Deer Flow

What is Deer Flow?

The project documentation describes Deer Flow as an open‑source SuperAgent harness developed by ByteDance. It is designed to research, write code, and produce artifacts by combining sandboxed execution environments, persistent memory, plug‑in tools, and the ability to delegate work to sub‑agents. The repository, hosted at github.com/bytedance/deer-flow, shows 77,978 stars, indicating substantial community visibility. The tool is not a single monolithic agent but a framework for orchestrating multiple agent-like components that can interact, share state, and pass tasks between each other. Sandboxes allow each component to run with isolated filesystem and network access, which the documentation notes is intended to reduce unintended side effects during agent execution. The memory component stores conversation history and results across sessions, while the tool system lets agents call external APIs, run shell commands, or invoke other functions. Sub‑agents can be spawned to handle subtasks in parallel, returning results to the parent agent. Deer Flow is built in Python and its source code is publicly available under a license that permits free use, modification, and redistribution.

Who is Deer Flow for?

Based on the description in the repository, Deer Flow is aimed at developers and researchers who need a structured way to build multi‑step, tool‑augmented agent workflows. It is relevant for teams building coding assistants, automated research pipelines, or any system that requires an agent to break a complex task into smaller steps, execute them in sandboxed environments, and maintain long‑term memory. The sub‑agent capability makes it suitable for scenarios where parallel execution or separation of concerns is beneficial. Because the project is open source and does not require paid accounts or API keys to run (beyond any external tools the user chooses to connect), it can be used for prototyping, academic research, or production deployments where the team prefers to control the infrastructure. The sandbox feature also appeals to organisations with security requirements that need to limit what an agent can access or modify.

How much does Deer Flow cost?

The project is offered under a free pricing model. The pricing page lists the starting cost at $0 per month. Being open source, there are no license fees, subscription tiers, or usage‑based charges from the project itself. Users can download the source code from GitHub, run it on their own hardware or cloud instances, and modify it as needed. Any costs incurred are limited to the infrastructure used to host the agents, such as compute time, storage, and network bandwidth, as well as any third‑party services (e.g., language model APIs, external databases) that the user configures Deer Flow to call. The project does not bundle or require any paid add‑ons directly from ByteDance.