fanbox
vibe coding 的驾驶舱：左边文件，右边/下边终端，中间看清每一次改动。 / The cockpit for vibe coding: browse files on the left, command agents on the right, watch
What is fanbox?
vibe coding 的驾驶舱：左边文件，右边/下边终端，中间看清每一次改动。 / The cockpit for vibe coding: browse files on the left, command agents on the right, watch every change in between.
Overview of fanbox
What it is
FanBox is a cockpit for coding agents. It allows users to command Claude Code or Codex locally, see every file and line they change, and take over anytime. Users can browse, preview, and edit local files on one side while running any coding agent in a real embedded terminal on the other. Every time the agent writes a file, its card lights up. All operations are contained in one window.
Key features
- Global fuzzy search (⌘K): Search by file name fragment or use
content:keywordfor full-text search. - Preview in place: Rendered Markdown, live HTML, syntax-highlighted code, inline images/video/PDF (including HEIC), archive content listing, and checkerboard backing for transparent images.
- Live dashboard: Every file the agent writes makes its card ripple and glow by change frequency; the light follows wherever the agent goes.
- Follow mode: One click makes the file view and preview track whatever file the agent edits. Code scrolls with freshly written lines flashing, HTML renders live while being written (double-buffered, zero white flash), Markdown renders live.
- This round panel: Shows which files the agent changed, which terminal/agent did it, +/- per file; click for a side-by-side diff, restore any single file, roll back the whole round, scrub the replay timeline.
- Project memory: Open any project folder to see past AI sessions (first message as title), files changed, skills triggered; one-click resume via
claude --resume/codex resumein the embedded terminal. - AI organize: AI proposes a cleanup plan from metadata only (never reads content or touches the filesystem); you approve each move; FanBox executes with a rollback log and one-click undo.
Use cases
FanBox is designed for developers who use AI coding agents to spin up multiple projects in a short time. It solves the problem of scattered files, unrecognizable project names, and invisible changes. The typical workflow — digging through Finder, switching to iTerm to launch an agent, then switching to a browser to check results — is folded into one window. It focuses on the chain of finding files, previewing, light editing, and commanding the agent.
Getting started
See the repository README for installation instructions.
License and pricing
Licensed under MIT. Open source.
Based on the project's public README and repository metadata.
