// the find
Nagi-ovo/voyager
Voyager — an all-in-one enhancement suite for AI Studio, Gemini, Claude & ChatGPT: timelines, folders, prompts, usage tracking, chat export, plugins, and more. / 面向 AI Studio、Gemini、Claude 与 ChatGPT 的全能增强套件。
Voyager is a browser extension that adds organization and navigation features to Gemini, Claude, and ChatGPT — folders, timeline navigation, prompt libraries, chat export, and usage tracking. It started as a Gemini-focused tool and has since grown into a multi-platform extension with a plugin engine for extending support to other AI interfaces. The target user is someone who lives in AI chat interfaces daily and finds the native UIs too bare.
The plugin architecture is the smart move here — instead of hardcoding Claude and ChatGPT support, they built an engine that lets the community extend Voyager to other AI interfaces, which is why you see DeepSeek and Doubao forks spawning naturally. Cloud sync via Google Drive or iCloud on Safari is genuinely useful since extension storage is ephemeral and most users lose their prompts eventually. The watermark removal for Gemini-generated images is a clean adaptation of existing open-source work, properly attributed. The Bun-based build pipeline with per-browser targets (chrome/firefox/safari/edge) is well organized for what is inherently a messy multi-target build problem.
The dependency on DOM scraping and injecting into AI interfaces that change without notice is the core fragility — every major UI update from Google or Anthropic is a potential breakage event, and with five browsers and three platforms, regression surface is large. The Safari path requires building through Xcode, which means contributing to that target is effectively macOS-only and will always lag the Chrome version. The plugin engine docs appear to live mostly in the VitePress docs site rather than in-repo specs, so plugin authors are one doc deployment away from being stuck. There is no mention of automated cross-browser testing; the `.github/docs/REGRESSION_NOTES.md` file existing suggests manual regression tracking, which does not scale well as features accumulate.