<![CDATA[Let AI agents work on huge datasets without transferring them]]>https://keibidrop.hashnode.devhttps://cdn.hashnode.com/res/hashnode/image/upload/v1593680282896/kNC7E8IR4.pngLet AI agents work on huge datasets without transferring themhttps://keibidrop.hashnode.devRSS for NodeSat, 19 Sep 2026 00:37:29 GMT60<![CDATA[The same 16 MiB block, fetched once a second for a minute]]>https://keibidrop.hashnode.dev/the-same-16-mib-block-fetched-once-a-second-for-a-minutehttps://keibidrop.hashnode.dev/the-same-16-mib-block-fetched-once-a-second-for-a-minuteSun, 06 Sep 2026 18:31:03 GMTHow do you work on a cut in DaVinci Resolve while traveling, when the footage sits on the studio NAS? KEIBI DROP shows the NAS folder on your laptop, over the internet, and fetches only the parts you touch, so you open the footage on demand straight from Resolve, Final Cut or Premiere. To show that, we recorded a Resolve demo: a NAS in Timisoara serves a shoot, a laptop in Bucharest, 400 km away, edits from it. The first take moved 3.17 GB for 146 MB of clips, and Resolve froze inside a read. By the end of the day four bugs were out. All four had been in the code since July and August. They showed up when an editor's read pattern met a saved contact, a daemon restart, a relayed lane and a Mac that went to sleep. This is what each one was, how the trace found it, and what the same take measures now.

DaVinci Resolve on the Deliver page scrubbing a 4K clip that lives on the NAS; the filmstrip thumbnails fill in as the playhead moves.

Measured, with the conditions on the original page:

  • Import four clips | 2 s | 0.1 MB
  • Build the timeline (filmstrip thumbnails) | 8 s | 199 MB
  • Land on clip 2, play 5 s | 8 s | 32 MB
  • Jump to the 4K clip, play 6 s | 13 s | 78 MB

Full version, with the limits and the method: The same 16 MiB block, fetched once a second for a minute

Source, MPL-2.0: github.com/KeibiSoft/KeibiDrop

]]><![CDATA[The files are on the NAS at home and the laptop is somewhere else]]>https://keibidrop.hashnode.dev/the-files-are-on-the-nas-at-home-and-the-laptop-is-somewhere-elsehttps://keibidrop.hashnode.dev/the-files-are-on-the-nas-at-home-and-the-laptop-is-somewhere-elseSat, 05 Sep 2026 12:24:25 GMTYour files live on a box at home or at the studio. Your laptop is somewhere else. KeibiDrop puts the box's folder on the laptop as a normal folder. You open a file, and the box sends the parts the program reads. The new container runs KeibiDrop on the box. The box is on all day, so the folder is there whenever you open the laptop, wherever you are.

Two terminals. Left: the box

Full version, with the limits and the method: The files are on the NAS at home and the laptop is somewhere else

Source, MPL-2.0: github.com/KeibiSoft/KeibiDrop

]]><![CDATA[Let AI agents work on huge datasets without transferring them]]>https://keibidrop.hashnode.dev/let-ai-agents-work-on-huge-datasets-without-transferring-themhttps://keibidrop.hashnode.dev/let-ai-agents-work-on-huge-datasets-without-transferring-themFri, 04 Sep 2026 13:39:23 GMTKeibiDrop connects two machines directly and gives both sides the same folder, encrypted end to end, with no cloud storage in between. kdmcp lets AI agents drive it as an MCP server: the remote folder mounts as a normal path, the agent reads files where they sit, and only the bytes it reads cross the wire. In a measured run, reading 1 MB out of a 256 MB file moved 16.8 MB instead of the whole file, and a 2,184-file dataset was listed 0.53 seconds after pairing. Setup is one config entry, and pairing needs a code from each side, so a single leaked code connects nobody.

Two terminal panes. The left agent offers a dataset folder over KeibiDrop, the right agent lists 2,184 files and reads one off the mount, with a running clock along the bottom.

Measured, with the conditions on the original page:

  • 1 MB read from the middle of a 256 MB file | 16.8 MB on the wire, 6.25% of the file, content matched
  • 40 MB download cut at 8 MB, then resumed | Finished, checksums matched on both sides

Full version, with the limits and the method: Let AI agents work on huge datasets without transferring them

Source, MPL-2.0: github.com/KeibiSoft/KeibiDrop

]]>