KeibiDrop

Website · Download · Blog

DaVinci Resolve scrubbing a 4K timeline whose clips live on a NAS 400 km away. The thumbnails fill in as the playhead moves; only the touched parts cross the internet. Click to watch the two minute demo on YouTube.
_{How to edit in DaVinci Resolve, Premiere Pro or Final Cut while traveling. Your footage stays on the studio NAS, you open it on demand from the app, no waiting for a download or an upload. Click for the two minute demo.}

Another computer's files show up as a folder on yours. Open and edit them in your own apps, with nothing to download first.

• Open a 40 GB file and start working instantly. Your edits sync back.
• End-to-end encrypted (post-quantum). No cloud, and the relay never reads your files.
• Same network: one click. Over the internet: swap a code once, then save the contact.
• Their files become a real folder on your machine. Open, edit, git clone, anything.
• macOS, Linux, Windows, Android. iOS coming soon. Open source engine (MPL-2.0).

KeibiDrop connection screen

KeibiDrop demo, Mac and Windows side by side. Click to watch on YouTube with sound.
_{Mac and Windows side by side. Click to watch on YouTube (with sound)}

brew tap keibisoft/keibidrop
brew trust keibisoft/keibidrop
brew install keibidrop

wget -O keibidrop.deb $(curl -s https://api.github.com/repos/KeibiSoft/KeibiDrop/releases/latest \
 | grep -o 'https://[^"]*_amd64\.deb')
sudo apt install ./keibidrop.deb

choco install keibidrop

Or download from GitHub Releases.

We host our own F-Droid repository. Open fdroid.keibisoft.com on your phone, scan the code, and check the fingerprint before you add it.

To add it by hand, use https://fdroid.keibisoft.com/fdroid/repo in F-Droid under Settings, Repositories. The fingerprint to check is on the same page.

git clone https://github.com/KeibiSoft/KeibiDrop.git
cd KeibiDrop
make build-kd # CLI daemon
make build-cli # Interactive CLI
make build-rust # Desktop UI (needs Rust + Slint)

Same network (LAN):

1. Both peers launch KeibiDrop
2. Devices discover each other automatically
3. One click to connect
4. Share files

Over the internet:

1. Both peers launch KeibiDrop
2. Copy your fingerprint and send it to your peer (Signal, Telegram, anything)
3. Paste each other's fingerprints and connect
4. Share files

Save a peer as a contact to skip the fingerprint exchange next time. On the same network, saved contacts connect with one click using pseudonyms.

It works through firewalls automatically. If direct connection fails, KeibiDrop falls back to an encrypted relay. No port forwarding, no router configuration.

1. Peers exchange fingerprints (or discover each other on LAN)
2. KeibiDrop finds the fastest path: LAN, direct IPv6, or encrypted relay
3. Files transfer over an encrypted channel the relay cannot read
4. Keys rotate automatically for forward secrecy

The encryption is post-quantum (ML-KEM-1024 + X25519) with AES-256-GCM or ChaCha20-Poly1305. Full protocol details in Security.md.

./keibidrop

./keibidrop-cli

./kd start # Start daemon
./kd show fingerprint # Get your fingerprint
./kd register <peer-fingerprint> # Register peer
./kd create # Create room (or: kd join)
./kd add /path/to/file.zip # Share a file
./kd list # List shared files
./kd pull file.zip ~/Downloads/ # Download a file

See TROUBLESHOOTING.md.

See CONTRIBUTING.md.

Go engine, CLI, and mobile bindings: Mozilla Public License 2.0 (per-file copyleft)

Rust UI, mobile apps, and brand assets: Proprietary - see DUAL-LICENSE.md

Desktop UI built with Slint

Built by KeibiSoft SRL.

Remote triage without a VPN: 20 GB mounted, 606 MB moved

Mount an evidence share read only across NAT and extract the artifact set you need, with no inbound port, native on Windows, and no kernel driver on the evidence machine. There is a write-up and a DFIR page.

Mount the camera originals. No proxy transcode before the handoff, and no relink after it.

Mount the workstation that holds the data. No bucket, and no upload before the first batch.

Two terminals. Left: a container on a NAS prints its code and Peer connected. Right: a laptop adds the box as a contact, connects, lists the shared folder and hashes one photo through the mount

A session needs both machines online, and a NAS, a mini PC or a home server is online all the time. The container in docker/ runs the serving side on that box from plain Docker with the stock kernel. Your laptop mounts the folder from anywhere and reads on demand, so only the bytes a program reads cross the wire. In that run the box, a container on a server in Timisoara, served 161 files to a laptop in Bucharest; the laptop paired once, listed the folder and hashed one photo. Setup is one compose file, in the guide. An Unraid template is in docker/unraid/.

The raws and the footage stay on the box at the studio. The laptop mounts the folder, Lightroom or Resolve opens the files from the mount, and each file moves only when it is opened; a 1 MiB read out of a big file moves one 16 MiB block. The catalog stays on the laptop, as the post-production page and the video and audio guide describe. Set the share read only when the box is the archive and the laptop must change nothing.

Two agents pairing over MCP: the laptop offers a 2,184 file dataset in place, the gpu box lists it and reads shards off the mount

Two agents, one per machine, each driving a KeibiDrop peer through the MCP server. The laptop offers a folder where it sits. The gpu box mounts it, lists 2,184 files six folders deep, and checksums 24 shards. In that run 88,581 bytes crossed the network out of 6,784,286 on disk, because that is what it opened. The whole thing, pairing included, took 12.5 seconds.

Build it with make build-kdmcp, then kdmcp install prints the config snippet for your MCP client. Sending stays off until KD_SHARE_ROOT names one directory to send from, so a receiving agent cannot be talked into shipping files out.

• Website
• Documentation
• Compared with sshfs - measured on a 200 ms link
• Technical deep dive
• Blog posts (53 posts on FUSE, crypto, performance)
• FAQ
• Comparison with alternatives
• About and credentials