Deer-Flow is an open-source AI agent harness and orchestrator developed by ByteDance. It provides a structured runtime environment for executing, managing, and extending autonomous agent workflows directly from the command line and terminal interfaces. Day-to-day interaction occurs through an interactive Terminal User Interface (TUI) and an operator CLI (deerflow). The harness manages agent execution via configurable skills stored in repository directories, routing tasks through backend gateway services. Dependency synchronization and environment isolation are handled via uv, with runtime behavior declared in a central config.yaml. Deer-Flow is distinguished by its strict modularity and transactional extension management system. It supports gateway contribution points and packaged extensions from local paths or remote Git repositories, using an ExtensionManager that audits lockfiles and rolls back configuration changes if environment reproducibility fails. Deer-Flow is an open-source project hosted on GitHub with an active commit history, reproducible Docker builder support, and multi-platform packaging capabilities.
Tags: agent harness, ai agent, cli, coding agent, developer tools, OpenSource
deerflow extensions CLI for installing, listing, enabling, disabling, and auditing plugins. - Gateway Contribution Points — Extensible backend services and router hooks for contributing custom routes and services to the agent gateway. - Structured Agent Skills — Directory-based workflow definitions (.agent/skills) for executing reproducible system modifications and verification pipelines. - uv and Docker Integration — Locked, reproducible dependency resolution backed by uv environments and container builder targets. - Transactional Configuration — Safe plugin updates that validate top-level YAML configurations and roll back transactions on lockfile discrepancies.deerflow extensions install. The transactional installer validates package requirements, updates pyproject.toml, and registers plugins into config.yaml safely.Deer-Flow is designed for platform engineers, backend developers, and AI engineers building modular agent infrastructure with strict dependency and configuration requirements. It is not intended for non-technical users seeking a simple browser-based AI chat tool.
bash git clone https://github.com/bytedance/deer-flow.git cd deer-flow # Sync dependencies using uv uv sync --lockedbash cp config.example.yaml config.yaml 2. Inspect available extensions using the operator CLI: bash # Using the operator CLI directly or via uv uv run deerflow extensions listbash # Install an extension via HTTPS Git URL uv run deerflow extensions install https://github.com/example-org/deerflow-sample-extension.git # Launch the agent TUI uv run deerflow tuiPart of CLI Coding Agents
Categories: AI Coding Agents, Code Assistants, Developer Tools