Player not loading? Watch on YouTube
This tutorial walks through DeerFlow setup and a research task using a local model endpoint. The speaker introduces the ByteDance project as an open source agent for long research workflows. The metadata frames the setup around Mac and MLX, though the spoken walkthrough does not explain MLX installation or identify the model.
The setup starts with cloning the repository and running make setup. The speaker selects an OpenAI-compatible provider, enters a local URL, and supplies a Tavily API key for web search. For web fetching, he chooses the reader provider transcribed as "Gina AI." He keeps the local sandbox and enables bash commands and write tools, then runs make install and make dev. The frontend opens at localhost:3000, where he selects "get started with 2.0" and creates an account.
The demo asks for research on AI tooling trends for developers in 2026. DeerFlow displays expandable steps before producing a report with summaries and sections on tooling, frameworks and vector databases. The speaker checks Tavily usage before and after the run: the counter rises from 23 to 30, a difference of seven credits. This is a local AI workflow with external search and fetching services; the demonstration does not establish fully offline operation or that all research data stays on the Mac.
