DeerFlow (Deep Exploration and Efficient Research Flow) is an open-source agent that orchestrates sub-agents, memory, and sandboxes to perform a wide range of tasks, powered by extensible skills.
Orchestrating sub-agents, memory, and sandboxes with extensible skills to perform a wide range of tasks, including deep research, with support for various LLM providers, web search, sandbox execution, and long-term memory.
DeerFlow 2.0 explicitly describes itself as a "super agent harness" that orchestrates sub-agents, memory, and sandboxes, and the code surface confirms it executes arbitrary code and shell commands. The harness spawns and coordinates sub-agents and dynamically executes code derived from LLM output, which is the defining signal of Tier 5 (self-modifying / agent-spawning). It runs an autonomous agentic workflow with no code-enforced approval gate evident in the surface analysis. The entrypoint heuristics flagged nothing, but the broader code surface (executes arbitrary code + shell, multi-agent orchestration, filesystem writes to variable paths, integrations with messaging platforms) overrides the README and entrypoint absence of signals.