harness (v2.0 ground-up rewrite) that manages multi-step tasks via LangChain/LangGraph integrations, isolated execution sandboxes, persistent memory systems, and pluggable skill modules. Supports Claude Code, DeepSeek, Kimi, and OpenAI models with configurable execution policies and tracing backends (LangSmith, Langfuse).

Quickstart

Get the deer-flow source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/bytedance/deer-flow.git2cd deer-flow3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Deep Research & Report Generation

Multi-hour research workflows combining web search (InfoQuest integration), memory recall, and structured output generation. Handles iterative refinement and context compaction for long conversations.

Code Generation & Automated Development

End-to-end coding tasks from requirements to sandbox execution, with Claude Code integration, file system access, and skill-based tool composition for validation and iteration.

Scheduled Autonomous Task Execution

Background agents handling periodic tasks (data collection, report updates, API integration tests) with sandbox isolation, memory persistence, and audit trails via IM channel integrations.

Implementation considerations

Interactive setup wizard (`make setup`) generates config.yaml and .env; Docker recommended for deployment. Local dev requires Python 3.12+ and Node.js 22+.
LLM provider selection is mandatory; wizard guides choice but no default fallback. API keys and cost management are operator responsibility.
Sandbox mode, bash access, and file-write tools are configurable execution policies; security posture depends on deployment context and chosen policies.
Memory and sub-agent coordination rely on internal state management; no documented distributed-tracing or multi-instance consistency guarantees.
Skills are extensible but framework expects developer to implement, test, and validate custom tools; no official marketplace or pre-built library documented.

When to avoid it — and what to weigh

Strict Real-Time Latency Requirements — Long-horizon tasks and context compaction add latency; unsuitable for sub-100ms response SLAs or synchronous API gateways requiring immediate replies.
Highly Regulated Security Environments — README explicitly warns improper deployment introduces security risks; sandboxing and memory isolation require careful configuration. Not suitable without thorough security review and hardening.
Production without Vendor Lock-In Tolerance — Heavy reliance on ByteDance-recommended models (Doubao, Kimi) and proprietary integrations (BytePlus InfoQuest, Volcengine). OpenAI/OSS alternatives work but are not primary targets.
Limited DevOps/Observability Budget — Requires Docker, tracing backend setup, LLM provider accounts, and configuration management. Minimal prerequisites but non-trivial operational overhead for small teams.

License & commercial use

MIT License (permissive; allows commercial use, modification, distribution, and private use with attribution requirement). No proprietary restrictions on code; derivative works may be distributed under MIT or compatible licenses.

MIT License permits commercial deployment without royalties or vendor approval. However, integration with BytePlus/Volcengine services (InfoQuest, Doubao models, Coding Plan incentives) may introduce commercial terms. Requires review of service ToS for any bundled or recommended proprietary tools. Open-source code itself is commercially viable.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Clear
Deployment complexity	Moderate
DEV.co fit	Good
Assessment confidence	High

Security considerations

README includes explicit ⚠️ security notice: 'Improper Deployment May Introduce Security Risks' with recommendation for security review. Sandbox mode, bash access, and file-write tools are configurable; default posture is not stated. No details on: isolation guarantees, code injection prevention, credential handling, audit logging, or compliance frameworks. Evaluate threat model, LLM provider trust, and sandbox implementation before production use. ByteDance sponsorship does not imply third-party audit.

Alternatives to consider

LangGraph / LangChain Agent Framework

Lower-level orchestration; requires more custom wiring for memory/sandboxes but avoids vendor integration and gives fine-grained control.

AutoGen (Microsoft)

Multi-agent conversation framework with similar long-horizon task support; more mature, broader LLM provider support, but less focus on sandboxing and research workflows.

Crew AI

Role-based agent framework with task composition; simpler mental model for sequential workflows but lacks deep-research, sandbox, and memory specialization.

Software development agency

Build on deer-flow with DEV.co software developers

Start with DeerFlow's interactive setup wizard (make setup) and deploy your first agent in Docker. Review the security notice and configuration examples, then connect your LLM provider and optional web search.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

mlflow langflow promptflow

Related on DEV.co

Explore the category and the services that help you build with it.

AI Frameworks AI Application Development API Development DevOps & Cloud Deployment Talk to DEV.co

deer-flow FAQ

Can I use DeerFlow with my own LLM (e.g., local Ollama)?

Yes, vLLM provider is documented (e.g., Qwen3 32B over vLLM 0.19.0). OpenAI-compatible APIs can be routed via `base_url`. Local inference is supported but must be explicitly configured.

What happens if my LLM provider goes down or API limits are hit?

Not documented in README. Retry logic, fallback models, and rate-limiting are not mentioned. Operator must implement external circuit-breakers or multi-provider failover.

Is DeerFlow suitable for production chatbots with real-time user interaction?

No. Long-horizon tasks and context compaction add latency. Better suited for async workflows (scheduled tasks, research agents, batch processing). Real-time bots should use lighter frameworks (LangChain, LangGraph).

What is the cost to run DeerFlow?

Depends entirely on chosen LLM provider(s) and usage. No built-in cost controls or quota management documented. ByteDance's Coding Plan offers incentives for recommended models (Doubao, Kimi) but terms are external.

Software developers & web developers for hire

DEV.co is a software development agency delivering custom software development services to companies building on open source. Our software developers and web developers design, integrate, and ship production systems — spanning web development, APIs, AI, data, and cloud. If deer-flow is part of your ai frameworks roadmap, our team can implement, customize, migrate, and maintain it.

Ready to automate research and coding workflows?

Talk to DEV.co Schedule a Consultation