DEV Community

Marius-Florin Cristian
Marius-Florin Cristian

Posted on • Originally published at keibidrop.com

KEIBIDROP: Every tool just works

We connected two machines with KEIBIDROP and ran every program we could think of on the shared virtual folder. Compilers, version control, video processing, databases, scripting languages, security scanners. Everything worked. File integrity held across every operation.

KEIBIDROP presents the peer's files through the operating system's filesystem interface. Programs do not know the files are remote. They read and write to a folder. The data arrives from the other machine over an encrypted channel.

How the shared folder works between two machines

Machine A adds files to the shared folder. They appear on Machine B within a second. Machine B opens them, compiles them, processes them, and saves the output back to the shared folder. Machine A sees the output immediately. No upload, no download, no sync button. Both machines read and write to the same folder in real time, over an encrypted peer-to-peer connection.

We tested this with 45+ programs. Alice writes source code, Bob compiles it. Bob writes analysis results, Alice reads them. Alice has a video, Bob creates thumbnails from it. Every tool that reads files from a folder works. The programs do not know the files are on another machine.

Build tools

Tool Test Result
git Clone repo, run git status and git log from the other side Works
make (C) Compile 87 source files inside the shared folder Works
cargo (Rust) cargo build --release on the shared folder Works
go build Init module, build binary on the shared folder Works
swiftc Compile Swift program on the shared folder Works
xelatex Compile LaTeX slides across mounts Works

Media tools

Tool Test Result
ffprobe Inspect 192 MB video from peer's folder Works
ffmpeg Extract clip, create thumbnail grid from peer's video Works
ImageMagick Read PNG, resize, identify 2048x5568 tileset Works

Data and databases

Tool Test Result
SQLite Create DB on one side, JOIN query from the other Works
SQLite (multi-table) RPG save with players + inventory, JOIN from peer Works
jq Write JSON on one side, parse from the other Works
Python csv Write CSV, read with DictReader from peer Works
Node.js Write file on one side, read from the other Works
Jupyter Write .ipynb on one side, parse from the other Works

Languages

Tool Test Result
Python Write module on Alice, import from Bob Works
Perl SHA256 hash with Digest::SHA across mounts Works
Ruby JSON + Digest::MD5 across mounts Works
Node.js npm install on one side, require from the other Works
Bash Write script on one side, execute from the other Works
Swift Compile and run across mounts Works

File operations

Tool Test Result
zip / unzip Zip repo on one side, unzip on the other Works
tar / gzip Create archive, extract on the other side Works
rsync --checksum Checksummed copy from shared folder Works
dd 1 MB + 100 MB random data, SHA256 verified both sides Works
split + cat Split file on one side, reassemble on the other Works
base64 Encode on one side, decode on the other, hash match Works

Integrity verification

Tool Test Result
sha256 Same file hashed from both sides Identical
sha512 Same file hashed from both sides Identical
openssl dgst SHA256 digest from both sides Identical
md5 (bulk) All 71 .c files hashed, aggregate compared Identical
diff Source files compared across mounts Identical
crc32 CRC32 checksum from both sides Identical

Security tools

Tool Test Result
golangci-lint Lint Go source from peer's mount Works
gosec Security scan from peer's mount Works
openssl enc/dec AES-256-CBC encrypt on one side, decrypt from the other Works
exiftool Read image metadata from peer's file Works

Collaborative workflows

The shared folder is bidirectional. Both sides can write. Two people can each handle a different step of the same project.

Workflow Alice does Bob does Result
C compilation Writes main.c + headers Compiles from his view Alice runs the binary Bob built
LaTeX slides Writes .tex source Compiles PDF with xelatex Alice opens the PDF Bob produced
Data pipeline Writes 100-row CSV Runs Python analysis, writes JSON Alice reads Bob's results
Video processing Has a 190 MB video Creates thumbnail grid with ffmpeg Alice sees the thumbnails

Large files

Test Size Result
100 MB random file, SHA256 from both sides 100 MB Identical
192 MB video, ffprobe from peer's mount 192 MB Works
ffmpeg transcode: Bob processes Alice's video, output appears on Alice 192 MB in, 153 KB out Works
50 files written by Alice, aggregate hash verified by Bob 50 files Identical
71 source files tar'd, extracted on other side, all hashes verified 71 files All match

What does not work

Operation Reason
Symlinks / hardlinks Not implemented. Tools like uv, npm with bin links fail. Use --no-bin-links or install outside the shared folder.
Execute binaries from mount (macOS) macOS blocks execution from FUSE mounts. Copy the binary locally first. Linux does not have this restriction.

How it works

KEIBIDROP presents the peer's files through the OS filesystem interface (FUSE on macOS/Linux, WinFsp on Windows). Every program that reads files through standard system calls works without modification. The file metadata syncs when the connection establishes. File contents stream on demand as programs read them. The encryption overhead is under 2%.

The connection is end-to-end encrypted with post-quantum cryptography (ML-KEM-1024 + X25519). No server stores your files.

Top comments (0)