jiyu.

labs

  1. building an IAM agent

    what it means to give an agent a real identity — accountability, delegation, and a bounded scope of action.

  2. speech to code, speech to IGA command

    turning natural language into safe access decisions — understand first, propose second, never act alone.

  3. treating humans as (almost) normal identities

    starting from agent-centric identity, and treating the human as the special case rather than the centre.

  4. secure pseudonyms

    letting people and agents use models without the model ever learning who they are — ephemeral, single-use identities.