Player not loading? Watch on YouTube
This Spanish-language video reviews the DeerFlow 2.0 documentation. DeerFlow 2.0 is presented as an open source AI agent framework from ByteDance for tasks that require more than a chat response. The speaker describes a rewrite that, according to the repository, shares no code with version 1.x. Its main agent can split work among subagents with separate contexts and tools, then combine their structured results.
The setup overview covers cloning the repository and using an interactive assistant to choose a model provider, optional web search and an execution mode. The speaker says the documentation recommends Docker and also describes a local installation path. The interface runs at localhost:2026. This is a self-hosted application, but the transcript does not establish that model inference runs locally: it describes support for models compatible with OpenAI-style APIs.
Tasks have their own file space, uploads and outputs. Examples include market research, reports, websites and slides. The speaker also describes persistent memory for preferences and accumulated knowledge, which the documentation says remains under local user control.
Installation requires comfort with configuration and environment variables. The review cites warnings about command execution and recommends a trusted local environment rather than exposure to untrusted networks. Host Bash is disabled by default, and the speaker points to isolated sandboxes for safer shell execution. Reported issues include permission problems and provisioner restarts.
