Home / Platform / GaiaKeep
Platform Layer: Durable data active

GaiaKeep

Documentation GitHub

GaiaKeep is the team’s distributed, durable, agent-based storage system: a sovereign, agent-focused storage platform that runs locally over Cresco, and a global file system built for agents rather than people.

  • Versioned and citable. Every read names a version, and each version, or a frozen extract of one, is citable with an identifier that follows the Research Data Alliance data citation recommendation, so each evaluation records the exact data it used.
  • Secure by design. Data is encrypted at its origin, so storage sites hold only ciphertext and no content keys. Every request is signed, roles are scoped to each tenant, the system administrator has no access to data, and the audit log is tamper-evident. Data can be forgotten by destroying its keys, and a legal order needs at least two distinct approvers and a 24-hour lock.
  • One storage continuum. Memory, NVMe, disk, and LTO tape form one continuum, with tape as the durable tier across independent sites around the state.
  • Data stays where it lives. GaiaKeep follows an edge paradigm in which agents use datasets where they live, and any version streams to a GPU node, a partner site, or an agent’s working cache.

All of the platform and agent systems use GaiaKeep as versioned storage for datasets, models, and code. The operator runs its own storage hardware and software, holds its own keys, and can keep data in the state; capacity is purchased rather than metered.