ARIS-Code, developed by Yang Ruofeng, is a command-line AI research tool built for users who want to organize longer research workflows around language models. It combines Markdown-based skills, multi-provider model support, MCP tool connections, persistent research state, and token tracking in one technical workspace. The focus is less on one-off prompts and more on repeatable research tasks that can be inspected, adjusted, and reviewed over time.

One model works, another checks the homework

The core workflow pairs an Executor model with a separate Reviewer model. One model can handle a research or drafting task, while the other examines the result for weak reasoning, missing support, or inconsistencies. This cross-model review adds a useful verification layer for AI-assisted work, although it does not replace source checking or human judgment when a result matters.

Skills live in Markdown, not a mystery box

ARIS-Code uses Markdown files as reusable workflow skills for tasks such as literature work, experiment support, drafting, citations, and figure preparation. That approach keeps instructions visible and editable instead of hiding the process behind a visual automation interface. Users who like inspecting and adapting their workflows can benefit from that transparency, while those expecting drag-and-drop setup can find the text-based approach demanding.

MCP opens the door to outside tools

The runtime can connect workflows to external tools through the Model Context Protocol. This gives users a way to extend research tasks beyond basic model prompts and build more connected processes around their chosen tools. Persistent state also helps longer projects retain context between sessions, while token and cost tracking gives researchers a clearer view of how much automated work is being used.

The terminal is part of the deal

ARIS-Code is designed for technically confident users. Its CLI-based workflow requires model configuration, provider credentials, and comfort with editing skill files when a process needs adjustment. There is no visual workflow builder to soften that learning curve, so setup can feel like a project of its own before the research automation begins to earn its keep.

Better for builders than button-clickers

ARIS-Code fits researchers and engineers who want more control over AI research automation than a standard chat interface provides. Its editable skills, review loop, and tool integration give it a strong foundation for structured experimentation and drafting workflows. The tradeoff is a technical setup that asks users to understand models, credentials, and command-line tools before the system becomes useful.