AI models & datasets
Your models and data, wherever the GPUs are
GPU capacity is spread across clouds, regions and labs, and the data has to follow it. fft moves model weights, checkpoints and training sets straight from your machine to theirs at up to 94 Gbps, encrypted end to end, so nobody in between ever holds a copy.
$ fft send ./model-weights fft:eyJ2Ijox… -r
GPUs are scarce. Idle GPUs are expensive.
GPU capacity is where you can get it: a GPU cloud in one region, a reserved cluster in another, your own racks on premises. Checkpoints from a large training run are terabytes and training corpora are larger, and every hour spent copying them is an hour of paid-for GPUs sitting idle. Copies across regions and providers slow down with distance, and routing weights through third-party storage gives one more party a copy of your most valuable asset.
Falcon fft moves weights, sharded checkpoints and datasets straight between machines: 100 GiB in 11 seconds over 100 Gbps networking, encrypted and verified, and 33–40 Gbps across a 100 ms path with 1% loss. Model weights compress on the wire when the network is the bottleneck: bf16 weights moved at 11.8 Gbps over a 10 Gbps link.
Folder sync keeps checkpoint directories in step between clusters, sending only new shards, so a run can fail over or move to cheaper capacity quickly. Delivering a model to a customer's own VPC or data center goes from your disk to theirs, encrypted end to end, with no middleman. It's built for automation: machine-readable progress and reports, and the full command contract as JSON for your pipelines and AI agents.
Tell us where your GPUs and data live. Request early access.
Your weights stay yours
Peer to peer with no cloud, relay or storage in the path. Keys exist only on the two machines.
Stop paying for idle GPUs
Fills 10–100 Gbps pipes with nothing to tune, so clusters start working sooner.
Across oceans, at speed
Holds its rate over long, lossy paths where ordinary copies slow to a crawl.
Built for pipelines and agents
Machine-readable output, and the full command contract as JSON for an LLM to read.
Where it fits
Stage training data
Copy datasets into a GPU cloud or a rented cluster without first uploading to someone else's storage.
Replicate checkpoints
Keep checkpoints in step between clusters and regions, so a run can move or recover fast.
Deliver models privately
Ship weights to a customer's or partner's own infrastructure, encrypted from your disk to theirs.
Share research data
Move large datasets between national labs, universities and the teams that train on them.
Related use cases
Falcon fft and Makai Systems: not in the path
No middleman. We never see your data.
Every transfer goes peer to peer, straight from one machine to the other, encrypted with keys that exist only on those two machines. There is no Falcon cloud, relay or storage in between, so neither Falcon fft nor Makai Systems ever sees your files, their names, or who you send them to.
Try it on your own network
Falcon fft is in early access. Leave your email and we'll set up a trial on your own links.