Several times a day someone asked me for "the files from the NAS". The answer was never about storage. It was about opening a folder and seeing 40 files called IMG_4821.JPG with no idea what any of them were until you downloaded them on a laptop, one by one.
So I wrote Banyan: one Go binary, no database, no PHP, no cron. It prints an address, you open it in any browser on the LAN, and you get thumbnails, previews for 100+ formats, and playback for video and audio - straight from a phone or a tablet.
It runs headless: ./banyan -headless. That single file is the whole server.
The trade I made, stated plainly
I did not build a product and then trim it. I picked what a file browser is for, and deleted everything else. Three deletions did most of the work.
1. Accounts and roles. In a home or a small studio, permissions are theater. I removed users, groups, and ACLs, and replaced them with share links that carry an expiry and a scope. The security boundary moved outward, to the network: bind to the LAN interface, put it behind TLS or a VPN, and let the OS decide who can reach the port. Fewer concepts inside the app means fewer places to get an authorization check wrong.
2. Two-way sync. Sync needs vector clocks, conflict resolution, and a story for "which copy wins" - and it drags in a stateful service that must always be running. A file browser needs something weaker: read a directory, show what is there, edit in place, write back to the same path. That is idempotent and stateless. Deleting sync also deleted the whole category of "the index is out of date" bugs, because there is no index.
3. A hosted version. The moment I wanted a hosted tier, I needed tenancy, per-tenant storage quotas, billing, backups, and an abuse process. I dropped it. Banyan stays something you run: your disks, your machine, your backups. This is also why the binary matters more than a docker-compose file - one artifact you can copy to a Raspberry Pi, a mini PC, or a spare Windows box and double-click.
What those deletions bought
- Server-side previews for 100+ formats. Images, HEIC, RAW, PDF, Office documents, archives, subtitles, fonts, and more. The conversion pipeline runs on the box that has the files, so a phone does not download a 60 MB TIFF to look at it.
- Share links without accounts. A URL with a token and an expiry, good for sending a folder to a client for a week.
- Edit inside the preview. Fix a typo in a document or crop a photo, and it writes back to the original path. No download-edit-upload cycle.
- PDF watermark and signature stamping, directory-level AES-256-GCM encryption, and a drag-and-drop upload that keeps folder structure.
- Browse on a TV or an old tablet with no client install, because the client is a browser.
The counterintuitive part
People assume the HTTP server is the heavy piece. It is not. The binary is dominated by the format-conversion pipeline - decoders, renderers, font handling. That is why, when I had to cut something to stay in one file, I cut sync and kept formats. Sync is a feature users ask for; formats are the reason they opened the folder in the first place. Cutting the wrong one would have produced a tidy sync tool nobody needed.
The other thing single-binary changes is distribution: no "install the runtime first", no "the version of libvips your distro ships is too old". The trade is that updates are a file replacement, and that any bug in a decoder is one you ship yourself.
Where it does not fit
If you need multi-user permissions with separate logins, or true bidirectional sync with offline edits, this is the wrong tool - use Nextcloud or Seafile. If you have a folder on a disk and you want to see inside it from any device on the network, that is exactly the case Banyan was built for, and deleting the other three features is what makes it small enough to be that.
Site: https://banyan.ikootoo.com/
Code: https://github.com/nbmany-web/banyan
Downloads: https://banyan.ikootoo.com/release/

