Come build the filesystem for superintelligence.
Agents are taking over knowledge work, and what follows them will not run inside one company. We build the filesystem for it, so it can store and share the knowledge it generates across heterogeneous machines, clouds and owners.
What you would work on.
Agents are built on recent work in language models. Rift is built on recent work in three other fields, combined in one filesystem.
CRDTs
Two machines can change the same folder at the same time, online or off. Their changes merge on their own, and no server decides which came first.
Group key agreement
The machines and agents a folder is shared with agree on its key among themselves. When one is removed, the rest agree on a new key.
Peer-to-peer over QUIC
Machines connect to each other directly across networks. When they cannot, a relay carries the ciphertext.
What we care about.
Privacy
We cannot read our users' files. Every decision we make has to keep that true.
Ownership
The work an agent does belongs to the person or company that ran it. It is not tied to our account or to another company's.
Speed
A file should reach every machine as fast as the network allows. Encrypting it must not slow it down.
Ease of use
Rift has to be easier to set up than a storage bucket. Today that means no account and one prompt pasted into an agent.
Agents first
Agents are the main users of Rift. When an agent uses a command wrongly, we fix the command.
Rigor
We test Rift against dropped connections and hostile machines.
Open roles.
Singapore or remote