Suzuri: A Local-First, AI-Native Research Development Environment

AI has already changed how research is done: agents search literature, draft prose, and write and run analysis code. What has not changed is where. The work remains scattered across a reference manager, a notes app, a writing tool, a code editor, and a chat window: what human and agent produce crosses tool boundaries as pasted text and exported files, the author pays in constant application-switching, and the agent never sees the whole project. Part of the solution has already emerged bottom-up: local-first, plain-markdown knowledge management — vaults of files that humans read rendered and agents read raw — has gained wide traction and popularity. But writing still happens in another tool and code in a third, and agents are native to the code editor and the chat window — not to the place where notes, sources, and the manuscript live. We name the missing tool the Research Development Environment (RDE) — the scholarly analogue of the IDE, defined by five properties — and present Suzuri, an open-source RDE built as a fork of the Zed editor, in which agent chat, skill runs, and Model Context Protocol (MCP) calls to external services, scientific writing and typesetting, and code and data analysis all happen in one folder on local disk. Suzuri is extensible by design: we demonstrate Kiroku, a skill for human–agent behavior analytics that reconstructs and analyzes the research process from Claude Code logs, and present key features under development and the future roadmap.