Below is not request as much as a concept that will become relevant once everyone starts to use automated web browsing.
0: Kagi has inexpensive hardware attestable minimum spec devices somewhere.
1: Ultimate enables you to gain access to a Linux privileged user in a device.
2: This happens such that first the device is attested against a Kagi talisman or DIY device, with attestation keypair & storage.
3: Attestation successful => Locally decrypt storage & tell you did to Kagi => Kagi elevates you to virtual non-attestable more powerful runtime. This still cannot run hardware wise LLM inference or do heavy computing. LLM runtime comes from providers.
4: Your local store information populates now Kagi's virtual runtime, that is for now online & SSH-able if you want, such that there is Kagi plain assistant discussions as one other account.
5: Each agentic custom assistant, e.g. normal discussion can be you spawned its own user, turning it agentic.
6: The Lens of that assistant determines its users permissions & Terminal Emulator tools available.
7: The assistant may now interact with the world by using the terminal.
8: An assistant may if given permission to message other agents of alter their permissions or create new asisstants.
9: Underlying all of this is a Kagi system-wids package.
10: For example, when creating a video game, a primary assistant could coordinate with specialized designer, coder, and playtester assistants. The workflow proceeds like this: designer creates concepts, passes them to main asistsant whhok passes them to coder, the coder for implementation, and the main assistant then forwards the result to the playtester for evaluation.
(speculative concept)
